Mind and Meaning 0521338263, 9780521338264

Is linguistic meaning to be accounted for independently of the states of mind of language users, or can it only be expla

911 125 83MB

English Pages 284 [142] Year 1986

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Mind and Meaning
 0521338263, 9780521338264

Citation preview

CAMBRIDGE STUDIES IN PHILOSOPHY

ND

BRIAN LOAR

STUDIES IN

CAMBRIDGE

Genera/editor Advisory G I L B E R T J O N A T H A N

editors } .

E .

H A R M A N , L E A R ,

J O H N

j .

S Y D N E Y

S H O E M A K E R

A L T H A M ,

M A R T I N

SIMON

H O L L I S ,

P E R R Y ,

J A M E S C A R G I L E Paradoxes:

T .

PHILOSOPHY

J .

B L A C K B U R N ,

F R A N K

S M I L E Y ,

J A C K S O N ,

B A R R Y

S T R O U D

a study inform and predication

P A U L M . C H U R C H L A N D Scientific Realism and the plasticity of mind N . M . L . N A T H A N Evidence and assurance WILLIAM LYONS

Mind and meaning Brian Loar Associate Professor, School of Philosophy,

University

of Southern

California

Emotion

P E T E R S M I T H Realism and the progress of science J . P. R O S S Portraying analogy DAVID HEYD

Supererogation

P A U L H O R W i C H Probability and evidence E L L E R Y E E L L S Rational decision and causality H O W A R D R O B I N S O N Matter and sense E . H . B O N D Reason and value D. M. A R M S T R O N G

What is a law of nature?

H E N R Y K Y B U R G J R Theory and measurement N . J . H . D E N T The moral psychology of the virtues M I C H A E L H . R O B I N S Promising, intending, and moral autonomy

The righl of ihc Univcrsily of Cambridge to prim and sell all manner of books was granted by Henry Will in 1534, The University has printed and published toiitimiously since 1584.

Cambridge University Press Cambridge London

New

York

Melbourne

New Sydney

Rochelle

Published b y the Press Syndicate o f the U n i v e r s i t y o f Cambridge The Pitt B u i l d i n g , T r u m p i n g t o n Street, Cambridge C B 2 I R P 32 East 5 7 t h Street, N e w Y o r k , N Y 10022, U S A 10 Stamford Road, O a k l e i g h , M e l b o u r n e 3166, Australia © Cambridge U n i v e r s i t y Press 1981 First published 1981 First paperback edition 1986 Printed i n Great B r i t a i n at the U n i v e r s i t y Press, Cambridge British Library Cataloguing in Publication Data Loar, B r i a n M i n d and meaning. (Cambridge studies i n philosophy) I . Intellect 2. Psycholinguistics I. Tide 128'.2 ISBN ISBN

BF431

81-6121

o 521 2 2 9 5 9 rationality correspond structurally to purely physical constraints pj^ysj^al states. A n d there is no i m p e d i m e n t to that. For constraints^^^^^^^^^ ^ ^ related as the theory says the belief physica^ ^ belief that ~ p are counterfactually related is, i n that ^^jg •£ ^ were t o occur then y w o u l d not P^^^' j j j g rationality constraints generate a vast n e t w o r k o f such '"''"'^'^ rfactual relations among physical states, ultimately w i t h the effect ' ' f describing a system o f physical state types whose counterfactual 'nterrelations m i r r o r the relevant logical relations among beliefs and '^^When Davidson writes that there is no echo o f the rationahty constraints i n physical theory, and speaks o f the disparate commitments o f the mental and the physical, there is the suggestion o f something like the f o l l o w i n g argument. In the f o r m u l a t i o n o f the laws o f the physical sciences there is n o t h i n g that resembles the constraints o f rationality. Therefore no reduction o f rational psychology t o physical science is possible, since reduction requires a subsumption o f the laws o f one theory under the laws o f the other. Perhaps there is an i m p o r t a n t kind o f inter-subject reduction that is law—law i n the classic sense. B u t it is quite unmotivated to infer f r o m the absence o f a psychophysical reduction o f that k i n d , that " c o m p l e t e understanding o f the w o r k i n g s o f the body and brain w o u l d not constitute knowledge o f thought and action". For functionalist reduction is not law—law reduction, w h i c h can be seen as follows.

N o w t o the m a i n question, posed first under an innocuous guise. Can a functionalist theory, that each attitude-sentence ascribes a second-order property expressible i n a physical vocabulary, recognize the constitutive force o f rationahty ? O f course i t can, and must. Let us anticipate i n a rough and general way the functional interpretation o f attitude-ascriptions o f chapter 4. The idea is that each predicate o f the f o r m ' z believes that p' (for variable z and specific p) ascribes a distinct functional state, w h i c h is determined b y the predicate's role i n a theory that includes constraints o n rationality (as w e l l as certain perceptual i n p u t conditions, and o u t p u t conditions to decision and behavior). The theory is counterfactual i n f o r m , and w h a t is required o f first-order state-types i f they are to have the right functional roles is that they stand to each other, to input and to o u t p u t , i n those counterfactual relations o f causation, transition and co-occurrence that the theory requires. In particular, a system o f physical state-types satisfies the constraints o n rationality provided they are all related counterfactually as the theory says the beliefs to w h i c h they correspond ought rationally to be related.

The functional organization o f the physical states o f a person results from many contingencies o f physiology o f basically t w o kinds. First, even those structural features o f the anatomy o f the brain that are genetically determined are not a matter o f law, for any " l a w s " o f genetics w o u l d be quite conditional i n f o r m . Secondly, as has already been pointed out, the psychological functional roles o f brain states may well be a matter o f developmental contingencies. So, that a system o f physical states is organized functionally according to the constraints o f rationality w o u l d be a quite non-nomological fact. N o t h i n g like the constraints o f rationahty shows up i n physical theory •,hut counterfactual constraints that are isomorphic t o rationahty constraints may show up in the physical facts.

This must be sayable i n a physical vocabulary, and the question arises h o w we can get r o u n d propositions to achieve that happy result. B u t

H o w can contingent facts about a physical system amount to rationality, whose ingredients are far f r o m contingent? It is a priori that

22

23

i f certain states are to be counted as beliefs and desires they must satisfy the constraints o f rationality. B u t that they do satisfy them can be as contingent as y o u l i k e ; i f they fail to do so they are not beliefs and desires. It is a fallacy to argue that, since rationality has constitutive force, a physical system w o u l d have to conform to rationality non-contingently for its w o r k i n g s to constitute the workings o f a" mind. I have been taking the key arguable point i n Davidson's argument for the irreducibihty o f the mental to be that n o t h i n g i n physical theory corresponds to rationality. B u t others may have located the crucial point elsewhere — namely i n the claim that rationality has constitutive force in rational psychology, w h i c h implies a certain unrevisahility for psychology. For i t is possible to read Davidson's argument thus: there can't be psychophysical laws, because no psychophysical generalizations can be indefinitely revisable, or " h o m o n o m i c " , w h i c h is a key feature o f scientific laws. The requirement o f rationahty is an impediment t o revisabihty i n certain directions. A l t h o u g h I do not t h i n k this is Davidson's main argument, i t deserves discussion. First, w e must distinguish revisabihty and revisabihty w i t h o u t change o f subject matter. That rationality is constitutive i n psychology rules out the indefinite revisabihty o f psychophysical generalizations w i t h o u t change o f subject matter, w i t h o u t their ceasing to be about beliefs and desires. B u t is that sort o f revisabihty really a constraint on being a law? I t w o u l d also rule out inter-subject laws connecting narrower and broader theoretical frameworks, w h e n the initial constraints o f one framework are so entrenched i n our semantics that their revision counts as a change o f subject matter. Accepting a generalization as indefinitely revisable w i t h o u t change o f subject matter is not a necessary condition o f accepting i t as b o t h true and projectible. Suppose basic theoretical physical concepts and certain laws governing them were as entrenched i n c o m m o n sense as rational psychology; the laws o f physics w o u l d not then be indefinitely revisable w i t h o u t " c h a n g i n g the subject". B u t w e could still regard them as true and as projectible. Secondly, the functionalist reply to Davidson does not deny the constitutive force o f rationality or his claim that psychophysical generalizations are not revisable w i t h o u t changing the subject. O n the contrary, the functional theory I envisage takes that as quite central. A n d t h i r d l y , the argument f r o m unrevisahility has a far less interesting conclusion than the one w h i c h seems to be Davidson's basic p o i n t ; for i t w o u l d show at most that there are no psychophysical laws. That w o u l d not i m p l y that psychophysical reduction is n o t o n the cards, for functionalist reduction

24

not psychophysical laws, but psychophysical t y p ^ t y p e requires. ^^i^j-jyized to individuals at times. So the non-existence o f correlation* -j^jg . ^ j t h our being able to read off, f r o m a person's laws IS anization and state-tokens, a w h o l e mental life, physica^^^^^^^.^^ j^3S f j g g j only obliquely a certain epistemological point dson's argument - namely, that we w o u l d never be in a position i " ^^Jentiy to regard a psychophysical correlation as lacking exceptions 'are required by consistency and cogent rationalization. O n a f nctional reduction the problem is not to verify one by one orrelations between single mental and physical properties. Functionahst reduction is rather holistic. W h a t is essential is whether a w h o l e system o f physical states exhibits a certain overall counterfactual structure. Moreover, functionalism does not i m p l y that we could ever i n practice discover or master all psychophysical correlations for a person at a t i m e ; perhaps the brain's w o r k i n g s are too complex for any o f us to have a total grasp o f them. W h a t is essential is evidence that there are such correlations, even t h o u g h i t is not evidence for specific correlations. Finally, reducibility is more a metaphysical than an epistemological matter. The key question regarding the rationality constraints is whether truth conditions for attitude-sentences can take them as constitutive, and at the same time describe circumstances o f physical organization that w o u l d appropriately satisfy them. B u t given the t r u t h conditions, the epistemology falls i n t o place. It is quite the w r o n g approach to consider unexplicated psychophysical generalizations, and to wonder about getting evidence for their strict t r u t h and p r o j e c t i b i l i t y : the o v e r w h e l m i n g l y i n t i m i d a t i n g difficulty o f the empirical question could seem an a priori impediment. The fundamental point is that there IS no incoherence i n the idea that certain physical states meet the demands o f rationahty.

25

2

is somehow^ true, but 1

tfrwU

Explicating attitude-ascriptions

Vh

nts ^

intentionahty corresponds to objective

and

determinatc

constitute the behef

t justifies the relational assumption? Consider first these t w o ( i ) Attitude-ascriptions have their meaning

^^^(_^esiTe Whether

relational - e.g. to an internal "sentence". T h a t m i g h t be jmphed b y calling belief-ascriptions " r e l a t i o n a l " . For jhat-clause may be quite extrinsic to those natural facts

from

the

theory, and so they must logically interact w i t h i t

opriately. It consists o f generalizations, w i t h variables i n place o f

Tat^lauses-

But, assuming

first-order

variables (on w h i c h

mo re

features o f our psychology and its extramental connections depends on

shortly), the relational construal follows. (2) T o explicate attitude-

w h a t naturalistic sense can be made o f propositional attitude-ascriptions,

ascriptions

sentences o f the f o r m ' x beheves that 5'. N o t every current theory o f

attitude-ascriptions true o f us, to assign them t r u t h conditions o f a

is to point to natural properties

w h i c h could

make

belief focuses on ascriptions o f belief For some, that w o u l d smack o f

certain sort. That, it seems, requires the devices o f a Tarski-type t r u t h

a concern for philosophical analysis, a supposedly discredited enterprise;

theory, which means f i n d i n g logical structure i n ' believes t h a t . . . ' ; each

but there is also something m or e specific. A naturalistic explication o f

such predicate must be logically complex. B u t , since ' . . . ' is not

' X believes s' is possible o n l y i f interpersonal non-extensional synonymy

truth-functional, and given the first-order assumption, that-clauses are

is possible. B u t i t may be t h o u g h t that the latter has been shown by

either complex names ( o f sentences or propositions), or demonstratives

Quinean arguments not to make naturahstic sense, and hence that t r y i n g

referring to the utterance o f ' . . . ' , or sequences o f names (which w o u l d

to explicate attitude-ascriptions is pointless. (This w o u l d still a llow

permit taking the that-clause to name the sequence o f the things thus

scope for saying w h a t beliefs are.)

named).

