the journal of philosophy - david chalmersconsc.net/papers/revisability.pdf · the extraordinary...

THE JOURNAL OF PHILOSOPHY V O L U M E C V I I I , N O . 8, A U G U S T 2 0 1 1

R E V I S A B I L I T Y A N D C O N C E P T U A L C H A N G E I N " T W O D O G M A S O F E M P I R I C I S M "

W . V . Quine's article "Two Dogmas o f Empi r i c i sm" is one o f the most i n f l uen t i a l works i n twentieth-century philosophy.^ The article is cast most expl ici t ly as an a rgument against logi

cal empiricists such as Carnap, a rgu ing against the analytic/synthetic d i s t i nc t i on tha t they appeal to a l ong w i t h the i r ve r i f i c a t i on i sm. B u t the article has been read m u c h more broadly as an attack o n the n o t i o n o f the a p r i o r i and o n the p rog ram o f conceptual analysis.

I w i l l address Quine's article construed as a cr i t ique o f the not ions o f analyticity and aprior i ty . I w i l l focus especially o n the most i n f l u e n tial part o f Quine's article: the arguments i n the final section concerni n g revisability and conceptual change. I n addressing these arguments, I w i l l adopt a l ine o f response g rounded i n Carnap's underappreciated article "Mean ing and Synonymy i n Natura l Languages."^ I wi l l argue that an analysis inspired by this article, when con jo ined wi th tools drawn f r o m contemporary two-dimensional semantics and f r o m Bayesian confirmation theory, provides what is needed to reject Quine's argument.

I w i l l n o t give a posit ive account o f the analytic o r the a p r i o r i b e y o n d the s tandard d e f i n i t i o n o f analyt ic i ty as t r u t h i n v i r tue o f mean ing and o f apr ior i ty as knowabi l i ty w i t h jus t i f i ca t ion independent o f experience. I am more inc l i ned to de fend the n o t i o n o f apr ior i ty than the n o t i o n o f analyticity, so I w i l l focus more o n the fo rmer , bu t the response that I w i l l develop can be used to d e f e n d ei ther n o t i o n f r o m Quine's cr i t ique.

' W. V. Quine, "Two Dogmas of Empiricism," Phibsophical Review, L X , 1 (January 1951): 20-43.

^Rudolf Carnap, "Meaning and Synonymy in Natural L3.ngu3.ges," Philosophical Studies, VI, 3 (April 1955): 33-47. Reprinted as Appendix D of Carnap's Meaning and Necessity: A Study in Semantics and Modal Logic (Chicago: University Press, 1956).

0022-362X/11/0808/387-415 © 2011 The Journal of Philosophy, Inc.

387

388 T H E J O U R N A L O F P H I L O S O P H Y

I . T H E A R G U M E N T S O F " T W O D O G M A S "

I n sections i t h rough i v o f "Two Dogmas," Quine argues that i f one tries to make sense o f the n o t i o n o f analyticity, one ends up moving i n a circle t h r o u g h cognate notions (synonymy, de f in i t i on , semantic rules, meaning) , and one cannot break ou t o f the circle. Many philosophers have been unmoved by this worry, as i t seems that one finds a similar circle f o r all sorts o f philosophically impor tan t notions: consciousness, causat ion , f r eedom, value, existence. So I w i l l set these criticisms aside here.

I n section v o f the art icle, Q u i n e makes points tha t specif ical ly address Carnap's logical empi r i c i sm, c r i t i c i z ing his cons t ruc t ion o f physical concepts f r o m p h e n o m e n a l concepts i n the Aupau, a n d his ver i f i ca t ion theory o f meaning . I w i l l set these points aside here, as I am n o t concerned to d e f e n d Carnap's cons t ruct ion or the verification theory o f meaning .

T h e ex t r ao rd ina ry i n f l u e n c e o f Quine ' s ar t ic le can be t raced i n large par t to its short final section. Part o f this in f luence stems f r o m the positive p ic ture that Q u i n e offers i n the first paragraph o f the section, character izing the totali ty o f ou r knowledge as a "man-made fabr ic w h i c h impinges o n experience only a long the edges," i n w h i c h "no par t icular experiences are l i n k e d w i t h any par t icular statements i n the in t e r io r o f the field, except ind i rec t ly t h r o u g h considerations o f e q u i l i b r i u m , a f fec t ing the field as a whole ." This p ic ture serves as a p o w e r f u l alternative to the ver i f ica t ionis t p ic ture p rov ided by some logical empiricists, bu t i t does n o t con ta in any direct a rgument against the analyt ic/synthet ic d is t inc t ion or the related n o t i o n o f apr ior i ty .

T h e most i n f l u e n t i a l arguments against an analyt ic /synthet ic dist i nc t i on are f o u n d i n the second paragraph, w h i c h I quote i n f u l l :

I f this view is right, i t is misleading to speak of the empirical content of an individual statement—especially i f i t be a statement at all remote f r o m the experiential periphery of the field. Furthermore it becomes fol ly to seek a boundary between synthetic statements, which ho ld contingently on experience, and analytic statements which hold come what may. Any statement can be held true come what may, i f 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 to 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 in principle between such a shift and the shift whereby Kepler superseded Ptolemy, or Einstein Newton, or Darwin Aristotle?"^

^ Quine, op. ciL, p. 40.

R E V I S A B I L I T Y 389

I w i l l focus o n these critical arguments. There are two crucial points.

( Q l ) "Any statement can be held true come what may, i f we make drastic enough adjustments elsewhere in the system."

(Q2) "No statement is immune to revision."

I f ( Q l ) and (Q2) are read as mere psychological claims, saying that as a matter o f fact someone migh t h o l d onto or revise any statement, then they are h igh ly plausible, bu t no t m u c h o f interest w i l l f o l l o w f r o m them. Quine is saying something more than this. We can understand ( Q l ) as saying that any statement can be rationally he ld true come what may, and (Q2) as saying that that no statement is i m m u n e to rational revision. These points have interesting consequences.

Many have taken these points to suggest e i ther that no sentences are analytic, o r tha t no d i s t i nc t ion can be d r awn between analytic and synthetic sentences. One possible connec t ion goes via the theses that analytic sentences are those that can be rat ionally h e l d true come what may and that a l l analytic sentences are i m m u n e to ra t ional revision. I f so, ( Q l ) suggests that by the f i rs t c r i t e r ion , al l sentences w i l l coun t as analytic. A n d (Q2) suggests that by the second c r i t e r ion , no sentence w i l l coun t as analytic. E i ther way, there is no useful distinct i o n between analytic a n d synthetic sentences to be had. Similar ly, i f we assume that a p r i o r i sentences are those that can be rat ionally he ld true come what may and that a l l a p r i o r i sentences are i m m u n e to revision, ( Q l ) and (Q2) suggest that there is no useful d is t inct ion between a p r i o r i and a poster ior i sentences to be had.

One c o m m o n response to the a rgumen t f r o m (Q2) is to suggest that revisability is quite compatible w i t h apr ior i ty (or analyticity), o n the g rounds tha t a p r i o r i j u s t i f i c a t i o n (or the j u s t i f i c a t i o n we have f o r bel ieving analytic sentences) is defeasible."^ For example, I m i g h t k n o w a ma thema t i ca l c l a im a p r i o r i , b u t my j u s t i f i c a t i o n m i g h t be defeated i f I learn that a leading mathemat ic ian thinks that the c la im is false. I t h i n k that this response is correct as fa r as i t goes, b u t to rest ent irely o n i t w o u l d be to concede a great deal to Quine . O n a c o m m o n t r ad i t i ona l concep t ion , at least some a p r i o r i j u s t i f i c a t i o n (and some ju s t i f i c a t i on f o r beHeving analytic t ruths) is indefeasible.

^See, fo r example, Hartry Field, "The A Prioricity of Logic," Proceedings of the Aristotelian Society, n. s., x c v i (1996): 359-79. Philip Kitcher, "A Priori Knowledge Revisited," in Paul Boghossian and Christopher Peacocke, eds.. New Essays on the A Priori (New York: Oxford, 2000), pp. 65-91, defends a conception of the a priori that requires indefeasi-bility, while Peacocke, The Realm of Reason (New York: Oxford, 2004) defends a conception that does not. For present purposes I will remain neutral on whether apriority entails some sort of ideal indefeasibility; the observation about testimony in section v i contains some relevant discussion.


O n e reasonably m i g h t h o l d that some a p r i o r i j u s t i f i ca t ion ( i n logic or mathematics, say) yields n o t j u s t knowledge bu t certainty, at least o n ideal ref lect ion. These claims are no t obviously correct, bu t they are also no t obviously incorrect, and I do no t t h i n k that Quine's argument establishes that they are false. So I wi l l take another l ine o f response.

The response I w i l l develop takes o f f f r o m the response given by Grice and Strawson at the end o f their article " I n Defense o f a Dogma." This response holds that ( Q l ) and (Q2) are compatible w i th an analyt ic / synthetic dis t inct ion, f o r a reason quite d i f f e ren t f r o m the one given above. Here is a passage addressing the argument f r o m (Q2):

Now for the doctrine that there is no statement which is in principle immune f r o m revision, no statement which might not be given up i n the face of experience. Acceptance of this doctrine is quite consistent with adherence to the distinction between analytic and synthetic statements. Only, the adherent of this distinction must also insist on another; on the distinction between that k ind of giving up which consists i n merely admitt ing falsity, and that k ind of giving up which involves changing or dropping a concept or set of concepts. Any form of words at one time held to express something true may, no doubt, at another time, come to be held to express something false. But i t is not only philosophers who would distinguish between the case where this happens as the result of a change of opinion solely as to matters of fact, and the case where this happens at least partly as a result of a shift in the sense of the words. Where such a shift in the sense of the words is a necessary condition of the change in truth-value, then the adherent of the distinction will say that the fo rm of words in question changes f rom expressing an analytic statement to expressing a synthetic statement.... And i f we can make sense of this idea, then we can perfectly well preserve the distinction between the analytic and the synthetic, while conceding to Quine the revisability-in-principle of everything we say.̂

Here the central p o i n t is that our judgments about any sentence, even an analytic sentence, w i l l be revisable i f the meaning o f the words change. For example, i f ' bache lo r ' changes f r o m a t e rm f o r u n m a r r i e d m e n to a t e r m f o r sociable men , then we w i l l no longer j udge that ' A l l bachelors are u n m a r r i e d ' is true. Bu t this observation is j u s t what an adherent o f the analytic/synthetic dis t inct ion should expect. Analytic sentences instead should be unders tood as those sentences that are i m m u n e to revision whi le the i r mean ing stays constant. M o r e precisely, they are those that are i m m u n e to rat ional re ject ion whi le their mean ing stays constant. (There is a sense i n wh ich an analytically false sentence m i g h t

^ H . P. Grice and P. F. Strawson, " In Defense o f a Dogma," Philosophical Review, Lxv, 2 (Apri l 1956): 141-58, at pp. 156-57.


be i m m u n e to ra t ional revision, bu t f r o m here onward I w i l l understand 'revision' as r equ i r ing rejection.)