I n 2.5 I discuss a certain sort o f theoretical activity, conservative

N o w consider this reply to the first argument. " W e could take those

exphcation, w h i c h appears inevitable and sensible regardless o f one's

generalizations to be axiom

views about analyticity. As for interpersonal s y n o n y m y ,

not a variable, and universal quantifiers ' ( V ; j ) ' ' (Vq)' may be dropped

I shall be

schemata;

the 'p'

i n ' x believes that p'

is

proposing, pace Q u i n e , a physicahst conservative explication o f attitude-

without loss. The relevant interaction o f the theory and a particular

ascriptions.

attitude-ascription is just a matter o f the ascription's relations t o all other

Concentrating

o n attitude-sentences

is inevitable;

the

content o f beliefs is whatever that-clauses ascribe. T o account

for

intentional forms o f psychological

explanation is to account for a

certain sort o f theoretical discourse. This is not to say that, were a

instances o f the schemata. The systematic connections between that behef and other beliefs and desires w i l l thereby be c a p t u r e d . " This can be reinforced "Why

by

the f o l l o w i n g reply t o the second

naturalistic conservative explication o f attitude-ascriptions not f o r t h -

argument.

can't 'believes t h a t . . . ' be non-complex?

c o m i n g , no m o r e radically eliminative reconstruction could go part

infinitely many predicates have t o be accommodated is not obvious,

That

o f the w a y towards answering the question. Explication is a matter

for i f ordinary mortals are capable o n l y o f behefs o f finite c o m p l e x i t y , the set o f belief-ascriptions to be accounted for is finite. T h e n every

o f degree.

such (superficially complex) predicate can be a non-complex theoretical 2.1

BELIEFS

AS

RELATIONS

term whose meaning is determined by the " t h e o r y " that results f r o m conjoining all instances o f the a x i o m schemata. There exists a lucid

that is, sentences o f the f o r m

evice for generating explicit definitions o f theoretical terms o n the

'x believes that 5' have the f o r m B( x , that-s), where the that-clause

asis o f h o w they occur i n a theory - namely, D a v i d Lewis's m o d i f i -

I take belief-ascriptions to be relational;

as a whole is a singular te r m (or, as o n Davidson's theory (1969), ' t h a t '

cation o f Ramsey's m e t h o d . " ( T o be discussed i n chapter 3;

is a singular t e r m , referring to the utterance o f the rest). This does not

1970.)

i m p l y that i f a belief is a physical state its underlying physical structure 26

Lewis,

Apart f r o m a denial o f the first-order assumption, this conjunction 27

o f replies constitutes, as far as I k n o w , the o n l y w a y o f avoiding thesis that belief-ascriptions are relational. Is i t a reasonable theory; W e l l , i t is rather artificial; its m a i n attraction w o u l d be its avoiding the relational interpretation i f that were desirable. Otherwise it ,^ unattractive. According to i t , w h e n we seem to be generalizing about beliefs and desires, we are really talking about belief- and desir^ ascriptions. Moreover, all the structure i n our generalizations about behefs and desires becomes an accident o f the syntax o f those simple theoretical predicates 'x believes t h a t . . . ' . B u t surely i t can't be an accident that persons w h o believe that s^ and S2 tend not to believe also that not Si; there must be a principle. A n d the i m p l i e d account o f our mastery o f the concepts o f belief and desire is o d d ; for the meaning o f ' x believes that snow is w h i t e ' is supposed to depend upon a vast conjunction o f schema instances. N o w , the second part o f the theory, w h i c h treats each predicate o f the f o r m ' x believes t h a t . . . ' in terms o f its role i n a theory interpreted a la Ramsey, is on the right track; but on the account o f chapter 4, n o r m a l intuitions about generalizations about beliefs are preserved, w i t h i n a relational framew o r k . Perhaps i t is bound to be inconclusive to argue i n the abstract for a relational theory w i t h o u t a specific proposal. B u t given the first-order assumption i t is the intuitive w a y o f treating generalizations about beliefs and the p r o b l e m o f systematic t r u t h conditions for behef-ascriptions. As for the first-order assumption, let me say this. Alternative ways o f generahzing, substitutional and higher-order quantification, make perfectly good sense; b u t b y m y lights that is because their semantics is i n a first-order metalanguage. Higher-order quantification is supposed by some to be ontologically conservative — that is, to avoid c o m m i t ment to properties or propositions as entities. B u t naturally that is not so i f one assumes a first-order semantic metalanguage w h i c h is not ontologically neutral. Thus i n the current context there seems to mc no advantage i n proceeding o n other than a first-order basis.

(in the ascri

's and not the behever's language), b u t their independent ^^^^^ invoked. The theory is a sentential theory,

semantic P'^J^^tic.relation but not a sc 2.2

THE

theory,

PROPOSITIONAL

THEORY

bad argument for propositions that, because beliefs are not sically linguistic states, belief-ascriptions must assert relations to '"^"^"linguistic items. T o describe a non-linguistic state o f y's via its relation to something linguistic o f ours is perfectly possible. Similarly, that we can ascribe beliefs to speakers o f other languages w i t h o u t mentioning their sentences does not i m p l y that w e refer to something non-hnguistic. . . , . , , More cogent arguments tor the propositional v i e w are based on negative logico-semantical reflections upon semantic-relation analyses, coupled w i t h the unsurprising assumption that the only alternative is the propositional theory. T h e centerpiece is Church's argument against Carnap's sentential analysis (Church, 1950; Carnap, 1947). Suppose ' Columbus believed that the earth is r o u n d ' asserts a relation between Columbus and some sentence o f ours that depends upon w h a t we mean by that sentence: ' C o l u m b u s beheves something equivalent to ' t h e earth is r o u n d ' as w e use i t ' . H a d we meant b y that w h a t we n o w mean by ' s m o k i n g causes cancer', the truth-value o f ' C o l u m b u s beheved that the earth is r o u n d ' w o u l d have been affected. As Columbus did not believe that smoking causes cancer, the sentence ' Columbus believed that the earth is r o u n d ' w o u l d have been false on that analysis. B u t clearly (the argument proceeds) i t w o u l d not thereby have been false: whether Columbus believed that the earth is r o u n d is quite independent o f w h a t w e English speakers mean b y ' the earth IS r o u n d ' . The belief-sentence has different possible w o r l d t r u t h conditions than the semantic-relation analysis gives i t . N o w the argument depends o n l y u p o n the analysis i n v o k i n g contingent semantic properties. So Davidson's theory (1969) that, on an utterance o f that behef-sentence, one refers to one's o w n utterance u o f ' t h e earth is r o u n d ' , and asserts a relation between Columbus and u that depends in part o n w h a t u means (evidently a contingent property o f «), falls w i t h i n the scope o f the argument.

Classically there are t w o types o f relational theory o f behefascriptions: they assert relations to propositions, or to sentences or utterances. I n the latter case, the relation depends on w h a t the sentence or utterance means, i n the m o u t h or language o f the ascriber; so I shall call such theories semantic-relation theories. N o w I am going to argue that, given the constraint o f physicahsm, neither sort o f theory is attractive. That may be surprising, for what further possibihty is there ? The answer is that o n m y account that-clauses name linguistic entities

The interesting upshot o f the argument seems to be that i t is the meaning o f ' the earth is r o u n d ' , and n o t the sentence, that should be carried t h r o u g h other possible worlds i n determining the t r u t h

28

29

conditions o f ' C o l u m b u s believes that the earth is r o u n d ' . But ^ " m e a n i n g " qua entity is a proposition. Another argument is this. I f s is literally ambiguous, so is believes that s'*. B u t o n the semantic-relation theory exactly one relation is involved, and one sentence s; so there is no r o o m for ambiguity. The propositional theory handles i t nicely: i f 5 is ambiguous i n sense, then •^that is ambiguous i n reference. T h e f o r m o f b o t h arguments is: the propositional theory accounts for the semantic facts about belief-sentences; semantic-relation theories do n o t ; there are no further theories that d o ; so belief-sentences assert relations to propositions. N o w , i t is not true that they are the only t w o relational theories available. A theory that takes belief-sentences to assert relations to linguistic entities, but not by v i r t u e o f contingent semantic facts about t h e m , is straightforwardly i m m u n e to Church's argument and the a m b i g u i t y argument (as I show at the end o f 7.2.) Moreover, there are rephes to the C h u r c h argument at the logico-semantical level. ^ But, most i m p o r t a n t l y , those arguments are not motivated at a certain foundational level. For i f one has one's eye o n problems about explicating semantic concepts, about the compatibihty o f intentionahty and physicalism, or about the vindication o f t r u t h as correspondence, a certain amount o f regimentation and reconstruction m i g h t w e l l seem justified i f w e could then get an otherwise satisfactory theory that preserved the basic structure o f belief-desire psychology. A n argument for the propositional analysis that w o u l d be more to the point w o u l d be that propositions are needed to capture intentional explanations o f behavior. Suppose there were a persuasive argument o f this k i n d : h o w should a physicahst react ? I t w o u l d remain an open question whether intentionahty is compatible w i t h physicalism (as Brentano denied i t was). So a physicahst could not accept the propositional theory o n the basis simply o f that argument, for i t w o u l d remain to be shown that such relations to propositions can be accounted for w i t h i n a physicahst framework. M y reason for rejecting propositions, put baldly, is this: to explain the contributions o f propositions to the explanation o f behavior is to ' Invoke a trans-world equivalence-relation: let indexically denote the actual world. T h e n , l e t ' x believes that s' have the form ' B ( x , s, '. Its truth conditions, in a possible w o r l d w', depend on a trans-world relation between C o l u m b u s , as he is in w', and s, insofar as s means what it means in w ^ - i . e . in the real w o r l d and not in w'. T h e explication is committed to the real w o r l d , and not to other possible worlds, w h i c h are invoked only at the meta-level, in considering whether the explication satisfies C h u r c h ' s intensional standards o f explication, as it seems to.

30

to eliminate them. The point does not involve a shoW .^jjist—reductionist formula such as: ' t o justify i n v o k i n g A's verifieat|° j^yjical objects, mental states) is thereby to show h o w to (electrons,^^^^, _ ^ot: 3 very happy principle. T h e difference is this, eliminate^ .^^ arguing for, the adequacy o f a physicahst ' w o r k . (By 'bat I do not mean one that b y definition excludes ^'^^'"ositions; some physicahsts believe i n other possible worlds.) A '"^"^patibihst about physicahsm and the existence o f behefs w o u l d then deem it necessary to show w h a t physical facts about x w o u l d constitute f having the belief-relation to a proposition. N o w I suggest that that would be to show either that one could separately, independently o f propositions, specify the relevant physical features; or that a person's physical properties are somehow structurally isomorphic to some systematic framework that can be specified by m e n t i o n i n g propositions, their logical interconnections, etc. B u t i n the latter case, i t is difficult to avoid conjecturing a framework o f more mundane entities than propositions and relations on t h e m , suitably structurally isomorphic to the propositional framework. The second strategy is the one I shall pursue ( c f 7.2). Since I can t h i n k o f no way, apart f r o m those t w o , in which the relevant relations to propositions can be physically grounded, i t seems to me that a physicahst, even one w h o is independently prepared to accept possible worlds, ought n o t to regard beliefs as, irreducibly, relations to propositions.

2.3

SEMANTIC

RELATION

THEORIES

The classical alternative has been to regard the that-clause as referring to a sentence or utterance o f the ascriber that is equivalent i n meaning or content to the behef O n some theories, Pierre's belief involves Pierre's accepting a sentence o f his o w n , w h i c h we characterize o n l y indirectly by producing something o f ours as equivalent i n meaning to i t : a p r o positional attitude-ascription indirectly describes a sentential attitude. But there is a more general idea, for one can envisage a theory o n w h i c h beliefs are not necessarily linguistic, not sentential attitudes, but can still, m some w a y to be clarified, be described as equivalent i n content to something linguistic. O n either theory, regardless o f whether the that-clause refers to a sentence, a sentential complex (an open sentence paired w i t h a sequence o f referents) or an utterance, the relation asserted by 'Pierre believes that s' obtains in part b y virtue o f w h a t 5 means. Hence such theories may be called semantic-relation theories. 31

Linguistic items have meaning o n l y relative to a language, a speaker or a population o f speakers. This is evident for sentence-types, whiclj could mean different things i n different languages or populations. Equally obviously, an actual utterance's meaning depends o n the utterer's intentions, the population's conventions, or the rules o f the language. So, a semantic-relation theory must rely o n some n o t i o n of meaning-relative-to-a-language, or person or population. T h a t is, the relation asserted b y ' x believes that s' is such that one o f its terms is a sentence or utterance, and another, i m p l i c i t t e r m is a language, person or population, and its obtaining depends i n part o n the meaning o f the one i n the other. N o w I want t o eliminate immediately one o f those possibilities - namely, that languages are usefully invoked at this point. I n chapter 9 there is an account o f w h a t a language is; but it presupposes the theory o f attitude-ascriptions developed i n this book. Otherwise there are, I t h i n k , o n l y t w o notions o f a language that m i g h t be relevant. O n b o t h a language is something i n v o l v i n g a set o f sentences as interpreted i n a certain w a y . O n the first, to interpret sentences is t o assign propositional meanings, and a language is a function f r o m sentences to meanings (Lewis, 1976). B u t that yields no real alternative to the propositional theory and w e may dispense w i t h it f o r t h w i t h . O n the second, to interpret is t o assign t r u t h conditions i n the form '5 is true i f f . . . ' . A language is then the set o f sentences, ^ , as interpreted thus - that is, as interpreted by a t r u t h theory ^ w h i c h imphes t r u t h conditions for each sentence i n So a language could be identified w i t h the pair < ^ , ^y. W h a t sort o f entity is ^ ? I t is a theory, a certain set o f sentences. B u t reference to a set o f sentences 3r, i f i t interprets 5^, must be to ST under a certain interpretation o f ^ itself Suppose ^ implies '5 is true i f f electrons have negative charge'; the meaning that equivalence assigns t o s w o u l d depend on what the right-hand side means. N o w we are here referring t o or mentioning 3' and not using i t - that is, referring to a set o f entities whose intrinsic properties are syntactic and not semantic. Initially, i n interpreting ' x believes that s\e reference to the syntactic entity 5 is not taken t o determine h o w 5 is to be taken; that requires further, i m p l i c i t , reference t o a language or population or person. B u t then the mere reference t o the syntactic entity ^ is not enough to determine its meaning and hence not the meaning o f 5. W e must invoke something further that interprets ^ . Unless w e arbitrarily n o w b r i n g i n reference to a person or population, w e are launched on a series o f i m p h c i t 32

an infinite hierarchy o f t r u t h theories — and all o n our references^^^^^^^^ o f x beheves that s'! O f course, those references must origin* 'j^^-.^^jy ineffectual, since they are always t o purely syntactic be cuniu ^^^^ evident that this is not an argument that w e can entities^^'i^^.j^^ ^ ^ ^ ^ j ^ conditions w i t h o u t launching a self-defeating "'^^^'^ _ o f course we can, but by using 3^, not by m e n t i o n i n g it.) '^'^^If^reference to IT d i d not also require reference to something further J interprets i t , then neither should the original reference to 5: i f an ^ ^ nretation for ^ could simply be taken for granted, then so for ' But 'X beheves that s' does not then refer to s as meaning w h a t i t does relative to a language, w h i c h is contra hypothesem. I f we are to persist i n the apparently inevitable i n t u i t i o n that meaning is always meaning-relative-to-something, the semantic-relation theory must presuppose some pragmatic n o t i o n , o f meaning-relative-to-aperson or population. O n e such theory is that i n uttering 'x believes that s' the speaker refers to that very utterance o f 5 (Davidson, 1969), either w i t h an i m p l i c i t direct reference to himself ('insofar as that utterance o f s means what i t means i n m y m o u t h ' ) or a quantification over speakers ('insofar as that utterance o f s means w h a t i t means i n the m o u t h o f whoever uttered i t ' ) . Another theory is that 'x believes that s' refers to a sentence (or sentential complex), w i t h an i m p h c i t direct reference to the speaker ('insofar as s means w h a t i t means for me') or to a specific population o f speakers. N o w the p o i n t is not that the pragmatic concept o f meaning for a person or population cannot be explicated w i t h i n a physicahst framework; o n the contrary. Rather i t is that the o n l y promising explication requires an independent explication o f propositional attitudeascriptions. W h a t m y sentences and utterances mean involves relations to m y beliefs and, perhaps, communicative intentions; short o f that, the prospects for a reductionist account o f the relevant pragmatic concepts are bleak. T h a t w i l l be no surprise to many, b u t let us get the reasons out i n the open.

I

Semantic-relation theories take the t r u t h o f 'x believes that 5' to depend on our sentence or m y utterance o f 5's equivalence i n meaning or content to something o f x's, some linguistic or non-linguistic state ^ '• s* has for x the same content or meaning as 5 has for us — M{s*, s, us)'. Thus the general f o r m o f semantic-relation theories is: 'x believes that 5' is to be explicated as ' f o r some s*, B*{x, s*) and M{s*,

s, us)'. T h e question n o w is whether M is to be explicated i n terms o f some more fundamental pragmatic semantic concept. Short o f 33

tWo

t h e r e is s o m e t h i n g fishy h e r e ; f o r t h i s last p r o p o s a l a d v e r t s t o n o

candidates f o r semantic concepts m o r e f u n d a m e n t a l t h a n equivalence

tic s e m a n t i c p r o p e r t y o f our sentence s. ( H o w c o u l d t h i s h a p p e n

bringing

in

prepositional

attitudes,

there

seem

i n m e a n i n g — n a m e l y , s o m e t h i n g intentional, p f o r u s ' , o r s o m e t h i n g extensional,

to

be

only

i n the f o r m ' s means that

i n t h e f o r m ' i is, f o r us, t r u e i f f . . . '

N o w , ' 5 m e a n s t h a t p f o r u s ' is n o h e l p h e r e , g i v e n t h e classical t r e a t m e n t s o f t h a t - c l a u s e s ; e i t h e r 'p'

ranges o v e r p r o p o s i t i o n s , o r w e

n e e d a s e m a n t i c - r e l a t i o n analysis o f ' s m e a n s t h a t p' itself, w h i c h leads t o a v i c i o u s regress. C o u l d t h e e q u i v a l e n c e o f m e a n i n g o f s* a n d 5 b e c a p t u r e d b y facts o f t h e f o r m ' 5 * is t r u e f o r ;c i f f . . . ' a n d ' 5 is t r u e f o r us i f f . . . ' ? C e r t a i n l y n o t i n t h a t f o r m ; f o r t h e n a n y m a t e r i a l l y e q u i v a l e n t s* a n d s' w o u l d c o u n t as e q u i v a l e n t i n m e a n i n g , w h i c h w o u l d m e a n t h a t i n ' x believes t h a t s',

we

c o u l d s u b s t i t u t e , salva

t r u t h - v a l u e as s? is e q u i v a l e n c e

veritate,

a n y 5' t h a t has t h e same

B u t t h e i n t u i t i v e idea is t h a t e q u i v a l e n c e i n m e a n i n g in truth

conditions,

and

perhaps

there

is a

more

s o p h i s t i c a t e d w a y o f c a p t u r i n g t h a t . W h y n o t say t h a t s* a n d i are e q u i v a l e n t i n m e a n i n g , r e l a t i v e t o x a n d t o us r e s p e c t i v e l y , j u s t i n case t h e c o r r e c t t r u t h theory

^ *

o f x and the correct t r u t h t h e o r y ^

of

t h e r e is a sentence s' i n b o t h ^ * (1)

^ *

T h ' a n s w e r lies i n t h e e q u i v a l e n c e o f ""on t h e c o r r e c t t r u t h t h e o r y ^

of

t h e sentence s is true"" a n d s i t s e l f ) B u t t h e t h e o r y has, i n a n y case, be t u r n e d i n t o a p r a g m a t i c t h e o r y i f i t is t o b e a d e q u a t e mean b y ^ *

has s o m e h o w

what

t o be r e l i e d u p o n , f o r t h e f o l l o w i n g

reason. Earlier, i n c o n n e c t i o n w i t h languages, i n the f o r m < 5 ' ' , seen t h a t t h e m e r e r e f e r e n c e t o a b o d y o f sentences ^ What ^

was insufficient.

means relative t o s o m e t h i n g f u r t h e r h a d t o be

S i m i l a r l y h e r e . F o r i t is c r u c i a l t h a t ^ * theory

i t was

f o r x,

and

the

invoked.

s h o u l d be the c o r r e c t t r u t h

n o t i o n o f c o r r e c t n e s s is i n p a r t

semantic:

u n i n t e r p r e t e d t h e o r i e s are n o t c o r r e c t o r i n c o r r e c t . C o n s e q u e n t l y m u s t be s o m e n o t i o n o f w h a t ^ *

there

means relative t o s o m e t h i n g further,

a n d , as i n t r o d u c i n g h i g h e r l e v e l t r u t h t h e o r i e s is n o w q u i t e p o i n t l e s s , that w o u l d a m o u n t t o t h i s : ' r e l a t i v e t o w h a t w e m e a n b y ^ * ' .

So

we have:

us are s u c h t h a t :

(2) ^

^rfis equivalent to the preceding pragmatic semantic-relation theory?

and ^

'x

such t h a t

i m p l i e s ^'s* is t r u e i f f 5'^, a n d implies

beheves t h a t s'

is t o be e x p l i c a t e d as: ' (3 s*) B*{x, s*) a n d ( f o r a n y ^ * ) i f ^ * is t h e c o r r e c t t r u t h t h e o r y f o r X, r e l a t i v e t o w h a t we m e a n b y , ^ * , t h e n ^ * i m p l i e s ^s*

is t r u e i f f s'~^

( t h e bars o v e r ' 5 * ' a n d ' 5 ' i n d i c a t e t h a t t h e i r p r o p e r subtituends

i n that

is t r u e i f f 5 ^ ' .

c o n t e x t are s t r u c t u r a l d e s c r i p t i o n s o f sentences r a t h e r t h a n t h e sentences

N o t i c e t h a t , i f y o u d r o p t h e q u a n t i f i e r ' ( 3 5 * ) ' , w h a t is l e f t has t h e f o r m :

themselves).