Fo l lowing standard practice, we can say that when the mean ing o f a sentence changes, there is conceptual change, some expression i n the sentence at f irst expresses one concept and later expresses another. W h e n the meaning o f a sentence stays the same, there is conceptual constancy: the expressions i n the sentence wi l l express the same concepts throughout. T h e n Grice and Strawson's po in t could be pu t by saying that an analytic sentence is one that is i m m u n e to revision wi thou t conceptual change. M o r e cautiously, the p o i n t could be pu t by saying that the fact that a sentence is revisable under condit ions o f conceptual change does no t entail that i t is no t analytic. Something similar applies to apriority.^

A t this po in t , Qu ine has two obvious replies. The f i rs t is to say that the appeal to mean ing i n characterizing the class o f analytic sentences is c i rcular , as the n o t i o n o f m e a n i n g is as poo r ly under s tood as the no t ion o f analyticity. The same could be said fo r the appeal to concepts and to propositions. This reply w o u l d be i n the spirit o f the f irs t f o u r sections o f "Two Dogmas." Bu t then this a rgument w i l l n o t be m u c h o f an advance o n the arguments i n the f irs t f o u r sections, and anyone who is no t moved by those arguments w i l l no t be moved by this one.

T h e second, m o r e i n t e r e s t i ng r ep ly is to cha l lenge Gr ice a n d Strawson to provide a p r i n c i p l e d d is t inc t ion between cases o f revision that involve conceptua l change and those that involve conceptual constancy. Q u i n e m i g h t argue that cases that are p u r p o r t e d to be o n e i ther side o f this division are i n fact cont inuous w i t h each other, and that there is no p r i n c i p l e d d is t inc t ion to be had. Something hke this t hough t m i g h t even be read in to the last sentences o f the paragraph f r o m Qu ine quoted above.

N o w one m i g h t suggest that Grice and Strawson are no t obl iged to provide a reductive characterization o f the dist inct ion—that is, one that does no t use 'meaning' and cognate notions—any more than they are requi red to provide a reductive de f in i t i on o f mean ing or analyticity

'̂We also might allow that there is conceptual change in this sense when the proposition expressed by an utterance of a sentence changes because of a shift in context. For example, 'Someone is bald i f f they have no hairs' might be accepted in one context and rejected in another. It is not clear that a mere contextual shift could change the status of a sentence as analytic, as arguably the meaning of such a sentence stays constant throughout. But i f we say that a sentence is a priori i f it expresses a proposition that is knowable a pr ior i , then i t is natural to hold that a sentence might be a pr ior i in one context but not in another. It is for reasons like this that I speak of 'conceptual change' rather than 'meaning change' or 'semantic change'; the latter phrases tend to suggest changes in standing linguistic meaning (thereby excluding mere contextual shifts), but it is changes in the propositions and concepts expressed that matter most for our purposes.


to answer the challenge i n the first f o u r sections. Again , this suggestion seems correct as far as i t goes. Nevertheless, i f Quine's opponen t cann o t say m u c h to characterize the p r i n c i p l e d d i s t inc t ion here, he o r she is at least i n the awkward dialectical posi t ion o f leaving a challenge unanswered, and o f leaving doubts about the dist inction unassuaged.

M y view is that m u c h can be said to f lesh ou t a p r i n c i p l e d distinct i on . I t h i n k that the basic tools f o r d o i n g so can be f o u n d i n Carnap's "Mean in g and Synonymy i n Na tu ra l Languages."

I I . C A R N A P O N I N T E N S I O N S

Carnap is Quine's major target i n "Two Dogmas o f Empiricism." I t is no t always appreciated that "Meaning and Synonymy i n Natura l Languages" can be read as a sustained response to Quine , perhaps because Carnap spends l i t t ie t ime discussing h i m explicitiy. Nevertheless, Carnap says enough to make clear that a response to "Two Dogmas" is in tended.

Carnap 's a r t ic le sets o u t to analyze the n o t i o n o f m e a n i n g a n d rela ted not ions such as synonymy. His a im is to provide a "scientific p r o c e d u r e " by w h i c h m e a n i n g a n d synonymy can be analyzed i n broadly naturalistic terms. Impor tan t ly , he aims to explicate n o t only the n o t i o n o f extension, b u t the n o t i o n o f in tens ion (the "cognitive or designative c o m p o n e n t o f meaning") , w h i c h he notes has been c r i t i cized by Q u i n e as "foggy, mysterious, and n o t really understandable."^

Carnap's key idea is that we can investigate the in tens ion that a subject associates w i t h an expression by invest igat ing the subject's j u d g m e n t s about possible cases. T o de te rmine the in tens ion o f an expression such as 'Pferd ' f o r a subject, we present the subject w i t h descriptions o f various logically possible cases, and we ask the subj e c t whe ther he or she is w i l l i n g to apply the t e r m 'Pferd ' to objects specif ied i n these cases. I f we do this f o r enough cases, t hen we can test a l l sorts o f hypotheses about the in tens ion o f the expression.

I n this article, Carnap takes the t e r m ' intension' as pr imit ive and does no t b u i l d possible cases in to the very nature o f intensions. B u t f o r our purposes i t is useful to adopt a suggestion that Carnap makes elsewhere, and simply define an intension as a f u n c t i o n f r o m possible cases to extensions. For a t e rm like 'Pferd', the intension wi l l be a f u n c t i o n f r o m possible cases to objects characterized i n those cases. For a sentence such as 'Grass is green', the intension wi l l be a f u n c t i o n f r o m possible cases to truth-values. T h e n Carnap's procedure above can be regarded as a way o f direct ly ascertaining the values o f the in tens ion that a subject associates wi th an expression, by presenting the subject w i t h a possible case and no t i ng the extension that the subject associates w i t h the case.

' Carnap, op. cit., p. 36.


O f course one cannot actually present a subject w i t h a l l possible cases to de termine every aspect o f an in tension. Bu t Carnap suggests tha t the i n t e n s i o n tha t a speaker associates w i t h an express ion is de te rmined by the speaker's l inguistic dispositions. For a given expression E used by a given speaker, the speaker w i l l have the disposit ion to associate a given extension w i t h E, w h e n presented w i t h a possible case. For example, i f 5 is a sentence, the speaker w i l l have the disposit i o n to j u d g e the sentence as t rue or false o f a possible case, w h e n presented w i t h that case. The intension o f an expression can then be seen as a f u n c t i o n that maps possible cases to the extension that the speaker is disposed to ident i fy , when presented w i t h that case.

I n this way, Carnap defines an expression's in tension i n naturalistic a n d even opera t iona l terms. We can go o n to de f ine synonymy: two expressions are synonymous ( fo r a speaker at a t ime) when they have the same intension ( for that speaker at that t ime). A n d we can define analyticity: a sentence is analytic ( for a speaker at a t ime) when its intension has the value ' true' at all possible cases ( for that speaker at that t ime).

W i t h this d e f i n i t i o n i n hand, we can go o n to provide a p r i nc ip l ed cr i ter ion f o r conceptual change over t ime. A n expression undergoes change i n mean ing between and f o r a speaker i f f the speaker's intension f o r E at differs f r o m the speaker's intension f o r E at fe. I f we accept Carnap's dispositional account o f intensions, i t follows that E undergoes change i n meaning between and ^ i f f there is a possible case such that the speaker is disposed to associate d i f fe ren t extensions w i t h E when presented w i t h the case at and ^ .

O f course there are many immedia te questions about Carnap's account. W h a t is a possible case? I n what vocabulary are these cases specified? H o w can we d e t e r m i n e w h e t h e r the m e a n i n g o f this vocabulary has changed? Can speakers make mistakes about in tensions? Can they change thei r m i n d about a case w i t h o u t a change i n meaning? Can mean ing really be operat ional ized this easily? A n d so on . Carnap's account may need to be m o d i f i e d or at least r e f i n e d to answer these questions.

Before addressing these matters, I wi l l illustrate how Carnap's account m i g h t be used to address the challenge i n section v i o f "Two Dogmas" directiy. I n my view, the essential aspects, i f no t the specifics, o f the resul t ing response are sound. These essential aspects carry over to more ref ined analyses couched i n terms o f two-dimensional semantics (section i v ) and Bayesian conf i rma t ion theory (section v and v i ) .

I I I . A C A R N A P I A N R E S P O N S E

I n "Mean ing and Synonymy i n Natura l Languages," Carnap does no t m e n t i o n the arguments i n section v i o f "Two Dogmas"; n o r does he


address revisabil i ty o r concep tua l change. Nevertheless, his f r a m e w o r k can be used to give a response to these arguments that is broadly i n the spir i t o f Grice and Strawson's response, fleshed ou t w i t h a p r i n c ip led c r i t e r ion f o r conceptual change.

W e can start w i t h Quine 's observat ion that any statement can be h e l d t rue come what may. This seems correct. Even a paradigmatical ly synthetic sentence such as ' A l l bachelors are unt idy ' can be h e l d t rue i n the face o f apparently counterva i l ing evidence, i f we al low suf f ic ien t ad jus tment o f ancil lary claims. T h e quest ion is whether such adjustments w i l l involve conceptual change, and whether we have a p r i n c ip led c r i t e r ion f o r d e t e r m i n i n g this.

W e m i g h t as wel l start w i t h a case. A t ^ i , Fred asserts " A l l bachelors are unt idy." A t ^ , Fred is presented w i t h evidence o f a t idy u n m a r r i e d m a n . Fred responds, "He's no bachelor! Bachelors must be over 30, and he's only 25." A t Fred is presented w i t h evidence o f a 35-year-old m a n w i t h a spotless apartment. Fred responds, "He's no t tidy! L o o k at the mess i n his sock drawer." h i this way, Fred holds the sentence true throughout , and t h rough similar maneuvers he can h o l d i t t rue come what may.

Does this case involve conceptual change? We can apply Carnap's analysis to see whether Fred's in tens ion f o r ' A l l bachelors are un t idy ' (call this sentence B) changes over the relevant time-span. Suppose that

is a detailed possible case i n wh ich there is an unmar r i ed 25-year-old m a n w i t h a tidy apartment. A t ^ , when Fred is presented wi th the in fo r ma t ion that c obtains, he responds that ' A l l bachelors are unt idy ' is true w i t h respect to c. By Carnap's c r i te r ion , Fred's intension f o r B is t rue w i t h respect to c at ^ .

W h a t about Fred's i n t ens ion f o r 5 at ^i? T h e key ques t ion is: i f F r ed h a d been presented w i t h a desc r ip t ion o f c at ^ i , b e f o r e he had evidence that the case was actual, w o u l d he have j u d g e d tha t ' A l l bachelors are unt idy ' was true w i t h respect to c}

I f the answer is yes, t hen Carnap's c r i t e r ion suggests that there is no relevant conceptual change between and ^ . I n this case, Fred w i l l s imply have had an unusual in tens ion f o r 'bachelor ' al l a long.

I f the answer is no , t hen Carnap's c r i t e r i o n suggests that there is relevant conceptual change between and T h e in tens ion o f ' A l l bachelors are un t idy ' w i l l have changed d u r i n g this t ime , p robab ly because the in tens ion o f 'bachelor ' has changed d u r i n g this t ime.

T h e same applies m o r e general ly . I f a speaker's j u d g m e n t conce rn ing a case at t2 is ref lec ted i n the speaker's dispositions to respond to such a case at ^ i , we can say that the speaker's j u d g m e n t conce rn ing that case is prefigured. I f a speaker's j u d g m e n t concern ing a case at 2̂ is n o t r e f l ec ted i n the speaker's dispositions at ^ i , we can say that the


speaker's j u d g m e n t conce rn ing the case is postfigured. O n Carnap's account, pos t f igured j udgmen t s involve conceptual change, bu t pref i g u r e d j udgmen t s do n o t

I n any case, we have what we need. Carnap's f r a m e w o r k allows us to see how any sentence can be he ld true come what may, whi le at the same t ime a l l o w i n g a p r i n c i p l e d way to d is t inguish between those cases o f holding- t rue that involve conceptual change and those that do not . Someth ing similar applies to cases o f revisability, t hough I w i l l no t go in to the details here.

I V . R E F I N I N G C A R N A P ' S A C C O U N T

Carnap's account o f m e a n i n g is remarkably simple, and one m i g h t reasonably wonder whether such a simple account can be correct. I t h i n k that wh i l e there are p rob lems w i t h the account, they can be addressed i n a way tha t preserves s o m e t h i n g o f the sp i r i t o f the account, i f no t the letter.

Perhaps the most obvious p r o b l e m is that subjects can make mistakes. A sub jec t m i g h t mi sca l cu la t e a n d j u d g e t h a t 36 + 27 = 73, and she m i g h t even be disposed to j u d g e this to be true w i t h respect to a l l possible scenarios. O n Carnap's account , i t w i l l f o l l o w tha t '36+27=^73' is analytic f o r the subject. Bu t this seems the w r o n g result: o n the face o f i t , the sentence is no t even true. Similar mistakes seem possible f o r non idea l subjects i n al l sorts o f domains.

T o handle cases o f this sort, we can m o d i f y the account to appeal n o t to what the subject w o u l d say i n response to the case, bu t to what the subject should say, or what she w o u l d say given ideal reasoning. We m i g h t say that the in tens ion o f E maps a possible case C to the extension that the subject w o u l d iden t i fy f o r E, i f she were to be presented w i t h C and were to reason ideally.

Construed this way, the account w i l l no longer y ie ld an operat ional d e f i n i t i o n o f meaning , at least unless we can f i n d an operat ional criter i o n f o r ideal reasoning. Bu t this is no t a bad t h i n g f o r those inc l ined to reject behaviorism i n any case. I t is also fa r f r o m clear that this account provides a naturaHstic r educ t ion o f meaning . I t w i l l do so only i f we already have a naturalistic r educ t ion o f ideal reasoning. Bu t the account need n o t be a naturalistic r educ t ion to be useful .

Someone m i g h t suggest that i n these cases, facts about mean ing de t e rmine facts about ideal reasoning ra ther than vice versa: i t is precisely because we mean such-and-such by 'Pferd ' that we should say such-and-such. We need n o t take a stand o n these questions about metaphysical p r io r i ty here. A l l we need is that i n these cases, there are facts about what subjects should say or what ideal reasoni n g dictates, and tha t we have some p re theore t i ca l g r i p o n these


facts. T h e n we can use these facts to he lp explicate a cor responding n o t i o n o f meaning , regardless o f w h i c h o f these not ions is metaphysically p r io r . I n effect , we are us ing an antecedent g r ip o n normat ive no t ions to h e l p expl icate semantic no t ions . O f course i t remains o p e n to a Q u i n e a n o p p o n e n t to reject normat ive not ions ent irely. I discuss oppos i t ion o f that sort later i n this article.

A n o t h e r p r o b l e m is t ha t o n the c o n t e m p o r a r y u n d e r s t a n d i n g , intensions o f t e n are inaccessible to a subject, even given ideal reasoning . For example, i f Kr ipke is r igh t , the in tens ion o f 'water' picks ou t H 2 O i n al l possible worlds, even f o r subjects who do n o t k n o w that water is H20.^ Such subjects w i l l no t be disposed to ident i fy H2O as the extension o f 'water' w h e n presented w i t h a possible case, so Carnap's d e f i n i t i o n w i l l get the intension wrong .

T o handle this p r o b l e m , we can take a leaf f r o m two-dimensional semantics, w h i c h recognizes two sorts o f in tens ion . Even i n l i g h t o f Kr ipke ' s ins igh t , Carnap's accoun t s t i l l m i g h t apply to one sort o f in tens ion , t h o u g h n o t the other . Kripke's p o i n t applies to secondary intensions, w h i c h govern possible cases considered as counterfactual : i f there had been XYZ i n the oceans and lakes, water w o u l d sti l l have been H 2 O . For the purposes o f Carnap's account , t h o u g h , we can focus o n p r imary intensions and stipulate that subjects consider the possible cases as actual. For example, we can ask t h e m to suppose that X Y Z actually is i n the oceans and lakes i n the actual w o r l d and ask f o r the i r ve rd ic t about the extension o f 'water' u n d e r that supposi t ion. Subjects plausibly w i l l h o l d that 'water' picks ou t XYZ i f that hypothesis is correct . This mi r ro r s the f ami l i a r suggestion that the p r imary in tension o f 'water' picks ou t XYZ i n a T w i n Ear th scenario, a l though the secondary in tens ion o f 'water' picks ou t H 2 O there. So i t is n o t ou t o f the quest ion that a Carnap-style account m i g h t w o r k f o r p r i m a r y intensions, w h i c h i n any case are o f t e n h e l d to be the sort o f in tensions that are t ied most closely to apr ior i ty and analyticity.

A t h i r d issue is the nature o f possible cases. For ou r purposes they s h o u l d be a k i n to possible wor lds . T h e y m i g h t be cen te red metaphysically possible worlds ( that is, worlds m a r k e d w i t h an i n d i v i d u a l a n d a t i m e ) , w i t h the c e n t e r i n g r e q u i r e d to hand le in tens ions f o r expressions such as T and 'now'. They also m i g h t be regarded as epi-stemically possible worlds, o r epistemically possible scenarios, w h i c h m i g h t be m o d e l e d by m a x i m a l consistent sets o f sentences that cannot be r u l e d ou t a p r i o r i . I w i l l n o t try to settle this issue here, bu t I w i l l use the w o r d 'scenario' as a generic t e r m f o r the entities involved. I n

^Saul Kripke, Naming and Necessity (Cambridge: Harvard, 1980).


order to suppose that subjects can reason about entire scenarios, we can once again appeal to the idealizat ion o f what the subject w o u l d say given ideal reasoning.

A f o u r t h issue is the vocabulary i n w h i c h a scenario is specif ied. Such a vocabulary w i l l need to be r i c h enough that a f u l l enough specif icat ion using this vocabulary plus ideal reasoning determines judgmen t s about o ther expressions' extensions, w i t h o u t be ing so r i ch that a specif icat ion builds i n al l those expressions directly. I t is a substantive c la im that some such vocabulary can be f o u n d , bu t p roponents o f t w o - d i m e n s i o n a l semantics have o f f e r e d a rguments f o r this c la im a long w i t h suggested vocabularies.^ The details o f such a vocabulary w i l l no t matter f o r ou r purposes.

A f i f t h issue is the wor ry tha t subjects m i g h t change the i r m i n d about a possible case w i t h o u t a change o f mean ing . Here , one can respond by r e q u i r i n g , as above, that the specif icat ion o f a scenario is r i c h e n o u g h that j u d g m e n t s about sentences are d e t e r m i n e d by the spec i f ica t ion and by ideal reasoning. I f so, then i f subjects are given such a specif icat ion and reason ideally th roughout , there w i l l be n o r o o m f o r t h e m to change t h e i r m i n d i n this way. Changes o f m i n d about a f u l l y specif ied scenario w i l l always involve e i ther a f a i l u r e o f idea l reasoning o r a change i n mean ing . O f course this c la im requires a version o f the substantive c la im i n the previous paragraph. I r e t u r n to a version o f this issue below.

The m o d e l we then reach is someth ing l ike the f o l l o w i n g . The (pr i mary) in tens ion o f an expression f o r a subject is a f u n c t i o n that maps scenarios to extensions, m a p p i n g a scenario w to what the subject w o u l d j u d g e to be the ex tens ion o f E u n d e r the suppos i t ion tha t w is actual, were she ideally ra t ional . This is no t a perfect d e f i n i t i o n , bu t i t is good enough f o r our purposes. This remains very m u c h i n the sp i r i t o f Carnap's d e f i n i t i o n , a l t h o u g h the invoca t ion o f ra t ional i ty makes i t a normat ive var ia t ion o n Carnap's account.

Impor tan t ly , we can use this account to provide a version o f the Carnapian response to Quine's arguments given i n the previous sect i o n . Conceptua l change ( o f the relevant sort) w i l l occur precisely w h e n an expression's p r i m a r y in tens ion changes across t ime . This w i l l h appen precisely w h e n the subject's dispositions to j u d g e the expression's ex tens ion i n a possible case (given idea l reasoning) changes. As i n the last section, we can f i n d cases o f ho ld ing- t rue

^See, for example, David J. Chalmers and Frank Jackson, "Conceptual Analysis and Reductive Explanation," Philosophical Review, cx, 3 (July 2001): 315-60. See also my Constructing the World (New York: Oxford, forthcoming).


where the dispositions change i n this way, and cases where they do not . W h a t matters is that we have a p r i n c i p l e d dis t inct ion.