M.*{s*,

N o w our correct t r u t h theory ^ s'

o f o u r s e l v e s is homophonic:

t h a t satisfies t h e a b o v e c o n d i t i o n is e i t h e r 5 i t s e l f o r

any

something

X, s, us). N o w

i t is i n c r e d i b l e t h a t t h i s is t h e

the m o r e e l e m e n t a r y c o n c e p t o f p r a g m a t i c e q u i v a l e n c e i n m e a n i n g :

s t r o n g l y e q u i v a l e n t t o s. B u t t h e n a f a r m o r e d i r e c t e x p l i c a t i o n o f ^x

M(s*,

believes t h a t

semantic terms, the apparent q u a n t i f i c a t i o n o v e r ^ *

'x

follows f r o m that account o f equivalence i n m e a n i n g :

believes t h a t s'

X, s, us). I n a n y case, i f M * w e r e t a k e n as u n a n a l y z a b l e i n f u r t h e r t u r n s o u t t o be

j u s t so m u c h s y n t a x , t h e w h o l e c o n s t r u c t i o n t o be u n d e r s t o o d as a

is t o be e x p l a i n e d as:

fS'

' ( 3 i * ) B * {x, s*) a n d t h e c o r r e c t t r u t h t h e o r y ^*

fundamental

p r a g m a t i c s e m a n t i c c o n c e p t ; t h a t is, n o t i n its t u r n t o be u n p a c k e d v i a

baroque w a y o f w r i t i n g '5* means f o r x w h a t s means f o r u s ' ; for the essential i n f o r m a t i o n c o n v e y e d is t h i s : s is t h a t sentence o f o u r s w h i c h ,

o{ X i m p l i e s ^s* is t r u e i f f s^'.

r e l a t i v e t o w h a t w e m e a n b y ( ^ * a n d h e n c e b y ) s, c a p t u r e s w h a t x's

I n o t h e r w o r d s , t h e f u n c t i o n o f a b e h e f - a s c r i p t i o n o n x is t o g i v e t h e t r u t h c o n d i t i o n s t h a t t h e c o r r e c t t r u t h t h e o r y o f x assigns t o w h a t x holds true.

sentence 5* m e a n s . T h e p o i n t o f t h e r a t h e r c o m p l e x a r g u m e n t I h a v e g i v e n is t h a t , i n the p r e s e n t c o n t e x t o f e x p l i c a t i n g b e l i e f - a s c r i p t i o n s ,

^ F o r p r e c i s e l y t h i s r e a s o n , w e c a n n o t e x p l i c a t e ''x b e l i e v e s t h a t s"" a s : ''{3s*)

B*{x,

s*) a n d

the a t t e m p t

a v o i d t a k i n g p r a g m a t i c e q u i v a l e n c e i n m e a n i n g as t h e

to

fundamental

s* i s , f o r X, t r u e i f f i " ! . A s s t a t e d , t h e c o r r e c t n e s s o f t h e c h o i c e o f s r e q u i r e s o n l y t h a t it

s e m a n t i c c o n c e p t b y i n v o k i n g t r u t h t h e o r i e s fails. So t h e final q u e s t i o n

b e m a t e r i a l l y e q u i v a l e n t t o s * . M o r e o v e r ' j ' d o e s n o t o c c u r i n t h a t f o r m u l a as a f i r s t - o r d e r

about

v a r i a b l e , s i n c e it o c c u p i e s a s e n t e n t i a l p o s i t i o n ; b u t t h e n t h a t f o r m u l a c a n ' t e x p l i c a t e ' x b e l i e v e s t h a t s',

i n w h i c h ' s ' is a f i r s t - o r d e r v a r i a b l e . B o t h p r o b l e m s s u g g e s t t h e n e x t

semantic-relation

theories

is w h e t h e r

t h e o r y discussed i n the text.

34

t h e r e is a

reductionist

e x p l i c a t i o n o f p r a g m a t i c e q u i v a l e n c e i n m e a n i n g , t h a t is, a p a r t f r o m

35

the exphcation that becomes available on the basis o f an independent account o f propositional attitudes. Equivalence in meaning Pragmatic equivalence i n meaning m a y be given the canonical form " " i * means for x the same as s means for y^, where s* is either linguistic or non-linguistic (an internal state perhaps), s a sentence or utterance, and y a person or population. Can this concept be accounted for w i t h o u t relying o n any further concept o f propositional content or meaning ? The first suggestion t o be considered is that the concept gets its identity f r o m its role i n the explanation o f behavior. The idea is this. T o explain Pierre's behavior w e must k n o w his reasons. N o w , suppose w e k n e w all his sentential attitudes — all the sentences he accepts as true and desires to be t r u e ; w o u l d we thereby k n o w his reasons? That could hardly be so w i t h o u t some account o f what his sentences mean. B u t , o n certain ways o f matching them w i t h our sentences Pierre begins to make sense to us. So perhaps the concept o f equivalence i n meaning gets its identity f r o m its explanatory role: it enables his sentential attitudes to count for us as his reasons. Let me indulge i n a parody. I f I am t r y i n g t o understand w h y Pierre looks groggy, stumbles about, and speaks incoherently, i t can be explanatory t o point out that his state is equivalent in degree of drunkenness to what mine w o u l d be after five double scotches. This shows that we can take the concept 'equivalent i n degree o f drunkenness' t o be foundational i n the theory o f intoxicated behavior; i t gets its identity f r o m that explanatory role. N o w , w h i l e analogies that we count as explanatory do rely o n equivalence-relations i n that way, i t is obvious, as I t h i n k the example makes clear, that their explanatory value depends o n an independent grasp o f the basic reference case — depends i n that example o n m y k n o w i n g or guessing w h a t i t is like to have d r u n k so many whiskies. M y understanding o f that equivalence-relation is simply m y grasp o f 'his state affects h i m like this, like that, etc.'. Some explanations account for the concepts thereby employed, and some presuppose an independent understanding o f t h e m . The former happens w h e n the explanatory context introduces the concept as ' the state, relation, etc. that has such and such causal etc. r o l e ' , where that role is somehow specified. B u t i n the whiskey example the equivalencerelation is not supposed to have any causal r o l e ; i t is appealed to extrinsically t o characterize something, an internal state o f our subject, that has the explanatory causal role. Similarly for i n v o k i n g equivalence 36

in explaining another's behavior. Suppose certain linguistic jn n i ^ * " * " ^ j j j o n g the causes o f a person's behavior. Instead o f characstates are a j-j.gg(.iy as having certain causal roles, w e m i g h t characterizing t ^ equivalent i n meaning t o certain states o f ours; and that terize t e^^^ ^ separate understanding o f equivalence i n meaning. TWs^leads naturally to another suggestion - namely, that pragmatic lence i n meaning is a certain equivalence i n explanatory or equiva ^^^^ Perhaps the idea can be made m o r e v i v i d as a proposal explicating attitude sentences - something l i k e : °^

'Pierre believes that snow is w h i t e '

is to be explicated as:

^ ^ r



,

,

r

' For some s, Pierre accepts 5, and the functional role of accepting 5 for Pierre is the same as the functional role o f accepting ' snow is w h i t e ' for us' But functional role is always relative to a particular system o f connections among i n p u t , o u t p u t , and internal states; there is no concept o f equivalence i n functional role tout court. I t is always relative to a specified functional system; t w o states, e.g. t w o physical states, may be functionally equivalent i n one functional system but n o t i n another. So the above c o n d i t i o n should read: ' Pierre's h o l d i n g - t r u e 5 is functionally equivalent to m y holding-true ' s n o w is w h i t e ' i n functional system F.' B u t a moment's reflection shows that to specify F essentially is t o give the w h e r e w i t h a l for specifying all the particular functional roles w h i c h F generates. B u t that implies being able t o specify the functional role o f Pierre's h o l d i n g - t r u e s independently o f its functional equivalence t o m y holding-true ' snow is w h i t e ' . A n d that functional role w o u l d seem t o be what elucidates ' Pierre believes that snow is w h i t e ' ; the content o f Pierre's belief is the functional role o f what he holds true. T h e details o f my psychology drop out o f the picture. This may initially be somewhat puzzling, for w h a t m i g h t a direct specification o f such a functional role l o o k like ? M i g h t i t be something like 'Pierre believes that snow is w h i t e ' ? W e shall see that indeed i t tnight. The analysis I shall propose takes' believes' to express a relation between Pierre and m y sentences; b u t the relation does not depend upon the functional role or meaning w h i c h those sentences have for me. I conclude that no semantic-relation theory o f behef-sentences can be maintained, if, that is, we are to avoid irreducible pragmatic semantic concepts. O n each w a y o f explaining the theory we come t o a dead etid. I f the basic concept is 'means that p\e theory is circular. I f i t 37

is ' is true i f f p', the conditions are either not sufficient or depend, if, effect, on equivalence i n meaning. I f it is equivalence i n meaning, then given that i t can't be accounted for b y its explanatory role, the only way to explicate i t w o u l d be i n terms o f equivalence i n functional role B u t that indirectly requires an independent specification o f the functional role o f Pierre's holdings-true, so that their equivalence to our sentences is beside the point. Semantic-relation theories are half-baked. T h e y rely at b o t t o m on taking interpersonal equivalence in meaning to be more basic than the specification o f particular meanings. B u t while that has a nominalist appeal, i t cannot be vindicated. I f equivalence i n meaning or content can be explained at all, that must be i n terms o f an independent account o f the content o f beliefs, and o f derivative concepts o f meaning.