A residual issue concerns the m e a n i n g o f the basic vocabulary. I f cases are speci f ied i n this vocabulary, t h e n we need to ensure tha t the basic vocabulary does n o t change i n m e a n i n g t h r o u g h o u t the process. I f we do no t requi re this, the resul t ing c o n d i t i o n f o r meani n g change w i l l be inadequate; a subject's dispositions to j u d g e that S obtains w i t h respect to a case specified by D m i g h t change over t ime , n o t because the mean ing o f 5 changes bu t because the m e a n i n g o f terms i n D change. I f we do require this, however, i t appears that we need some f u r t h e r c r i t e r ion f o r m e a n i n g change i n the basic vocabulary items used i n D, as the disposit ional m e t h o d w o u l d y ie ld t r iv ia l results. So i t appears that the disposit ional m e t h o d f o r d e t e r m i n i n g m e a n i n g change, even w h e n idealized, is incomple te .

A second residual issue concerns the role o f the a p r i o r i i n character iz ing this account. I t is na tura l to suggest that the ideal reasoning i n ques t ion mus t be res t r ic ted to idea l a p r i o r i reasoning. I n fact , some two-dimensional accounts use the n o t i o n o f apr ior i ty i n d e f i n i n g p r i m a r y intensions: the p r imary in tens ion o f a sentence S evaluated at a w o r l d w is t rue precisely i f a mater ia l cond i t i ona l ' I f D, t h en S' is a p r i o r i , where Z) is a canonical specif icat ion o f S}^ I f we take this route , t hen we have arr ived at a p r i n c i p l e d d is t inc t ion only by h e l p i n g ourselves to the contested n o t i o n o f apr ior i ty a long the way.

As b e f o r e , i t is n o t clear h o w b a d these res idua l p r o b l e m s are. O n e s t i l l m i g h t see the i n t ens iona l analysis as d e m o n s t r a t i n g tha t the Q u i n e a n p h e n o m e n a o f ho ld ing - t rue and revisability are qu i te compa t ib l e w i t h the in t ens iona l f r a m e w o r k a n d have n o power to r e fu t e i t . Even i f one has to assume some independen t g r ip o n the n o t i o n o f ap r io r i ty and o n the m e a n i n g o f expressions i n the basic vocabu la ry , one can use the f r a m e w o r k to p r o v i d e a reasonably en l igh ten ing analysis o f relevant cases. Still, we have no t b roken ou t o f

^*^This objection is related to Quine's argument f r o m the indeterminacy of translation in Word and Object (Cambridge: M I T , 1960). Quine took Carnap's account to be a serious challenge to his arguments in "Two Dogmas," and the indeterminacy argument can be seen in part as a response to it. There, Quine argues that no dispositional analysis can settle facts about meaning, because mult iple assignments of reference will always be compatible with a subject's behavioral dispositions. This applies even to Carnap's account, i f we allow multiple potential assignments of reference to the basic vocabulary. In effect, Carnap's account assumes that the meaning of the basic vocabulary is fixed, but i t is not clear why such an assumption is legitimate, and i t is not clear how this meaning itself might be grounded in dispositional facts. Thanks to Gillian Russell for discussion here.

See, for example, Chalmers, "Epistemic Two-Dimensional Semantics," Philosophical Studies, c x v i i i , 1/2 (March 2004): 153-226.


the Quinean circle. I t w o u l d be nice to be able to characterize the relevant distinctions wi thout such a direct appeal to the contested notions.

I t h i n k that such a characterization can be f o u n d . T h e key idea is to cast things i n terms o f cond i t iona l probabihty, rather than i n terms o f apr ior i ty .

A n i n i t i a l observation is that someth ing very m u c h l ike a p r imary i n t e n s i o n can be charac ter ized w i t h o u t appea l ing to ap r io r i t y , by appeahng to c o n d i t i o n a l p robabih ty instead. I n par t icular , one can de f ine the in tens ion o f a sentence S at a scenario w, f o r a subject, i n terms o f the subject's ra t iona l cond i t i ona l credence i n 5 given D cr\S\D), where / ) is a canonical specif icat ion o f w. We can say that the in tens ion o f 5 is t rue at w i f f cr'{S\D) is h i g h , and false at w i f f cr'{S\D) is low. H e r e we r equ i re an idea l iza t ion , so tha t cr'{S\D) is the c o n d i t i o n a l credence tha t the subject w o u l d have given ideal reasoning, o r someth ing a long those lines.

I f we do this, t hen we w i l l have a p r i n c i p l e d c r i t e r ion f o r conceptual change that does no t appeal to aprior i ty . O n this c r i t e r ion , a subject's in tens ion f o r S w i l l change between and t2 i f f there is a scenario w w i t h canonica l speci f ica t ion D such that cr'{S\D) changes f r o m h i g h to low or vice versa. One then c o u l d r u n the arguments o f the previous section again using this n o t i o n . This w i l l provide a reply to Quine 's chal lenge that gets a r o u n d the second residual issue above ( r ega rd ing a p r i o r i t y ) , t h o u g h i t s t i l l may be subject to a vers ion o f the first issue ( regarding the basic vocabulary).

A t this po in t , however, I t h i n k an alternative analysis involv ing cond i t iona l probabi l i ty is available. This analysis is closely related to the one j u s t m e n t i o n e d , and is a descendant o f the Carnapian analysis i n the previous section, bu t i t does n o t requi re any o f the semantic apparatus used i n that section and this one. This analysis proceeds us ing o n l y s t andard Bayesian cons idera t ions a b o u t evidence a n d upda t ing . I n add i t i on to the advantage o f fami l ia r i ty , this approach has o ther s ignif icant advantages i n responding to Quine's challenge. By avoiding the need f o r canonical specifications o f complete possible scenarios, i t avoids the large idealizat ion needed to handle enormous specif icat ions. I t also has the p o t e n t i a l to avoid o r m i n i m i z e b o t h residual issues above.

V . A B A Y E S I A N A N A L Y S I S O F H O L D I N G - T R U E

I n what fo l lows I develop a Bayesian analysis o f Quine 's arguments f r o m h o l d i n g - t r u e a n d f r o m revisabi l i ty . For the purposes o f this analysis, we can set the f r a m e w o r k o f scenarios a n d in tens ions to one side. A l l we need are o r t h o d o x Bayesian claims about credence and its revision i n l i gh t o f new evidence.


Le t us assume a standard Bayesian m o d e l o n w h i c h sentences are associated w i t h uncond i t i ona l and cond i t iona l credences f o r subjects at times. T h a t is, f o r a given subject and a given t ime, a sentence 5 w i l l be associated w i t h an uncond i t i ona l credence cr{S), and a pair o f sentences 5 and T w i l l be associated w i t h cond i t iona l credence cr{S\T). (These o r d i n a r y credences cr{S\T) s h o u l d be d i s t i ngu i shed f r o m the idealized ra t ional credences cr\S\T) f r o m the previous section.) Credences are standardly taken to be real numbers between 0 and 1, b u t f o r o u r purposes exact i tude is n o t r equ i r ed . I t is e n o u g h tha t some credences be h i g h and others low.

I w i l l also assume a version o f the s tandard Bayesian p r i n c i p l e o f condi t ional iza t ion: i f a subject has credence cri{S\E) at and acquires total evidence specified by the evidence sentence E between ti and ^ , t hen the subjecfs credence cr2{S) at ^ shou ld be equal to cri{S\E). I w i l l give a m o r e precise version o f this p r inc ip le below. T h e nature o f evidence sentences w i l l be discussed later i n this ar t ic le , b u t f o r now we can t h i n k o f t h e m as specifying ei ther that cer ta in experiences ob ta in o r that cer tain observable states o f affairs obta in .

We can start w i t h a typical case whereby an apparent ly synthetic sentence is h e l d t rue i n the face o f apparen t ly c o u n t e r v a i l i n g evidence, by appeal to appropr ia te ancil lary theses. As before , suppose that at ^1 Fred asserts, " A l l bachelors are unt idy." A t t^, Fred acquires evidence ind ica t ing that there is a tidy, u n m a r r i e d 25-year-old m a n , a n d responds by d e n y i n g tha t the m a n is a bachelor , as bachelors must be over 30.

L e t B be ' A l l bachelors are unt idy ' , and let Ehe Fred's total relevant evidence acquired between and ^ . Le t cri{B) and cr2{B) be Fred's credences i n B at and t2 respectively. T h e n cri{B) and cr2{B) are b o t h h i g h .

The crucial question is: W h a t is CTI ( ^ I^" ) , Fred's cond i t iona l credence i n B given £ a t ^ i , before Fred acquires the evidence i n question?

I f cri{B\E) is h i g h , t h e n Fred's j u d g m e n t at t2 reflects a cond i t iona l credence that he already had at ti. I n this case, the j u d g m e n t at t2 is prefigured, i n a sense analogous to the sense discussed earlier. H e r e , Fred's accept ing B i n l i g h t o f E accords w i t h the p r i n c i p l e o f condi t iona l iza t ion .

I f cri{B\E) is low, t h e n Fred's j u d g m e n t at t2 fai ls to r e f l e c t the c o n d i t i o n a l credence that he already had at ti. I n this sort o f case,

'^The arguments I present here also can be run using the principle of Jeffrey conditionalization, which allows conditionalization on evidence about which a subject is not certain. See Richard C. Jeffrey, The Logic of Decision, 2nd ed. (Chicago: University Press, 1983).


the j u d g m e n t at is postfigured, i n a sense analogous to the sense discussed earlier. Here , Fred's accepting B i n l igh t o f appears to violate the p r inc ip le o f condi t ional iza t ion.

O n standard Bayesian assumptions, there are two central ways one can ob ta in apparent violat ions o f cond i t i ona l i za t i on f o r sentences. First, this can happen w h e n the subject is no t f u l l y ra t ional th roughou t the process; perhaps at Fred has n o t t hough t things t h r o u g h properly, or at t^ he makes some sort o f reasoning error. Second, the content o f the key sentence B can change between ti and t^. This can happen i n cases involving indexicals, wh ich are no t relevant here, or i n cases o f conceptual change. I n these cases, i t remains possible that Fred's credences i n relevant propositions obey condit ionalizat ion, bu t that his credences i n associated sentences do not, because the associat ion between sentences and propositions changes over time.^"^

We m i g h t fo rmula t e this as a version o f the Bayesian p r inc ip le o f condi t ional iza t ion , f o r sentences:

(CS) I f a subject is ful ly rational, and i f the subject acquires total evidence specified by E between tx and t^, and i f the content of sentence Sdoes not change between t\ and t^, then cr^iS) = cri{S\E).

Perhaps the most f ami l i a r version o f the p r inc ip le o f conditionaUza-t i on is cast i n terms o f proposit ions: i f a f u l l y ra t ional subject acquires total evidence specified by p ropos i t ion e between ti and t^, t hen cr2{p) = cri{p\e). (CS) fo l lows f r o m this c l a i m i n c o n j u n c t i o n w i t h the plausible claims that w h e n sentence S expresses p ropos i t ion p f o r a subject at that t ime, cr{S) = cr{p) at that t ime , and that the content o f a sentence is the p ropos i t ion i t expresses.

I t fol lows that i f Fred i n the pos t f igured case above is f u l l y ra t ional , then this is a case o f conceptual change. O f course i t m i g h t be that Fred is no t f u l l y ra t ional , bu t this is o f no he lp f o r Quine . I t is unremarkable that i r r a t iona l subjects m i g h t h o l d o n to any sentence or reject any sentence, and this observation has no consequences regarding analytici ty o r apr ior i ty . For Quine 's observations about revisability and

A potential third way that conditionalization can be violated arises on views where sentences express certain relativistic contents. Consider a view on which utterances of the sentence ' I t is raining' always express the same temporal proposition It is raining, which can be true at some times and not at others. On Saturday, I might have a low conditional credence in It is raining given The weather forecast says rain on Sunday; then on Sunday I might acquire evidence that the weather forecast says rain on Sunday, resulting in high credence in It is raining, without irrationality. On a more standard view, on which the content of ' I t is raining' uttered at t is It is raining at t, this will be classified as a change in content, but on the temporal view the content stays the same. For present purposes, we can either count these as changes in content in an extended sense, or we can require in principle (CS) that the content in question is nonrelativistic content.


hold ing- t rue to have any bite , ra t iona l subjects are requ i red . So we may as wel l assume that Fred is f u l l y ra t ional .

I f we assume tha t the r e l evan t subjects are f u l l y r a t i o n a l , we n o w have a p r i n c i p l e d c r i t e r i o n f o r concep tua l change i n a case o f ho ld ing- t rue . Suppose that o u r subject accepts 5 at ^ i , acquires apparen t ly coun te rva i l i ng evidence E between a n d ^ , a n d continues to accept S at t^- T h e n we can say

(i) I f cri{S\E) is low, this is a case of conceptual change. (ii) I f cri{S\E) is high, this need not be a case of conceptual change.

O n e can now ask: is i t t rue that a subject can h o l d o n to any given sentence 5 come wha t may, i n l i g h t o f any evidence, w i t h o u t i r r a t iona l i ty or conceptua l change? By this analysis, this c l a im requires that f o r any given sentence 5 and any evidence £ , cr{S\E) is h i g h (or at least is no t low) . B u t this c la im is obviously false. For ra t ional subjects and most sentences ( i n c l u d i n g most paradigmatical ly empi r i ca l sentences), there w i l l be evidence sentences £"such that cr{S\E) is low.

T h e m o r a l here is that i n the general case, Qu inean hold ing- t rue come what may requires widespread v io la t ion o f condi t iona l iza t ion , w h i c h requires i r ra t iona l i ty or conceptual change. B u t the fac t tha t an i r r a t iona l subject m i g h t reject a sentence is no evidence that i t is n o t analytic or a p r i o r i , a n d the fact that a subject m i g h t reject a sentence af ter conceptual change is n o evidence that i t is n o t or ig inal ly analytic or a p r i o r i . So Quine's a rgument f r o m hold ing- t rue fails.

V I . A B A Y E S I A N A N A L Y S I S O F R E V I S A B I L I T Y

For ou r central example o f revisability, we can use a f ami l i a r case f r o m H i l a r y P u t n a m . L e t Cbe ' A l l cats are animals'. This m i g h t seem paradigmatical ly analytic or a p r i o r i . B u t let E specify evidence c o n f i r m i n g that that the f u r r y , apparently fe l ine creatures that inhab i t ou r houses are actually remote-cont ro l led robots f r o m Mars, whi le al l the o ther creatures we see are organic. Pu tnam argues that i f we discovered that E obtains, we w o u l d reject C. So let us suppose that Sarah accepts C at ^ 1 , acquires total evidence as specif ied by E, and rejects C a t t^.

Here , the diagnostic question is: W h a t is Sarah's in i t i a l cond i t iona l probabihty cri{C\E)}

I f cri{C\E) is low, t hen Sarah's j u d g m e n t at t^, reflects a cond i t i ona l credence that she had at ti. I n this case, the j u d g m e n t at

Perhaps there are certain strong conceptions of analyticity on which an analytic sentence cannot be rejected by any subject, rational or irrational. But these conceptions are not standard, and in any case no such constraint applies to apriority.

^^Hilar)^ Putnam, " I t Ain ' t Necessarily So," this J O U R N A L , L I X , 22 (Oct. 25, 1962):


t2 is p re f igured . Here , Sarah's accepting C i n hght o f E accords w i t h the p r inc ip le o f condi t ional iza t ion .

I f cr2{C\E) is h igh , t hen Sarah's j u d g m e n t at fails to ref lect the cond i t iona l credence that she had at ti. I n this sort o f case, the j u d g m e n t at 2̂ is p o s t f i g u r e d . H e r e , Sarah's accep t ing C i n l i g h t o f E appears to violate the p r inc ip le o f condi t ional iza t ion.

For exactly the reasons given before , the pos t f igured case requires ei ther that Sarah is n o t f u l l y ra t ional or that her use o f C undergoes conceptual change between ti and ^ . Cases o f this sort are o f no he lp to Q u i n e . A g a i n , the fac t tha t an i r r a t i o n a l subject m i g h t re jec t a sentence is no evidence that i t is no t analytic or a p r i o r i , and the fact that a subject m i g h t reject a sentence after conceptual change is no evidence that i t is no t or iginal ly analytic or a p r i o r i .

For Quine's a rgument to succeed, he needs to exclude cases o f this sort. T h a t is, he needs to make the case that any sentence can be r a t i o n a l l y revised w i t h o u t a v i o l a t i o n o f c o n d i t i o n a l i z a t i o n . T h i s requires that f o r a l l r a t iona l subjects and f o r a l l sentences S, there exists an evidence sentence £ such that cr{S\E) is low.

Th i s c l a i m is n o t so obviously false as the c o r r e s p o n d i n g c l a im about holding- t rue come what may. For this reason, one m i g h t regard the a rgument f r o m revisability as a stronger a rgument than the argum e n t f r o m holding- t rue . Indeed , supporters o f Qu ine such as Putnam and H a r m a n have concen t ra ted o n the a r g u m e n t f r o m revisability and made claims no t f a r f r o m the c la im i n question.

Sti l l , i t is n o t clear jus t what the grounds are f o r accepting the key cla im. A t this po in t , a n u m b e r o f observations can be made.

First, Quine's o f f i c i a l grounds f o r the revisability c la im involve the abil i ty to revise anci l lary claims w h e n necessary. These grounds are the same as f o r the holding- t rue c la im, and i t is clear that Qu ine sees the two as cont inuous. These grounds suggest that af ter ob ta in ing evidence, a subject cou ld use these features to revise a given sentence. B u t we have seen that revisions o f this sort typically involve violations

^̂ See Putnam, op. cit; and Gilbert Harman, "Doubts about Conceptual Analysis," in Michaelis Michael and John O'Leary-Hawthorne, eds.. Philosophy in Mind: The Place of Philosophy in the Study of Mind (Boston: Kluwer, 1994), pp. 43-48. I t may be useful to distinguish a pragmatist reading of the arguments in "Two Dogmas," which stresses the freedom to adjust ancillary hypotheses as one chooses, f r o m an empiricist reading, which stresses the role of unexpected evidence i n driving us to revise our beliefs. Roughly, where the pragmatist reading turns on the claim that one may accept or reject certain statements, the empiricist reading turns on the claim that one should (or perhaps that one would). A pragmatist reading will put equal weight on the argument f r o m holding-true and revisability, while an empiricist reading will put more weight on the latter. The pragmatist strand is more central in Quine's text, but the empiricist strand has been more influential among later Quineans.


o f c o n d i t i o n a l i z a t i o n . These g rounds do very l i t t l e to suggest tha t be fo re acqu i r i ng the relevant evidence, a subject's c o n d i t i o n a l credence cr{C\E) w i l l be low.

Second, almost any c l a im c o u l d be ra t ional ly re jec ted given testim o n y o f an apparent epistemic superior. B u t this p o i n t has no bear ing o n apr ior i ty; that a c la im cou ld be rejected i n this way is n o evidence that i t is no t a p r i o r i . T h e p o i n t also does n o t establish that any c la im is revisable unde r ideal re f lec t ion , as i t is fa r f r o m clear that this sort o f revisability applies to ideally ra t iona l thinkers. Perhaps these th inkers ' grounds f o r accepting a mathemat ica l c la im, f o r example, w i l l always defeat any evidence conce rn ing an apparent epistemic superior.

T h i r d , even i f this sort o f considerat ion applies to many apparent cases o f a p r i o r i t ruths, there are a n u m b e r against w h i c h i t has n o purchase . Some such cases i n c l u d e the m a t e r i a l cond i t i ona l s discussed i n the previous section: ' I f D, t h e n 5' (equivalently, 'D—>S'), where Z) is a lengthy specif icat ion o f an arbi t rary scenario and 5 is a sentence such as 'Water is H 2 O ' such that cr{S\D) is h igh . Assuming a f u l l y r a t i o n a l subject , i t f o l l o w s tha t cr(D^S\D) is h i g h , so t ha t cr{D—>S) is also h igh . We can stipulate that D includes o r entails a f u l l specif icat ion o f evidence that obtains i n the scenario, so that D entails E f o r any evidence sentence E that obtains i n the scenario, and that D entails -^E otherwise (setting vagueness a s i d e ) . A qu ick two-case a rgument now suggests that n o evidence E cou ld lead us to ra t ional ly re jec t D—>S. First case: i f E does n o t ob t a in i n the scenario, t h e n D entails ^E. I n this case, cr{^D\E) = 1, so cr{D^S\E) = 1. Second case: i f E obtains i n the scenario, t hen D entails E. N o w cr{D^S\E) must he between cr{D—>S\E8c-^D) and cr(D—>S\EScD). B u t the f o r m e r is 1, a n d the lat ter is j u s t cr{D^S\D), w h i c h we have seen is h i g h . So cr{D^S\E) is h i g h . P u t t i n g the two cases together , cr(D^S\E) is h i g h f o r a l l E. I m p o r t a n t l y , m a t e r i a l c o n d i t i o n a l s very m u c h l ike these are the a p r i o r i t ru ths that are most i m p o r t a n t i n the two-dimensional f r amework .