2.4

BELIEFS

AND

DESIRES W I T H O U T

THAT-CLAUSES

Propositional and semantic-relation theories exhaust the hitherto available options for explaining behef- and dcsnc-ascriptions — i n other words, for explaining that-clauses. B u t there are other theories o f what beliefs and desires are. Suppose they are essentially hnguistic, that to believe or desire something is to stand i n a certain relation to a sentence o f one's o w n language,' h o l d i n g - t r u e ' or ' w a n t i n g - t r u e ' (' z holds-true s' does not mean 'z beheves that s is t r u e ' ; to hold-true 5 is not to have a metalinguistic belief about s). A c o m m o n v i e w nowadays explains those relations i n terms o f their systematic roles i n explaining behavior. A n d , some recent theories take sentential attitudes as p r i m a r y i n rational psychology, de-emphasizing propositional attitude-ascriptions on the grounds that non-extensional interpersonal s y n o n y m y can't objectively be accounted for. T r u t h conditions and references are then assigned directly to sentences as the objects o f sentential attitudes. There are t w o such theories, interestingly different; and w h i l e their details do not matter to the main p o i n t , it w i l l help to indicate their general f o r m . H a r t r y Field (1978) has proposed that w e regard beliefs and desires simply as internal representational states whose structure is that o f a language. W h i l e Field allows the internal "sentences" o f t w o persons to be alike i n reference and extensional t r u t h conditions ( w i t h predicates denoting properties and not sets), he is sceptical o f any stronger interpersonal s y n o n y m y , and therefore o f a determinate reading o f propositional attitude-ascriptions. D o n a l d Davidson

(1974) proposes a different reconstruction o f 38

chology: sentential attitudes are basic, not as internal states, rational psyc believer's public language sentences w h i c h w e but as rc iristrumentalist spirit, i n order to systematize behavior, ascribe- ^^^^^^^^ j^e determined b y a Tarski-type t r u t h theory on the "^'j^'^'er^s" language, and n o t h i n g like propositions come into the

^''rhe

point I wish to make is that eliminating that-clauses i n favor sentential attitudes and their extensional semantics carries away ° rtain apparently basic features o f rational psychology. The proposi'^kjnal attitude system contains resources for describing patterns o f motivation, learning, reasoning, etc., at a non-language-specific level. That is, we have general conceptions o f inductive and deductive tendencies, the role o f the practical syllogism i n m o t i v a t i o n , as w e l l as processes introduced b y more specific psychological theories having to do, e.g., w i t h cognitive dissonance or unconscious m o t i v a t i o n and so on, whose description involves no mention o f specific features o f the subject's language. Evidently history and the social sciences depend essentially on such explanations and descriptions. Conceived i n general terms, such patterns are at a level o f abstraction not to be found w i t h i n a framework o f sentential attitudes and extensional semantics; and, as I shall t r y to make clear, to supplement that framework suitably is to m o v e towards a basis for propositional attitude-ascription. T h e point is not that this is eo ipso a defect i n the sentential attitude f r a m e w o r k ; for i t could t u r n o u t to be the most that can be vindicated. B u t i t needs to be registered that a framework for psychological theorizing w h i c h lacks the resource o f propositional attitudes is bound to be strikingly different f r o m the one we operate w i t h . Substituting sentential attitudes w o u l d be quite radically e l i m i n ative, even w h e n the extensional semantics is taken i n t o account. Taking other creatures as subjects o f ordinary rational psychology turns on whether w e can describe such things as their m o t i v a t i o n , i n the pattern 'wants q, believes p, beheves that i f f then d o i n g A results ^ N o r do semantic-relations o f the sort I have discussed. H o w e v e r , Davidson has in another paper (1969) proposed an account of propositional attitude-ascriptions that goes roughly like this: on an utterance o f ' x believes that s' I assert that some state o f x's stands in a certain semantic equivalence-relation to m y utterance of s. N o w i f that state of x's is a sentential attitude, and i f the semantic equivalence-relation is equivalence in what Tarski-type truth theories ascribe to m y utterance and to xs state, the theories of the 1972 paper and of the 1969 paper fit together. In elfect I have discussed that sort of account 'n the preceding section. B u t it is possible also to read the 1972 paper as a proposal of a sentential attitude theory in w h i c h propositional attitude-ascriptions have no essential role.

39

i n q, does A'; or their generahzing f r o m experience; o r their reasoning via modus ponens or certain simple patterns o f non-contradiction; and so o n . B u t , evidently such description abstracts f r o m the lexical and syntactic features o f their sentences, that is surface features described in a formal syntax, 'e.g. as i n 's consists o f s^ follov^red b y ' a n d ' followed b y 5 2 ' . W e then have these questions, ( i ) Given that generalizations o f those types abstract f r o m syntactic details, are there still resources w i t h i n a sentential-attitude framework for capturing them? ( 2 ) W h a t about m i m i c k i n g them language b y language in generalizations that i n effect are about the practical syllogism, modus ponens, etc., b u t are couched i n the syntax n o w o f Hans's language, n o w Pierre's language, n o w Jose's language, etc? W o u l d that be a useful substitute for the more general framework? As regards ( i ) , perhaps w e can generalize i n terms oitruth conditions. O n the face o f i t , that w o u l d involve quantification and semantic description o f a sort that goes beyond the resources o f an ordinary truth theory. Thus, suppose w e w a n t t o say that Pierre has just reasoned by modus ponens: he holds-true a sentence S2 because he holds-true s^ and a sentence s^ w h i c h is a conditional o f 53 o n s^. T h e force o f 'is a c o n d i t i o n a l ' , o n the line being pursued, is semantic. B u t no such semantic predicate occurs i n an ordinary Tarskian t r u t h theory, and naturally i t is n o t unproblematic here. I f w e t r y t o define i t i n terms o f S j ' s being described i n a certain w a y b y the correct t r u t h theory T* o f Pierre, w e are back t o the semantic-relation theories o f 2 . 3 : the t r u t h theory T * is itself being referred t o as meaning something relative t o something further. I n any case, i f w e could unproblematically cash out the full array o f semantic predicates o f the f o r m 'is a c o n d i t i o n a l ' , etc. (including ' is a caf-word' and so o n for the non-logical words o f Pierre's language), w e w o u l d thereby have an effective account o f thatclauses.

, ' d' etc. w o u l d then somehow be captured, and provided >^^'/and '^^^^ happened across the board, w e w o u l d again have a the same^t explicating that-clauses. framewo^^^^^ positions. First, the sentential-attitude f r a m e w o r k ^•^'^ eTthose generahzations and forms o f description. T h e n w e have eliminat^^ something very basic indeed, and w e are left w i t h a dras^•^"^1" mincated rational psychology, i f indeed that term still applies, t " - * ^ j ^ j j j l y there is another more attractive possibility — a k i n d o f ^ mise. Perhaps some interlinguistically ascribable functional '^"les^can be captured b y such logico-syntactic descriptions o f sentential "^ttitudes as' is a c o n d i t i o n a l ' , w h i l e we cannot spell out interlinguistically or interpersonally ascribable functional roles for non-logical or n o n structural components o f sentences, hke ' c a t ' (for whatever reasons). Then we could capture those generalizations b u t could n o t fully exphcate that-clauses. This possibility has to be borne i n m i n d ; i t w o u l d constitute i n effect a partial explication o f propositional attitudeascriptions. The generalized functional account suggests a new w a y o f accounting for that-clauses: somehow they capture functional roles o f sentential attitudes that are interpersonally ascribable, and subsumable under those generalizations that are virtually constitutive o f o u r commonsense rational psychology.

2.5

PHILOSOPHICAL

EXPLICATION

The functional theory o f propositional attitudes o f chapter 4 depends on a certain attitude towards philosophical explication o n w h i c h i t is both flexible and conservative. The f o l l o w i n g comments on the general subject o f exphcation m a y clarify the constraints o n the project.

( 2 ) W h a t then about i m i t a t i n g those generalizations w i t h i n each language? That w o u l d be fine, b u t the functional similarities between

A n attempt at analytical explication m a y arouse t w o different negative reactions o f principle. There is anti-reductionism; a thing is what it is and not another t h i n g . A n d there is e l i m i n a t i v i s m : i f a concept is philosophically troublesome, get r i d o f i t and, i f necessary, replace it w i t h something mo r e lucid, mo r e rigorous, mo r e scientific, mo r e ontologically parsimonious. There is something simplistic i n b o t h extremes. T o make a concept sacrosanct, n o t illuminable b y reconstruction, m a y simply make i t uninteresting i n the light o f ongoing theory; and, i n any case, the drive towards explication m a y be irrepressible. O n the other hand, philosophical exphcation does n o t bave to be seen as M o o r e a n analysis. There is an apparently ineliminable

40

41

Something similar results f r o m (what appears t o be) the o n l y other possible w a y o f extending the sentential-attitude f r a m e w o r k t o capture those generalizations. Suppose one were t o say that whether Pierre's sentence s^ is conditional depends u p o n its role in reasoning, i.e. that i t behaves functionally thus and so. N o w i f we could treat all details o f Pierre's language i n such terms, w e could thereby construe that clauses — namely, as devices for ascribing functional roles t o w h a t the believer holds-true.

role for explication i n philosophy and science; so i t has to be at some level a naturahstically describable cognitive phenomenon. Let n\c explain w h y i t is inevitable, and w h y i t does not depend on any non-natural analyticity. ' E x p l i c a t i o n ' covers a broad spectrum o f philosophical constructions and n o t j u s t those that m i m i c Moorean analyses; they range from Quine's paradigm o f Wiener's and Kuratowski's definitions o f ' o r d e r e d pair', to philosophers' explications o f 'x knows that p'. N o t all exphcations have the same m o t i v a t i o n , or the same ideal o f w h a t is to be preserved; indeed not all explications serve specifically philosophical purposes. I t is the latter I shall be concerned w i t h here. Explication is elimination, according to Quine. A n d that must be the beginning o f a naturalistic account, given the i m p r o b a b i l i t y o f vindicating analysis as literally the unpacking o f what already is there i n meaning. T h e slogan could suggest what i t does not entail - that all exphcations are conceptual overthrows. B u t explication is elimination w i t h replacement; and the conservativeness o f some replacements is the point t o be made. Q u i n e ( i 9 6 0 , pp. 257-62) emphasized puzzle-solving motivations for the replacement o f troublesome concepts b y others that preserve theoretically useful features. W h i l e puzzle-solving and theoretical u t i h t y are certainly i m p o r t a n t i n m a n y central cases, they are not what I wish t o emphasize. Consider w h a t m i g h t be called an explicative dilemma, whose ingredients are as follows, ( i ) A theoretical framework A - for example, physical theory, extensional semantics, set theory, t r u t h functional constructions, quantificational constructions, or mechanistic biology. (2) A n imperialist inclination t o accept A as adequate for expressing all truths about a certain subject S. (3) A set o f propositions B, w i t h i n S, w h i c h are not i n A, but w h i c h there is a strong inclination to accept; for example, propositions about beliefs and desires, or meaning, or numbers, or w h i c h are conditional, or m o d a l , or about human autonomy.

• jjs" o v e r t h r o w i n g the imperialist pretensions o f ^ , and g i v i n g jesolu ^Q{xit is that there is more than one w a y to give up B. h first is ruthless r e v o l u t i o n : pursue theories i n A, and jettison B. ruthless r e v o l u t i o n is not always sensible for t w o reasons. First, is the one that Quine has emphasized, the theoretical u t i l i t y o f '''^Secondly, there is the matter o f entrenched cognitive attachment to ^ it is awfully difficult not t o see people as having beliefs and desires c making real choices, etc., quite apart f r o m w h a t those concepts ontribute to o n g o i n g theory. W e simply accept certain things, and overthrowing t h e m , even i n the light o f the imperialist theory, may be enormously difficult. There is then often ample incentive to seek a conciliation, and that comes i n degrees. Imagine t w o theorists early i n the century w h o react to Russell's analysis o f ' n u m b e r ' as " m a x i m a l set o f conumerous sets", i n these two ways. B o t h accept the replacement, but one feels his o n t o l o g y has changed, w h i l e the second does not. O r imagine t w o philosophers w h o react to compatibilism about free choice (in terms o f counterfactuals about wants and decisions) i n the f o l l o w i n g ways. B o t h accept the replacement, but one takes his v i e w o f the h u m a n c o n d i t i o n t o have changed, and the other does n o t . M y p o i n t is that the disagreement between these theorists need not be regarded as having a truth-value. For the latter member o f each pair the replacement counts as a conservative explication, w h i l e for the former i t is more or less radical. (There is still something factual i n w h a t they agree o n . For the replacement to be correct, truths o f B must give w a y t o truths o f ^ . )

Thus described, an explicative d i l e m m a does not entail explicit inconsistency, since i t involves strong inchnations and not o u t r i g h t acceptances; but perhaps some theorists' cognitive dilemmas involve o u t r i g h t inconsistency. Either w a y , an exphcative set-up is unstable i f faced up t o . Given a classical n o t i o n o f analysis, the inconsistency could be diagnosed as merely apparent: the propositions o f B are deemed equivalent i n meaning to truths o f A. Otherwise there are t w o possible

The theorist for w h o m the explication is conservative may have achieved the happier resolution o f the dilemma - not exactly because he hasn't had to change any beliefs, for that rests o n the dubious concept o f analysis. B u t his cognitive situation is as though he hasn't changed any beliefs. H o w can that be? I can suggest no more detailed account o f the phenomenon o f conservative explication; but i t occurs and can serve as a cognitive resolution t o w h a t is perceived as a serious theoretical p r o b l e m . A n exphcative dilemma is like a paradox; and w h y resolve paradoxes? There are t w o reasons: first, the cognitive discomfort may demand relief and, secondly, paradoxes can obscure the t r u t h . So w i t h exphcative dilemmas. This suggests that one p o i n t o f philosophical theorizing is cognitive therapy. B u t this is n o t the Wittgensteinian therapy o f insights w i t h o u t theories; i t suggests instead an i m p o r t a n t incentive to theory — even w h e n conservative explication is not fully on the cards: some cognitive concihation may be better than none.