F o u r t h , once one notes tha t this a r g u m e n t allows some t ru th s S such that cr{S\E) is h i g h f o r al l E, i t is clear that there is n o longer a sound p r i n c i p l e d a rgument that f o r al l S, there is an E such that cr{S\E) is low. As a result, we may expect to find many m o r e exceptions to this claim. Indeed, many Quineans have conceded such exceptions, especially i n the domains o f mathematics and logic, and there is no reason n o t to expect many more .

One can understand entailment here in a variety of ways. For present purposes we need only the claim that i f D entails E, then cr{E\D) = 1 and cr{~D\~E) = 1 for a ful ly rational subject.


F i f t h , i t is w o r t h stressing that even i f this hne o f a r g u m e n t succeeded, i t w o u l d be m u c h more conservative than Quine 's o r ig ina l l ine . I t leads natural ly to a view o n w h i c h there is an analytic/synthetic d is t inc t ion . A t worst, i t w o u l d be the case that most or al l sentences previously regarded as analytic (a p r i o r i ) , such as ' A l l cats are animals', w i l l be recons t rued as synthetic (a p o s t e r i o r i ) . B u t one c o u l d st i l l use the cur ren t f r a m e w o r k to characterize Carnapian intensions, once one acknowledges that the intensions f o r sentences such as ' A l l cats are animals' w i l l be false at some scenarios. One w i l l still have a p r i n c ip led d is t inc t ion between cases that involve conceptual change and cases that do not . I n this way, advocates o f analyticity, apriori ty, and conceptual analysis w i l l have m u c h o f what they want.

I n any case, the Bayesian analysis has g iven us wha t we wanted: a p r i n c i p l e d c r i t e r i o n f o r i d e n t i f y i n g cases o f concep tua l change. I t has only given us a suf f ic ien t c o n d i t i o n , rather than a necessary and suf f ic ien t c o n d i t i o n , bu t this is good enough f o r our purposes. W i t h this analysis i n hand, i t is clear that Quine's arguments f r o m revisability and holding-true fa i l .

V I I . Q U I N E A N O B J E C T I O N S

Objection 1: The Bayesian Analysis Begs the Question. I t m i g h t be suggested that the Bayesian pr inc ip le (CS) that I have appealed to simply assumes a n o t i o n o f conceptual change w i t h o u t argument , and therefore begs the question against the Qu inean skeptic about this no t ion . I do n o t t h i n k that this is qui te r ight . (CS) is itself a consequence o f the p r inc ip le o f condi t ionahzat ion f o r proposi t ions and o f two other weak assumptions, n o n e o f w h i c h say a n y t h i n g abou t concep tua l change. Sti l l , this hne o f a rgument assumes a n o t i o n o f propos i t ion , about w h i c h a Qu inean m i g h t be skeptical.

N o w , Quine ' s doubts abou t p ropos i t i ons have been m u c h less i n f l u e n t i a l than his doubts about the analyt ic/synthet ic d is t inct ion. I t is clear that to get o f f the g r o u n d , Bayesian accounts o f conf i rmat i o n require e i ther something hke proposit ions or someth ing l ike the n o t i o n o f conceptua l change. Bayesian credences w i l l be assigned e i the r to abstract enti t ies such as propos i t ions , events, o r sets; to l inguistic items such as sentences; or to menta l items such as beliefs.

'^Here we can distinguish between radical Quineans, who hold that there is no analytic/synthetic distinction, and moderate Quineans, who hold that there is such a disdnction but that very few sentences are analytic. I f the analysis given here is right, we might expect this distinction to correlate to some degree with the distinction between pragmatist and empiricist Quineans mentioned in footnote 16. This seems to be what we find in practice. For example, Quine in "Two Dogmas" takes the pragmatist and radical lines, while Putnam, op. cit, takes the empiricist and moderate lines.


I f we take the f i r s t route , t hen we can use these entities j u s t as we used proposi t ions to g r o u n d a n o t i o n o f conceptual change. I f we take the second o r the t h i r d route , we need to requ i re someth ing l ike concep tua l constancy to avoid counte rexamples to p r inc ip le s such as condi t ional iza t ion .

O f course a Q u i n e a n m i g h t s imply reject Bayesianism altogether, a long w i t h the associated p r inc ip le o f condi t ional iza t ion . This w o u l d seem rash, however, as Bayesianism is an extremely successful theory w i t h widespread empi r ica l applications. So by a Quinean's o w n lights, i t is h a r d to reject i t . Fu r the rmore , even i f one rejects Bayesianism, a successor theory is l ike ly to have c o r r e s p o n d i n g p r inc ip l e s o f dia-c h r o n i c r a t iona l i ty g o v e r n i n g h o w bel iefs s h o u ld be u p d a t e d over t ime i n response to evidence. Precisely the same issues w i l l arise f o r these pr inc ip les : i f they apply to abstract i tems we can use these to d e f i n e concep tua l change, a n d i f they apply to l inguis t ic i tems o r men ta l items we w i l l require a n o t i o n o f conceptual change.

I t h i n k the deeper m o r a l is that there is a constitutive l i n k between r a t i o n a l i n f e r ence a n d concep tua l constancy. Issues such as those floated here w i l l arise f o r any p r i n c i p l e o f d iachron ic ra t ional i ty . I f i t is a p r i n c i p l e tha t f r o m A a n d A—>B one s h o u l d i n f e r B, a n d i f the premises a n d conclusions here are sentences o r m e n t a l i tems, then to avoid obvious counterexamples the p r inc ip le should requi re tha t A a n d B have the same m e a n i n g o n each occasion w h e n they occur. A n d i f the p r inc ip le applies to abstract objects such as p ropositions, these can themselves be used to def ine conceptual change. So i f we are n o t skeptics about pr inciples o f d iachronic rat ionali ty, a n o t i o n o f conceptual change w i l l be h a r d to avoid.

Objection 2: Rationality Presupposes Apriority. I t m i g h t be suggested that i n appeal ing to the n o t i o n o f rat ionali ty, the n o t i o n o f apr ior i ty is smuggled i n . For example, someone m i g h t h o l d that al l pr inciples

A related objection is that the very idea of a credence or a conditional credence presupposes conceptual constancy. After all, one's credence associated with a sentence arguably is determined by one's dispositions to make certain judgments and decisions involving the sentence, for example, the odds one would take on a bet on that sentence i f i t were offered. But in considering such dispositions, we have to assume that the meaning of the sentence stays constant f r o m the initial moment to the bet. Likewise, a conditional credence cr(M\E) arguably is determined by one's dispositions to make judgments about M conditional on the supposition of E. This requires conceptual constancy: i f meaning changes between the initial moment and the judgment, a high initial credence might go with a negative judgment. Still, any conceptual constancy needed here is at best very local, within an episode of consciousness. In any case, i f it turns out that the notion of apriority is as secure as the notion of a credence, so that the Quinean can reject the former only by rejecting the latter, that should be good enough for the defender of apriority. Thanks to Ned Block and Kelvin McQueen for discussion here.


o f r a t iona l in fe rence depend o n u n d e r l y i n g pr inc ip les about the a p r i o r i ; f o r example, perhaps an inference f r o m some premises to a conc lus ion is r a t iona l precisely i f i t is a p r i o r i tha t i f the premises obta in , the conclus ion is l ikely to obta in . O r perhaps the distinctive idea l iza t ion made by the Bayesian involves some tacit assumptions about the a p r i o r i . For example, perhaps the Bayesian r equ i remen t t h a t r a t i o n a l subjects s h o u l d have c r e d e n c e 1 i n l o g i c a l t r u t h s depends i n some way o n the bel ief that logical truths are a p r i o r i . I f so, the appeal to ra t ional principles here presupposes one o f the key notions at issue.

T h e reply here is s t ra ightforward. W h e t h e r or n o t the objector is correct that rat ional i ty depends i n some way o n aprior i ty , the appeal to r a t iona l i ty is i nnocuous i n the c u r r e n t d ia lec t ica l contex t . T h e relevant opponen ts are those w h o accept the n o t i o n o f ra t iona l i ty bu t who quest ion the n o t i o n o f apr ior i ty . M y a rgument is i n t ended to establish that if one accepts cer ta in pr inc ip les c o n c e r n i n g rat ionality, t hen one should reject Quine's a rgument against the a p r i o r i . I f this objec tor is correct , t hen the opponen t e i ther should give u p o n the pr inciples concern ing rat ional i ty or accept the n o t i o n o f the a p r i o r i . E i the r ou tcome is su f f i c i en t f o r m y purposes. I am happy to concede that i f an o p p o n e n t rejects the n o t i o n o f rat ional i ty , or rejects al l relevant pr inciples o f d iachronic rat ionali ty, then the curren t a rgument has no purchase against her.

I t is also w o r t h n o t i n g that the principles o f rat ional i ty that I appeal to are pr inciples that many or most opponents o f the a p r i o r i accept. C o n d i t i o n a l i z a t i o n has n o obvious c o n n e c t i o n to the a p r i o r i , f o r example. I do n o t know whether the special status that the Bayesian gives to logical truths has a special connec t ion to the a p r i o r i , bu t i n any case this status plays no role i n my argument . Th a t is, the argum e n t does n o t requi re the Bayesian c la im that ra t ional i ty requires credence 1 i n logical truths. I n fact, the pic ture I have sketched appears to be compat ib le w i t h a view o n w h i c h logical t ruths deserve ra t ional credence less than 1, and o n w h i c h they can be revised given relevant evidence. A l l that is r equ i red is that such a revision should obey cond i t iona l i za t ion . N o t h i n g here smuggles i n any obvious presuppositions about the a priori.'^"^

Objection 3: A Principled Line between Conceptual Change and Irrationality Cannot Be Drawn. A Quinean may suggest that our concept o f rationality

•̂̂^ Of course, my argument appeals to logical claims at various points, but this does not require that logical truths are a priori or that they are unrevisable. I t merely requires that they are true. Likewise, my argument does not require that the principle of conditionalization is a priori or that i t is unrevisable. I t simply requires that the principle is true.


is n o t fu l l y determinate, and that as a result a clear division between cases o f i r rat ional i ty and cases o f conceptual change cannot be f o u n d . Some h a r d cases, such as revising logic i n l i g h t o f q u a n t u m mechanics, are no t easily classified as either.

However, my reply to Quine's a rgument does n o t requi re d rawing a l ine here. I t suffices f o r the purposes o f the a rgument that the violations o f condi t iona l iza t ion involve ei ther i r ra t ional i ty or conceptual change, and we do n o t have to classify these violat ions f u r t h e r . I n any case, as l o n g as there are clear cases o f ra t ional j u d g m e n t , the existence o f unclear cases entails at worst a vague d is t inc t ion , n o t a nonexis tent d is t inc t ion .

Objection 4: The Argument Requires Constancy in Evidence Sentences. Recal l the first res idual issue f o r the f r a m e w o r k o f in tensions discussed earlier: the f r a m e w o r k assumes conceptual constancy i n the base vocabulary, so the f r a m e w o r k canno t exp la in this concep tua l constancy. O n e m i g h t t h i n k t ha t an ana logous issue arises w i t h respect to the ev iden t ia l vocabulary used to specify evidence sentences such as E. A f t e r a l l , cond i t iona l i za t ion concerns wha t to do w h e n one has a certain credence cr{S\E) and then learns E. T h e cond i t i o n a l credence is i n par t an a t t i tude to a sentence E, and wha t one learns is also a sentence E. O n e m i g h t t h i n k i t is r equ i r ed that the sentence have the same mean ing o n b o t h occasions. I f so, t hen any apparent failures o f condi t ionahzat ion i n a ra t ional subject c o u l d be b l amed o n a change i n the mean ing o f terms i n E, instead o f a change i n the mean ing o f terms i n 5, and i t is no t clear that we have a p r i n c i p l e d way to choose.

As i t stands, this p i c tu re is n o t qu i te r i gh t . L e a r n i n g E does n o t typically involve the sentence E at a l l . Perhaps i f l ea rn ing were always by testimony, and i f E were a sentence used i n testimony, t hen the issue w o u l d arise. B u t f o r o u r purposes we can assume that the re levan t l e a r n i n g is by p e r c e p t i o n . H e r e , E w i l l be a sentence character izing the evidence that one learns, and the l ea rn ing process need n o t involve this sentence at a l l . So there is no use o f E at ^ that needs to be ahgned w i t h the use o f £ at ti. A t best we need to r equ i r e tha t E as used at ti correct ly applies to the evidence acqui red at ^ . B u t this is a m u c h weaker requi rement , conce rn ing only the extension o f E as used at ti, w i t h no role f o r any use o f E at ^ .

St i l l , i t can be a rgued tha t a c q u i r i n g evidence requires hav ing ce r t a in attitudes to the evidence. For example , the r a t iona l i ty o f Bayesian cond i t iona l i za t ion o n new experiences arguably requires n o t j u s t that one has the experiences, b u t that one is cer ta in that one has t hem. I f so, one m i g h t suggest that the f r a m e w o r k tacitly


requ i res tha t at one is c e r t a in o f the evidence s ta tement E (which says that cer ta in experiences obta in) . This issue is starker i n ahernative f rameworks such as Je f f rey condi t ionahzat ion , w h i c h accommodate uncer ta inty about evidence by giving an expUcit role to one's credence i n evidence statements such as E at /2- Does this require some sort o f constancy i n the mean ing o f E af ter all?

T h e issue is del icate . For the reasons g iven above, I t h i n k tha t the sentence E as used at plays n o essential ro le here. However, i t is arguable that at t^, the subject must be certain o f (or have other appropr ia te att i tudes to) the evident ia l p ropos i t i on expressed by E at ^ i : the p ropos i t i on that cer ta in experiences ob ta in , f o r example. O r w i t h o u t i n v o k i n g proposi t ions: subjects must be cer tain that the relevant evidence obtains ( that they are having certain experiences, say), where this is the same evidence c o n c e r n i n g w h i c h they had cond i t iona l credences at ^ i . W i t h o u t this a l ignment , one cou ld always respond to an apparent fa i lure o f condi t iona l iza t ion by saying that a l though a subject's i n i t i a l credence was cond i t iona l o n evidence e obta in ing , and a l though evidence e later obtained, the subject i n fact became certain that some other evidence ^* obtained. I f this were so, there w o u l d be n o v io l a t i on o f cond i t iona l i za t ion ( the subject w o u l d n o t acquire the evidence e), a n d there arguably w o u l d be no i r ra t ional i ty .

T h i s r e q u i r e m e n t o f a l i g n m e n t provides some r o o m f o r the Qu inean to maneuver, bu t the r o o m is extremely l im i t ed . T o e l imi nate this r o o m altogether, we need only suppose that we have a g r ip o n what i t is f o r a subject to accept o r suppose that certain evidence obtains. W i t h this m u c h g ran ted , we can s imply s t ipulate tha t f o r o u r purposes, the c o n d i t i o n a l credences cr{S\E) re levant at are credences i n S cond i t iona l o n the evidence that is actually obta ined at 2̂- Th i s removes any loopho le , and does so w i t h o u t m a k i n g any assumptions abou t constancy i n the m e a n i n g o f language across t ime. A t most, we have to assume an unders tanding o f certain beliefs and suppositions about evidence.

The r equ i r ed assumptions can be reduced f u r t h e r by n o t i n g that f o r ou r purposes, evidence can be h m i t e d to experiences, or at least to observa t iona l states o f a f fa i r s . W h i l e there is a sense i n w h i c h emp i r i c a l knowledge o f nonobservat ional states o f affairs can serve as ev idence f o r o t h e r c la ims, i t is p laus ib le t h a t th is k n o w l e d g e is g r o u n d e d i n evidence c o n c e r n i n g expe r i en t i a l o r observat ional matters. O n a Bayesian view, ou r credences i n these states o f affairs t h e n must m a t c h those d e t e r m i n e d by cond i t iona l i za t ion o n exper ien t ia l or observational matters. I t h i n k i t is also plausible that credences i n observational states o f affairs should match those determined


by condi t ional iza t ion o n exper ient ia l matters.^^ I f the latter c la im is granted, t hen f o r present purposes we can restrict the relevant evidence i n cases o f revisabihty and holding-true to experient ia l states. Even wi thou t i t , we can restrict the relevant evidence to observational states. So to answer the Qu inean worry, we need only suppose that we have a g r ip o n what i t is f o r a subject to accept or suppose that cer ta in exper ient ia l o r observational states o f affairs obta in . This is something that Quine's arguments i n "Two Dogmas" do no t give us reason to doubt .

T h e upshot o f al l this is that the residual issues about a base vocabulary are n o t e l imina ted al together o n a Bayesian approach, b u t they are m i n i m i z e d i n a way that brings ou t the severe costs o f the Qu inean posi t ion. A Quinean who rejects the notions o f analyticity and apriori ty a long present lines also must insist that there is no objective fact o f the matter about whether a subject accepts or supposes that a given observational state obtains. This view goes a long w i t h a generalized skepticism about the contents o f thought , perhaps i n the spir i t o f Quine's skepticism about m e a n i n g developed i n his arguments c o n c e r n i n g radical translation. I t likewise requires a certain skepticism about diachronic rationality, f o r reasons discussed earlier.

Qu ine h imse l f argues b o t h f o r skepticism about mean ing ( i n Word and Object) a n d f o r a sort o f skept ic ism about n o r m s o f r a t iona l i ty ( i n "Epistemology Naturalized"^^). Few have been prepared to f o l l o w h i m here, and even those w h o sympathize w i t h the Q u i n e o f "Two Dogmas" have tended to reject these later views. O f course Quine 's arguments f o r these views deserve a t t en t ion i n the i r o w n r igh t , b u t i t is clear that the arguments i n "Two Dogmas" provide l i t t le d i rec t suppor t f o r t h e m . Sti l l , the cu r r en t analysis suggests a deep l inkage be tween these views. D e f e n d i n g the a rguments o f "Two Dogmas" against a cer tain appeal to conceptual change leads natural ly to skept ic i sm abou t d i ach ron i c r a t i ona l p r inc ip le s a n d abou t the c o n t e n t o f language and thought . Contrapositively, once even m i n i m a l claims

For example, i f one is fully rational, one's credence that there is a red square in f ron t of one should match one's antecedent conditional credence that there is a red square in f ron t of one given that one is having an experience as of a red square. ( I f norms of radonaUty do not ensure certainty about the experiences one is having, one can move to a Jeffrey-conditionalization analog.) Theses of this sort have been denied by some dogmatists about perception (for example, James Pryor, "Uncertainty and Undermining," URL: wwwjimpryor.net/research/papers/Uncertainty.pdf). They also might be denied by some who think that perceptual knowledge is more secure than introspective knowledge (for example, Eric Schwitzgebel, "The Unreliability o f Naive Introspection," Philosophical Review, c x v n , 2 (April 2008): 245-73).

Quine, "Epistemology NaturaUzed," i n Ontological Relativity and Other Essays (New York: Columbia, 1969).


about rat ional i ty and though t are accepted, the arguments I have considered against analyticity and apr ior i ty dissolve.

Objection 5: There Can Be Rational Revision by Resetting Priors. Quineans o f a p ragmat i s t str ipe o f t e n appeal to the u n d e r d e t e r m i n a t i o n o f theory by evidence: mu l t i p l e theories are consistent w i t h the same evidence, and we have considerable lat i tude i n choosing between them. I n the Bayesian f ramework , where theory is de te rmined by evidence a l o n g w i t h p r i o r p robab i l i t i e s , this u n d e r d e t e r m i n a t i o n comes to u n d e r d e t e r m i n a t i o n o f probabi l i t ies that are p r i o r to any evidence. Th i s u n d e r d e t e r m i n a t i o n yields a p o t e n t i a l way tha t f u l l y r a t i ona l subjects m i g h t violate condi t ional iza t ion w i thou t conceptual change.

The relevant m e t h o d here is that o f resetting priors . This m e t h o d stems f r o m the observat ion that most Bayesians al low that there is some f l ex ib i l i t y i n one's u l t imate pr iors : the p r i o r probabi l i t ies that a subject should have before acqu i r ing any empi r i ca l evidence. ( O f course these p r io r s are s o m e t h i n g o f a f i c t i o n . ) For example , o n Carnap 's f r a m e w o r k f o r i n d u c t i v e log ic , equal ly r a t i o n a l subjects may have d i f f e r e n t values f o r X, the parameter that guides how quickly the subjects adjust the i r beliefs i n l i gh t o f inductive evidence, and this d i f fe rence can be traced to a d i f fe rence i n ul t imate priors. Two such subjects m i g h t acquire exactly the same evidence over t ime , whi le be ing l ed to quite d i f f e r e n t posterior probabil i t ies . I f G is the thesis that human-caused g loba l w a r m i n g is o c c u r r i n g , f o r example , one subject m i g h t be l ed to a h i g h credence i n G, whi le another m i g h t be l ed to a low credence i n G.