42

43

3 Functional theories 3.1

FUNCTIONAL

THEORIES:

A

GENERAL

CHARACTERIZATION

Functional theories are theories o f individuals, o f h o w the states o f an organism, or a machine, are causally or sequentially related. In cases that are n o r m a l l y o f interest, those states are internal. W h i l e a functional theory says n o t h i n g o f their composition or internal structure, i t docs far more than to summarize certain external i n p u t - o u t p u t correlations. For a functional theory o f a is falsified i f as internal states are not causally or transitionally related as i t requires. Such a theory does then c o m m i t itself to substantive claims about what is inside. Those "states" are state-types and n o t state-tokens. So a functional theory o f a does not describe the cause or effect o f any particular dateable state or action. Rather i t asserts that a's possible states are counterfactually related to each other, i n p u t and o u t p u t thus and so. 'Possible states' means (to restrict attention to physical things): state-types w h i c h a could be i n , given a's physical structure and composition. That a functional theory asserts counterfactual relations among state-types is clear f r o m its m a k i n g sense to describe functionally a machine that has never been turned o n . So explanations o f a's behavior i n terms o f a's being i n certain functional states are not part o f a " f u n c t i o n a l t h e o r y " o f a. T h e y presuppose a functional theory o f a, and are related to i t somewhat as explaining a trail i n a cloud chamber i n terms o f a positron's passage is related to a general theory o f particles. A functional theory does not then say what is going on inside, not even in functional terms. This is related to the earlier p o i n t that t o k e n - t o k e n psychophysical identities presuppose t y p e - t y p e correlations. I n w h a t follows, the "belief-desire t h e o r y " is true o f a at f i f a's possible state-types are organized functionally i n a certain w a y at t. Whether a has a specific belief or desire is a further question, about whether a actually is i n a state w i t h a certain functional role.

or a quite different matter, i t can have the same functional prganizaW" ^ ^,^ ^^^^ v i r t u e o f different underlying ojganization^^^^^ although t y p e - t y p e correlations are required b y a state-tyP^^^-^^^^^^j. j^gjjtal predicates, the requirement is m i n i m a l , functiona^^^^ all that is required for a to be i n M at f Fo"^ ^ t !t certain possible state-type has a certain counterfactual r 'Atonal role, and a is i n a state o f that type. This allows for m a x i m a l plasticity i n as w i r i n g : a certain mental • n can be switched f r o m one neurophysiological state to another. NJo'neurophysiological state-type is reserved for a certain mental role, and h o w i t comes to have a particular role may be radically contingent, for h o w i t would interact w i t h others m a y change f r o m m o m e n t to moment. A l l that is required for a to continue f r o m t^ to (3 to have a certain functional organization, and therefore to be capable o f a given functional state, is that at each t i m e between t^ and ^2 some states o f a's are thus organized, and these need not be the same states throughout. The states over w h i c h a functional theory quantifies are not themselves functional states. Suppose T is a functional theory o f a physical thing a, some machine say. So the states w h i c h w o u l d make T true o f a, by being suitably causally related,^ w o u l d be physical state(-type)s, w h i c h w e can t h i n k o f as properties o f a. T h e y are first-order properties, properties o f individuals, whose definitions do not quantify over other properties o f individuals. N o w i f a first-order property or state 9 is among those w h i c h makes T t r u e o f a, that is because 9 satisfies a certain counterfactual causal condition - cp has a certain functional role. This " r o l e " is a second-order property, a property o f a first-order property. N o w , because a is i n a first-order state w i t h a certain functional role, a itself may be said to have a second-order property in another sense o f that t e r m , on w h i c h i t means ' property o f having some first-order property that satisfies a given c o n d i t i o n ' . A functional state o f an individual isjust such a second-order state or property - being in a first-order state wilih a certain functional role. Functional states, then, are not the states a functional theory quantifies over, for the latter are first-order, and functional states are second-order. ' State-types stand in such counterfactual causal relations if their instantiations w o u l d stand in certain causal relations to each other.

Evidently the concept o f functional organization ought to be relativized to times. A n i n d i v i d u a l can cease to have a certain functional 44

45

3.2

DAVID

LEWIS'S

THEORY

OF THEORY

INTERPRETATION

W h a t is i t t o interpret a theory as a functional theory? W h a t logical f o r m do we thereby ascribe to it? So far I have indicated o n l y in j general w a y what a functional theory o f a says, that as states have certain counterfactual roles. N o w the question is how a functional theory expresses that. The most lucid account o f the structure o f some functional theories, o f h o w a theory's terms denote states via their functional roles, is a certain application o f D a v i d Lewis's general account o f theory interpretation (1970). B u t the m o d e l thereby generated turns out, as we shall see, not t o be completely general. I t captures o n l y one (albeit central) type, and the belief—desire theory is not thereby accommodated as a functional theory. B u t a reasonably efficient way o f reaching a more general conception o f a functional theory is via that m o d e l . First Lewis's account o f theoretical terms i n general needs explanation. Ramsey (1929) t o o k a theory's empirical i m p o r t to be isolated by m a r k i n g o f f observational f r o m theoretical predicates, and then existentially quantifying the latter, w i t h the (second-order) quantifiers prefixed to the w h o l e theory, to the conjunction o f its postulates. Ramsey's idea is quite i n t u i t i v e ; for i f T's theoretical terms are explained via their role i n T , then i t is natural t o regard the proposal o f the theory as being like this: 'There is a property (call i t 'being an electron'), and another property (call i t 'being a n e u t r o n ' ) , such that, etc.'. Ramsey's theory, thus stated, is realist about theoretical properties, but eliminativist about the predicates that express those properties, their i m p p r t being expressed entirely i n an observational and logical vocabulary (including the crucial n o t i o n o f a property). A Ramsey interpretation then takes a theory t o say there are properties or relations that are related t o each other and to types o f observables i n such and such ways. Lewis's account departs f r o m Ramsey'jBFirst, there is a useful syntactic revision o f the theory before i n t e ^ e t a t i o n . Each theoretical predicate, ' x is an electron' say, is converted into a name, prefixed by the instantiation-relation— ' X has clectronhood'. Secondly, the theoretical—observational distinction is dropped, i n favor o f one between those terms that are understood via their role in the theory, and those understood independently o f the theory. T h i r d l y , Lewis holds that a theory is not explained merely by existentially generalizing its theoretical terms. Let T [ t j , . . , f„] represent the occurrence o f all T's nominalized theoretical terms; and let 46

V ] represent the substitution o f distinct variables t h r o u g h o u t , sey's account, T says that T [ x i , . . . , x „ ] is realized by some On o f properties, relations, etc. Lewis claims rather that n o r m a l l y jequence^ ^^^^^^^^^ intends its theoretical terms to denote uniquely, * ^'^^t'hat therefore T is analytically equivalent to the assertion that ^""^ ^ * x ] is satisfied by a unique sequence o f properties, etc. I shall ^ ' * j { , j s by "saying that T is equivalent t o : P^' -yi^ere are items (properties, relations) X j , . . .,x„ w h i c h uniquely are such that T[xi,..., x„]. A n important effect o f the strong interpretation is the explicit definition o f each theoretical t e r m : ( j = the item X j for w h i c h there are X j , . . . , x„ such that they all are uniquely such that T [ x i , . . . , x „ ] . = the i t e m x„ for w h i c h there are , . . . , x „ _ j such that they all are uniquely such that T[xi,. ..,x„]. So the term tj m e a n s : ' that property, state, etc. w h i c h is the I'th member o f the unique sequence that realizes T[xi,.. .,x„]'. This interprets theoretical terms not j u s t as they occur w i t h i n the theory, b u t also, say, in singular predications. (Ramsey's theory i n this connection is awkward; c f 3.4.) Theoretical terms get their meaning f r o m their role in T ; b u t w h i c h are those? Lewis's answer is that a non-theoretical term is " o n e o f our original terms, an old t e r m w e already understood before the new theory T.. .was proposed". N o w this temporal criterion is not exactly right. For h o w relevant is i t to interpreting attitude-sentences, for example, whether historically the concepts o f belief and desire came later than the relevant concepts o f physical environment and behavior, or whether children learn the one set before the other, rather than all together? W h a t matters is that now the former concepts are relatively unproblematic, while thoje o f belief and desire need elucidation. This is h o w i t is w i t h explicaUifei^'ih general. Intuitions about the relative clarity o f t w o sets o f concepts, together w i t h metaphysical assumptions (e.g. physicalism), determine what is w o r t h explicating i n terms o f what. M o r e archaic concepts m i g h t be explicated in terms o f newer ones.-^ O f course, w h e n a theory has just been proposed, the o l d terms w i l l be the proper explicative basis. Indeed this often happens when sets and functions are involved in exphcation.

47

3-3

T H ELOGICAL FORM

OF FUNCTIONAL

THEORIES:

A

FIRj^

APPROXIMATION

Suppose the behef-desire theory is universally quantified, the quantifier being over persons w i t h postulates l i k e : ' for any z, i f z believes that p and q, then z believes that p', and so o n . N o t h i n g has been said about w h i c h expressions o f the belief—desire theory, i n w h a t syntactic form m i g h t be interpreted b y the Ramsey—Lewis m e t h o d , and w e shall see there is a p r o b l e m about that. T h e present p o i n t concerns merely the universal quantifier over the subjects o f the theory: w i l l applying Lewis's m e t h o d t o the theory i n that f o r m make i t a functional theory i n the i n t u i t i v e sense? T h e answer is n o . For i t w o u l d then have the f o r m (dropping uniqueness for s i m p h c i t y ) : 3^1 x^zT[z,Xi,...,xX This says that certain states Xi,.. .,x„ have a certain functional role for all persons. B u t that a functional theory is true o f t w o things does not require them t o have first-order states i n c o m m o n ; and so there need be no states that have the relevant functional organization i n all persons. T h e upshot is that functional theories are n o t i n the basic case universally quantified. Rather, w h a t gets interpreted are instantiations o f the theory b y particular things o r perons. I f w e wish t o speak o f the belief—desire theory, rather than the S m i t h or Jones theory, that " t h e o r y " is an open sentence; henceforth I take functional theories to have such a free variable. The Ramsey—Lewis m e t h o d is then directly applied. I f the original theory has the f o r m : T[z, t^,...,t„] then, i g n o r i n g uniqueness, the interpreted theory is: lx-^,...,x„T[z, Xi,...,x„] thus leaving i t open that different individuals satisfy the theory b y virtue o f different first-order states having the relevant functional roles.