Now, a subject w i t h a h i g h credence i n G m i g h t ref lect and observe that her h i g h credence is traceable entirely to the value o f X i n her u l t imate priors , and that this value was quite arbitrary. She may note that i t w o u l d have been equally ra t ional to start w i t h a lower value o f X and to end u p w i t h a lower credence i n G. A t this po in t , a b o l d subject m i g h t choose to change her credences wholesale. A t least i f she has a g o o d enough record o f her evidence, she can " u n w i n d " back to the u l t imate pr iors , reset 1 to a lower value, and reintegrate all the evidence by condi t ional iza t ion . She w i l l end up w i t h a new set o f credences, i n c l u d i n g ( a m o n g many o the r d i f fe rences ) a lower value f o r G.

A Q u i n e a n m i g h t suggest tha t there is n o t h i n g i r r a t i o n a l about d o i n g this, and that this m e t h o d m i g h t be explo i ted i n order that a subject c o u l d h o l d o n to almost any sentence come what may and also revise almost any sentence. A f t e r al l , f o r most nonobservational empi r ica l sentences S and most paths o f evidence, there is some u l t i mate p r i o r that w i l l lead to a h i g h credence i n S and some ul t imate p r i o r tha t w i l l l ead to a l o w credence i n 5. N o n e o f this requires


conceptua l change. So violat ions o f cond i t iona l i za t ion i n a r a t iona l subject do n o t provide a suf f ic ien t c o n d i t i o n f o r conceptual change af ter a l l .

Th i s pos i t ion requires r e j ec t ion , or at least revision, o f o r t h o d o x Bayesianism. O n the o r t h o d o x view, condi t iona l iza t ion is a constraint o n d iachronic rat ionali ty, a n d this sort o f revision w i l l be i r r a t iona l . F u r t h e r m o r e , the view tends to l ead to an anything-goes v iew o f ra t iona l belief. I f there are no constraints o n ul t imate pr iors , the view entails that at any m o m e n t , i f cr{p) < 1, t hen one's credence can be rat ional ly revised so that cr{p) is a rb i t rar i ly close to zero. A n d even i f there are constraints o n u l t ima te p r io r s , these constraints mus t be weak e n o u g h to vindicate the large violat ions o f cond i t iona l i za t ion that the Q u i n e a n a rgument requires, l ead ing natural ly to a view o n w h i c h most beliefs can be rat ional ly revised at any m o m e n t i n t o disb e l i e f Given this m u c h , i t is n o t easy to see how beliefs can consti tute knowledge at all.^^

Fur ther , i t is fa r f r o m clear that al l beliefs can be revised i n this way. For example, given that logical behefs, mathemat ica l beliefs, and eviden t ia l statements are constra ined to have credence 1, this m e t h o d w i l l n o t y i e ld revisabil i ty f o r these beliefs . M o r e general ly, there is n o t m u c h reason to h o l d that i t w i l l y i e ld revisions to those beliefs usually classified as a p r i o r i ( 'A l l bachelors are unmar r i ed ' , say), most o f w h i c h appear n o t to depend o n u l t imate priors . So this response is weakest where i t needs to be strongest.

Mos t f undamen ta l l y : as l o n g as we have a concep tua l d i s t i nc t ion between cases i n w h i c h beliefs are revised by this process and cases i n w h i c h they are no t , we s t i l l have e n o u g h to d raw a d i s t i n c t i o n between those violations o f condi t iona l iza t ion that involve conceptual change a n d those that do not . T h e Q u i n e a n w i l l have to insist that we do n o t have a g r ip o n this conceptua l d i s t inc t ion , so that there is n o d i s t inc t ion to be d rawn between cases o f reset t ing pr io rs a n d cases o f conceptual change. I t h i n k there is htt le reason to accept this. F u r t h e r m o r e , even i f this l i ne were accepted, i t w o u l d lead once again to an across-the-board skept icism abou t p r inc ip les o f b e l i e f u p d a t i n g and o ther f o rms o f d iachronic rationahty. So i f pr inciples o f d iachronic rat ionahty are a l lowed at al l—even the l ibera l p r i n c i ples suggested by the cu r ren t approach—then the d is t inc t ion between conceptual constancy and conceptual change remains intact.

In addition, this method is a sort of beUef revision that is not driven by evidence at all. So although this line of reasoning is perhaps the best way of preserving the pragmatist reading of Quine's arguments in light of the present analysis, it does not sit easily with the more influential empiricist reading.


Objection 6: Subjects Need Not Have Conditional Credences. I t m i g h t be objected that the Bayesian analysis requires the assumption that f o r every sentence S used by a subject and every possible evidence sentence E, the subject has a c o n d i t i o n a l credence cr{S\E). B u t this is an unrealistic ideal iz ing assumption.

I n response: the ideahzation is no t enormous. For most S and most E, the subject w i l l have some relevant dispositions involv ing S and E, f o r example, invo lv ing her willingness to accept various bets involvi n g S and E. I n many cases, these dispositions w i l l hne u p i n a clear enough way that cr{S\E) w i l l be h i g h . I n other cases, they w i l l l ine u p i n a clear enough way that cr{S\E) w i l l be low. I n o ther cases, the dispositions may be enough o f a m i x that i t is ha rd to say.

Quineans m i g h t suggest that i f cr{S\E) is indeterminate i n this way, and the subject later rejects S u p o n l ea rn ing E, this should no t count as a v io la t ion o f condi t ionahzat ion. I f so, they then m i g h t suggest that f o r any S, there is some E such that cr{S\E) is indeterminate i n this way, and such that the subject c o u l d later re ject S o n l ea rn ing E w i t h o u t v io la t ing condi t ional iza t ion. Perhaps this sort o f revisabihty is enough f o r thei r purposes?^^

I do no t t h i n k that this is enough, however. Cases o f this sort seem to t u r n essentially o n the sub jec f s n o t b e i n g f u l l y r a t i ona l . I f the subject is f u l l y ra t ional , t hen her dispositions to accept 5 o n supposing E and o n l e a r n i n g E shou ld be the same, assuming n o conceptua l change. T h a t is, i f a f u l l y r a t i o n a l subject rejects S o n l e a r n i n g E and t h i n k i n g things t h rough , then i f she were to have been presented in i t ia l ly w i t h the supposit ion that E and had though t things t h rough , she should have rejected S cond i t iona l o n that supposit ion. T o f a i l to meet this cond i t i on is a fa i lure o f f u l l rat ionali ty, j u s t as is an ord inary v io la t ion o f condi t ionahzat ion. So at best the Qu inean has presented us w i t h a k i n d o f revisability that can be exp lo i t ed on ly by subjects w h o are less than f u l l y ra t ional . L ike the sort o f revisability that can be e x p l o i t e d on ly by i r r a t i o n a l subjects, this sort o f revisabil i ty has no bear ing o n matters o f apriori ty.

V I I I . C O N C L U S I O N

Quine is r i gh t that any statement can be he ld true come what may and that no statement is i m m u n e to revision. Bu t as Grice and Strawson

'^'^It is especially likely that ordinary subjects will lack credences cr{S\D) involving the scenario specifications D discussed earlier, due to the enormous size of these specifications. This observation does not affect the use of conditional credences involving D to define intensions, as the credences used there are always idealized rational credences cr'{S\D), for which the current issue does not arise. Where nonidealized credences are concerned, these cases will not yield cases of revisability along the lines in the text, because the subject will be incapable of learning that D.


observe, these phenomena are quite compatible w i th a robust analyt ic/ synthetic dis t inct ion and a robust n o t i o n o f meaning. A Bayesian analysis reveals that Quine is no t r igh t that any statement can be he ld true come what may wi thou t conceptual change or irrationahty, and hkewise f o r revision. We can p i n down the dis t inct ion between cases that involve conceptual change and cases that do no t using ei ther the m e t h o d o f intensions or Bayesian analysis.

T h e m e t h o d o f intensions characterizes intensions i n terms o f certa in idealized dispositions, and uses this n o t i o n to dist inguish between cases tha t involve conceptua l change and cases that do no t . I n the centra l version that I examined , this m e t h o d assumes the n o t i o n o f apr io r i ty , so i t does n o t g r o u n d tha t n o t i o n independen t ly . St i l l , i t shows h o w a f r a m e w o r k invo lv ing apr io r i ty can accommodate a l l o f Quine 's data. A n d f o r the same reasons that most phi losophers reject Quine ' s a rguments i n the f i r s t f o u r sections o f "Two Dogmas," n o independen t g r o u n d i n g is requi red .

The Bayesian analysis takes a step f u r t h e r and defends the a p r i o r i o n partly independent grounds. This analysis assumes the n o t i o n o f condit ional probabihty and the normative n o t i o n o f rationahty to provide condi t ions f o r conceptua l change, b u t i t does n o t assume the n o t i o n o f apriority. I n effect, constitutive connections between rat ional inference and conceptual change are used to make inroads in to the Quinean circle.

T h e conclus ion should no t be too strong. WTiile I have responded to Quine 's arguments against the a p r i o r i and the analytic, I have n o t p r o v i d e d a posi t ive a r g u m e n t f o r the a n a l y t i c / s y n t h e t i c o r the a p r i o r i / a pos ter ior i d is t inc t ion , and I have no t t r i ed to g r o u n d these not ions i n whol ly independen t terms.

One migh t be tempted to take things a step f u r t h e r still, and at tempt to def ine aprior i ty i n terms o f condi t ional probabi l i ty and rationahty. For example, one m i g h t suggest that a sentence 5 is a p r i o r i f o r a subject precisely w h e n the ideal condi t ional probabihty cr{S\D) is 1 (or: is h igh) f o r a l l scenario specifications D. Bu t there wi l l be residual issues. For a start, i t is no t clear that one can def ine the class o f scenario specifications wi thou t using the n o t i o n o f apriori ty. I n addi t ion , the thesis is subject to various potent ia l counterexamples; f o r example, one m i g h t argue that w h e n 5 is a mathematical t r u t h and D specifies a scenario i n w h i c h one is a poor mathematical reasoner, cr{S\D) should be m u c h less than 1. M u c h more w o u l d need to be said to handle these issues.^''

I address the second issue in Constructing the World, arguing that there is an idealization on which cr{S\D) in this sort of case should be 1, where this idealization can be understood without appeal to the a pr ior i . I think that the role of the a pr ior i with respect to the first issue is more robust.


Sti l l , we have seen that these not ions can he lp us at least i n diagnosing issues regarding meaning, conceptual change, and the a p r i o r i . A n d we have seen enough to suggest that Quine's arguments i n the final sect ion o f " T w o Dogmas o f E m p i r i c i s m " do n o t th rea ten the d is t inc t ion between the analytic and the synthetic, o r the d is t inc t ion between the a p r i o r i and the a poster ior i .

D A V I D J . C H A L M E R S

Austral ian Nat iona l University and New York University

the journal of philosophy - david chalmersconsc.net/papers/revisability.pdf · the extraordinary...

Documents