tates B u t this has the aesthetic defect o f not being a direct fifst-or'^^'' ^j^g general m e t h o d . M o r e i m p o r t a n t l y , there is an gpplicatio'' j.g3Son t o take some functional theories n o t t o be iniieP^" . ygntified. Suppose a theory contains an i n p u t condition universe J^^^^^^^ contains no theoretical terms. Some things may then whose antecedent b u t n o t the functional theory. B u t o n the satisty j_^^jj(gntial proposal they falsify the theory. For example, i f universa psychological theory contains an antecedent that lacks an ''^^^j^gj^al terms, then an otherwise mindless m a n n i k i n that satisfies ^'^falsifies the theory. T h e general problem is n o t remedied b y "^'ecifying the intended subjects; for the o n l y discoverable relevant ^^^'^'erty they have i n c o m m o n m a y be the sharing o f that functional ganization. This is n o t t o deny that there are interesting universal ^pralizations o f the f o r m : ' A l l members o f species S instantiate the neralizations psychological theory T'. B u t that is not the basic f o r m o f a functional theory. As I have already pointed o u t , a functional theory is also relativized to times, that is, contains a free variable over times; i t is then satisfied by individual time-pairs. Earher w e saw that functional theories are counterfactual i n f o r m . The reason I gave was that a machine that has never run can have a certain functional organization. A n o t h e r reason now is the relativization t o times. So functional theories are n o t about merely statistical properties o f states. B u t consider Lewis's proposal that the concept o f a psychological state is o f a state that typically has such and such causes and effects (Lewis, 1971,1972). W i t h o u t the relativizing to times, a psychological state can't be realized b y different first-order states at different times. B u t i f w e do relativize t o times, then ' mental state M (for x zt t)' becomes: ' t h e state o f x w h i c h at t typically has such and such causal r o l e ' . That ' t y p i c a l l y ' makes n o sense. T h e problem evaporates i f the theory i n w h i c h ' M ' occurs is counterfactual. For then ' M (for x at t)' means: ' t h a t state w h i c h at t would play such and such role i f . . . ' .

There is another possible w a y t o apply the Ramsey—Lewis method to functional theories. The p r o b l e m was that i f the theory is universally quantified, the result o f applying the m e t h o d is an existential—universal proposition, w h i c h is too strong. W h y not instead place the existential quantifiers inside the universal quantifier t o get the weaker universalexistential proposition ? That says that for each relevant t h i n g or person some first-order states are functionally organized thus and so, n o t requiring creatures that share the functional organization t o share

This account allows for a's being i n a certain functional state, even i f the state that reahzes i t has a cause that is not among those mentioned m its definition. Suppose (what I doubt) that the commonsense concept o f pain is i n part the concept o f that state o f x w h i c h w o u l d at t be caused b y b u r n i n g , gouging, c u t t i n g ; a's pain m i g h t still be caused b y l o n e o f these, as an instance perhaps o f causalgia, for i t m a y be a state Aat would at ( have been caused b y b u r n i n g , etc.

48

49

A theory's (a) containing free variables over individuals and times.

(b) asserting counterfactuals relating inner states, input and output, ^ (c) being interpretable b y the Lewis m e t h o d , w i t h the theoretical tcr^^ denoting states b y virtue o f their functional roles — these w o u l d see ^ to be sufficient for its being a functional theory. B u t are they necessari^j This is i m p o r t a n t because i f they were, then, as we shall see, t i , belief-desire theory could not be interestingly construed as a functional theory. There are t w o respects i n w h i c h the model is not general enougl, First, the unique realization requirement is too strong; and in the f o l l o w i n g section I shall suggest a w a y o f liberalizing that. But that isn't the most i m p o r t a n t p r o b l e m . For, secondly, those conditions contain a restriction so natural one barely notices i t . I mean that the theoretical terms are construed as denoting the states to w h i c h the theory assigns functional roles. That does define an interesting class of functional theories, what w e m i g h t call simple black box theories. But some theories that are, b y a general and i n t u i t i v e criterion, functional theories do not satisfy that condition. Even so, the Lewis method will still i n its w a y be employed to explicate t h e m ; the belief^esire theory is a m o n g t h e m .

J to the functional t e r m as follows: as regards ( i ) , i t m i g h t corre*P°"r 2 vvere i n a,- then p w o u l d result and, i f z were i n fc, then j,e that, 1 ^ ^ ^ j ^ regards (2), sometimes i( q were the case then P ^ ]A be in'«, and sometimes i f q were the case then z w o u l d be ^ ° One m i g h t be tempted to say that a, and 6, were functionally in j ^ " ^ ^ j j j therefore T was m u l t i p l y reahzed. B u t that w o u l d be equiva e^^.^j^^^ actually satisfy (2), w h i c h is not statistically ^'^"hfie'd B"'^ disjunctive state [a, V 6,] does satisfy b o t h (2) and

Before considering m u l t i p l e realization, i t w i l l be useful to consider an i m p o r t a n t w a y i n w h i c h functional theories can be uniquely satisfied — namely, b y disjunctive states. Suppose Tcontains b o t h ( i ) ' i f z were i n (, then p w o u l d result' and (2) ' i f q were true then z w o u l d be i n r,'. T w o underlying physical states a, and i , m i g h t interestingly

^'^So disjunctive states can figure i n cases that appear to involve It'nle realization because o f an ill-defined conception o f functional realization. Characteristic o f such cases is that the disjuncts o f the j„nctive states do not realize the relevant functional states w h i l e they appear to do so because being i n each o f them implies being i n a disjunctive state that does realize the functional state. N o w let us see that disjunctive states cannot be invoked to forestall a d m i t t i n g m u l t i p l e realizations. Here are t w o examples o f multiple realizations o f functional theories. ( i ) Suppose T contains this clause: ' i f z were i n f, and z were i n t- then q\d suppose t, occurs nowhere else i n the theory. ( A n example i n the behef-desire theory is the concept o{ability: a decision leads to an action, o n l y given the ability to do the action; the sole function o f abihty i n that theory is to convert decision into action.) N o w suppose several underlying states have the role o f that is, they satisfy the condition o n x, that i f z were i n x, and z were i n tj then q. W o u l d we count the functional theory as thereby falsified? It appears not. Notice that the disjunction o f those states also satisfies the c o n d i tion; but i t can't be invoked to preserve unique realization, for i t and its disjuncts all satisfy the condition ( c f L u b o w 1977). (2) Let T be a deterministic functional theory that contains ' causes tj and tj causes tk'< where that is the only occurrence o f tj i n the theory. (This last proviso is not necessary for the general point, but i t makes the case simpler.) Suppose underlying states x, and satisfy all the other conditions the theory imposes on and t^, and suppose there are states y\d y2 such that x, causes y j , y^ causes y2, and y j causes x^,. T h e n , ( i ) Xj causes y j and, b y transitivity, y^ causes x^, and (2) x, causes y j (^y transitivity) and 72 causes X),. So t w o sequences that differ i n their Jth places satisfy T. (This point is due to N e i l L u b o w , 1977.) I f the functional theory were thereby counted as false the chance o f true functional theories o f organisms w o u l d be l o w , for sequentially linked

50

51

3.4

THE

MULTIPLE

REALIZATION

OF F U N C T I O N A L

THEORIES

T h e current concept o f a functional theory appears to allow two physical states to have the same functional role, to realize the samefunctional state, to be functionally equivalent. A functional theory T can be m u l t i p l y realized b y sequences o f z's states i n such a w a y that r is to be counted as true o f z. This is o f course impossible on the recently given interpretation o f functional theories. I shall present weaker conditions w h i c h still yield one o f the principle benefits of Lewis' m e t h o d — explicit definitions o f functional terms. (This section may be merely technical to those not interested in Ramseyfication as such; the reader w i l l not lose the thread o f the main account b y passing directly to chapter 4. T h e central idea here is that a nominahzed functional t e r m denotes not a single first-order state by virtue o f its functional role, but a set o f states.)

states must underlie v i r t u a l l y all such theories that we are likelv produce. W e must then allow for m u l t i p l e reahzation. A t first, that migj^^ seem to i m p l y that a functional theory T[ti,..., f„] is equivalent to Ramsey sentence ^x^,.. .,x„T[xi,.. .,x„], w h i c h asserts that one more sequences o f states realize the theory. B u t this leads to considerably difficulty i n interpreting functional terms one b y one, as they oceur outside the theory. W h a t the difficulty is w i l l emerge, the Ramsey version being i n a certain respect too weak. B u t a certain stronger condition generates explicit definitions o f functional terms while a l l o w i n g for m u l t i p l e realization. This I shall n o w present, and then give reasons for preferring i t to the Ramsey theory. I f the predicate ' z is i n ' o f theory T expresses a second-order property o f the sort usually suggested i n this connection, then, for any first-order property i n the (th place o f a realization o f T , having that property is sufficient for satisfying the functional predicate. Conversely, satisfying the /th functional predicate is sufficient for having some first-order property that is the i t h member o f some reahzation o f T. These t w o conditions are direct consequences o f a functional predicate's expressing the second-order p r o p e r t y o f having some first-order property that has the relevant functional role. Those conditions are captured b y taking tf to denote the set o f all and only rth members o f realizations o f T. That is expressed thus: ti = d e f {x,: 3 x i , . . . , x , _ i , x, This is equivalent t o : z is or, i f w e Ramsey's predicate

...,x„ T[xi,...,x„]}.

i n tj = d e f S x j , . . . , x „ ( r [ x i , . . . , x„] & y is i n x,) retain the second-order quantification i n terms o f w h i c h theory originally was presented, Pj being the (th theoretical o f T, i t is also equivalent t o :

P,z = d e f 3 9 i , . . . , 9 „ ( r [ ( p i , . . . , 9 „ ] & 9,.^). I n these last t w o forms, one m i g h t discern an intuitive account o f h o w the theoretical terms o f T are defined i f T is equivalent to its Ramsey sentence: i n t u i t i v e , but w r o n g . Evidently, any interpretation o f theoretical terms must account for their occurrence univocally, b o t h w i t h i n and w i t h o u t the theory i f logical relations between the theory and independent predications o f its predicates are to be maintained. O n Lewis's theory that all goes s m o o t h l y : the results o f substituting w i t h i n T t h e Lewis interpretations o f , . . . , ( „ is logically equivalent to the Lewis interpretation o f T. B u t o n m y proposal there is a snag w h i c h requires an adjustment. 52

the theory each (, is taken to denote a set o f first-order states jf wit gsented, and impermissible equivocation is generated. For, the way P o f the theory b y a sequence o f first-order in resards tnc j^jjsfaction ^ ^ „ as '^^^^'^^j-h ' z is i n X;' is interpreted as expressing a relation between states.^^^j^ and first-order states. B u t o n the proposed interpretation indivi ^ame not o f a first-order state b u t a set o f first-order states, ^*d^then is i n ^ i ' expresses a relation between individuals and sets * f first-order states. W e can't have i t b o t h ways. ° The problem is soluble. T h i n k o f each theory as having t w o variants follows. The open sentence that results f r o m substituting variables for theoretical terms can be t h o u g h t o f as having t w o interpretations. O n the first, every predicate '2: is i n x , ' is interpreted as meaning 'z is in some member o f x,', w i t h ' x , ' ranging over sets o f states; this is the upper variant o f the theory. O n the second interpretation, ' 2 is i n x ' is interpreted naturally, w i t h ' x , ' ranging over states; this is the lower variant o f the theory. N o w w i t h regard to a sequence o f states satisfying the theory, the lower variant is the relevant one. T h e n define each (, o f the original theory as ' the set o f states w h i c h are (th members o f sequences that satisfy the lower variant o f T ' . (The metalinquistic aspect o f this formulation can be ehminated.) The theoretical terms, thus defined, can then be substituted for the variables i n the upper variant o f the theory w i t h a fitting result. So the theory T is equivalent to T's upper variant. The lower variant is something the theory carries along w i t h i t ; i t is the lower variant that is, strictly speaking, satisfied or realized, m u l t i p l y or uniquely, and not the theory itself Identifying the theory w i t h its upper variant means that ( i ) theoretical terms occur w i t h i n and w i t h o u t the theory univocally, (2) m u l t i p l e reahzation is accommodated, and (3) exphcit definitions o f the theoretical terms are generated i n the non-theoretical vocabulary, thus explaining h o w a functional theory gives meaning to its theoretical terms. Is a theory on this interpretation i n general equivalent to its Ramsey sentence ? Suppose T contains a conditional whose antecedent contains a conjunction o f theoretical predicates: (A) i f z were i n f, and z were i n tj then z w o u l d be i n Suppose, also, the f o l l o w i n g conditions are satisfied: (1) T i s realized b y only the t w o sequences ; (2) none o f a,, Oj, hi, hj are identical w i t h each other; (3) for every other n, a„ = h„, i n particular, a^ = bi,; 53

.y and

(4) tj occurs nowhere else i n T. Evidently, the Ramsey sentence o f T is true since T is realized \ follows f r o m the proposed definition o f theoretical terms that f, = {„ i , } and tj = {aj, bj}. B u t that, together w i t h ( A ) , (3) and (4), entails that the sequences < . . . , a,, 6 , , . . . > and < . . . , 6,-, aj,... > must also satisfy t},^ theory, w h i c h is precluded by ( i ) and (2). So on the proposed interpretation T is false. O n the proposed is t r u e ; and, as we does not i m p l y the new interpretation sentence.

interpretation o f T , i f T is true its Ramsey sentence have just seen, the t r u t h o f the Ramsey sentence t r u t h o f T o n the proposed interpretation. O n T's T is logically stronger therefore than its Ramsey

Let us say that the sequences < . . . , « , , aj,... > and < . . . , & , , are related to the sequences bj,... > and < . . . , fe„ aj,...

bj,...) > by

recombination in thejth place. The reason w h y the Ramsey sentence isn't strong enough to capture the theory that contains (A) is this: for it to be true o n the proposed interpretation, the set o f its realizations must be closed under recombination i n the ^ t h place. That a theory is realized by several sequences does not guarantee that their i t h members each interact w i t h alljth members as the theory says tj and tj should interact; the Ramsey sentence may be true w i t h o u t such inter-realization interaction. B u t o n the i n t u i t i v e conception of a functional theory, i f a certain first-order state realizes a certain functional state, being i n the former state is sufficient for being i n the latter. So i f the theory says t w o functional states interact i n a certain w a y , all combinations o f the first-order states w h i c h realize those functional states must interact thus. It could appear that the new interpretation o f T can be captured thus: T i s realized and the set o f T's realizations is closed under recombination i n all places. B u t there are theories that are true under the new interpretation whose realizations are not closed under recombination. Suppose, for example, that T contains (B) i f X is i n f,-, then x is i n tj and that f, occurs nowhere else i n the theory. A n d suppose < . . . , a,, aj,... > and realize T , and that a,, ay, i „ bj are all distinct states. Does i t f o l l o w f r o m (,'s denoting {a,, h,} that np^ ^ "Hovf the p r o b l e m w i t h ^"'^'"ducing this functional state thus is that (B) is not strong enough '""^ nsure that any pair o f states having those functional roles is ' the ' " l i e f that p ' and 'p-attentiveness'. T h a t is a general feature o f the commonsense theory I am sketching; the p r o b l e m gets a general solution i n 4.6 N o w let me simply observe that the p o i n t is t o specify commonsense theory w h i c h implies something about each belief which i t imphes about no other. (B) is still too strong, for a n o r m a l observer can fail to accept the evidence o f his senses. H e can believe things not t o be as they appear, when i n fact they are. A non-scepticism c o n d i t i o n is needed, and so let us speak o f z's being open t o p , w i t h this result: (C) I f p is true, 0 ( p , z , t), z is p-attentive at t, and z is at < open to p , then z believes at t that p . H o w is ' z is open to p ' to be construed? There are t w o possibilities. First, i t is definable; secondly i t is a basic functional t e r m o f the theory on a par w i t h 'p-attentive'. As regards the first, w e take at face value an apparent observational fact p (i.e. w h e n p is true, O holds, and w e are p-attentive), unless countervailing beliefs forestall believing that p. This is not t r i v i a l , for many beliefs are not produced w h e n other beliefs do not forestall t h e m . So ' z is open t o p ' could be defined as 'z's other beliefs do not forestall z's believing t h a t p ' . (The resulting interpretation o f (C) leaves i t non-trivial.)

N o w (A) is still too strong, for observable facts can stare one i n the face, unnoticed t h r o u g h lack o f attention. I n our commonsense psychology we do employ the concept o f attention i n this connection. Mere general attentiveness is not enough: let us say that z is p-attentive i f z is attending to those aspects o f z's observational field that, given O (p, z, t), are n o r m a l l y sufficient for z's perceiving that p. W e then have:

This w o n ' t do as i t stands, for unqualified counter-beliefs may be absent while negative estimates o f the probabilities make one suspicious. Since in this functional theory the cognitive attitude is not belief-to-a-degree, but straight-out acceptance, this creates a complication. I am sceptical o f the psychological reality o f degrees o f belief defined as a probability measure, and even more sceptical that any specific decision function, such as m a x i m i z i n g expected u t i h t y , is generally ascribable. Perhaps a viable n o t i o n o f subjective probabihty can be generated f r o m c o m parative conditional p r o b a b i h t y : ' z believes that, given 5, q is more likely than r ' , i n some extension o f this functional theory, thereby enabling us t o define ' z is open t o p ' as 'z's subjective probabilities do not forestall z's beheving that p . '

68

69

T h e second m e t h o d is to treat ' z is open to p' as a basic functio term o f the commonsense theory, and i t is plausible that some su u unified factor is operative i n our n o r m a l dispositions to perceptujl belief A c q u i r i n g concepts like 'this is red, is a rabbit, is larger th t h a t ' seems to be i n part to acquire dispositions to make the relevant judgments, i n appropriate circumstances, provided one is receptive to p\g the case. Suppose q is logically weaker than p, and that an ordinary believer w o u l d believe q on the basis o f believing p; so, p m i g h t be 'there's a dog over there', and q 'there's an animal over there'. T h e n the belief that q satisfies (C) w i t h p i n the antecedent; so being i n a state which satisfies (C) is not sufficient for believing that p. The resources are at hand for dealing w i t h this. For every non-negative observational proposition p, let not-p also be observational. The input condition ' i f not-p is true, etc., then z believes that n o t - p ' w i l l not i n general hold o f the belief that not-^ (keeping the antecedent constant, that is), i f not-ij is stronger than not-p. N o w suppose internal constraints o n negation do adequately establish what i t is for t w o beliefs to be negations o f each other. The w h o l e functional theory then implies something about the belief that there's a dog over there that i t does not i m p l y about the belief that there's an animal over there, viz. that i t is functionally the negation o f a belief whose input condition is that there is not a dog over there. So the p o i n t has been to specify a unique functional condition for each observational belief; and since, i n the antecedent, ' p is t r u e ' does not occur i n the scope o f 'beheves', we help ourselves to it unproblematically. Relative to the independent enumeration o f observational propositions (C) is a schema for as many input conditions as there are observational beliefs.

nds i n part on our scientific theories. I t is i m p o r t a n t to jjjeories epe differences between explicating belief i n general and keep i*^ ^ functional system that determines interpersonally charaete ^^jj^£ content i n a certain society. A radical revision o f jscriba e j^gjjgf dispositions may block intentional synonomy observ j g j Q j i . (The general distinction between functional role and .,-rr)SStnt'

,

/-

1 1

n

r

.1

1

^'''^'h^conditions and reference w o u l d allow reference to be preserved '^'^^ revision, even though functional content is not. The question o f OS: *'^'^°rnniensurability for theories separated by conceptual r e v o l u t i o n is a"oubtl*ess affected b y that distinction.;

5

UNIQUE

SYSTEMATIC

ROLES

FOR

NON-OBSERVATIONAL

BELIEFS

The next step is to specify further conditions w h i c h secure systematic uniqueness for non-observational beliefs. The idea is to find interpersonally ascribable constraints o n m i n i m a l rationality w h i c h i m p l y , for each non-observational belief, a unique relation to a given «-tuple o f observational beliefs. There are t w o classes: those concerning logical form (L-constraints), and those concerning non-observational predicates (M-constraints). The former, together w i t h the input conditions, secure uniqueness for non-observational beliefs whose o n l y non-logical constituents occur i n observational propositions (A-type behefs). I t u r n in 4.7 to the question o f uniqueness for beliefs that have at least one non-observational component (B-type beliefs). L-constraints, and uniqueness for A-type

beliefs

I assume all beliefs are expressible i n a first-order language w i t h indexicals. For those w h o find this unrealistic, m y account can count as a fragment o f a fuller theory, w h i c h w o u l d contain constraints o n non-first-order factors — m o d a l operators, higher-order quantification, causal connectives.3

Does recognizing observational behefs c o m m i t me to the observational—theoretical distinction o f empiricist epistemology and theory o f meaning that has come under heavy attack ? I t is probably incompatible w i t h something i n the recent literature to identify some beliefs as functionally observational. B u t let us be clear about w h a t is not i m p l i e d . I t is not implied ( i ) that any beliefs are i m m u n e to counter-evidence, or even intrinsically credible, or (2) that acquiring observational concepts is just to acquire dispositions to produce such behefs i n the relevant perceptual circumstances; that also involves assimilating i t into a theoretical framework. A n d (3) i t is quite compatible w i t h m y account that w h a t counts as data for scientific

The L-constraints that f o l l o w are, as psychological assumptions, undemanding. Indeed, i f any o f them is not b y and large true o f z, beliefs o f the relevant logical f o r m seem not ascribable to z; the required rationahty is so m i n i m a l that their approximate satisfaction is partially definitive o f having the relevant beliefs. A n interesting

70

71

Naturally, the

first-order

quantification in terms o f

restriction allows for expressing modalities, first-order

higher-order

quantification over possible worlds, properties,

propositions, etc.

by-product o f their undemanding character is that they do not ii that the set o f a person's behefs is deductively closed — that is, that a perso^ beheves all the deductive consequences o f his beliefs. I n w h a t follows, person and t i m e variables are o m i t t e d , and ' =^ > is the counterfactual connective. So, ' Bp & ^ => Bp, B^' abbreviates ' i f z were t o believe that p & ^ at f, z w o u l d beheve that p at t and 2 w o u l d believe that q at t\e constraints are not supposed to be exceptionless; the whole theory is prefixed b y ' i n general'. Their consequents are negative, but I have appended to some, i n square brackets, positive consequents w h i c h may not be too strong. The L-constraints are supposed collectively to i m p l y something distinct about each logical f o r m . Constraint (yb), o n the existential quantifier, may seem out o f place, but i t plays a key role i n chapter 4. The

L-constraints

(la) (lb) (2a) (2b) (3a)

Not B ~ ( p B p => not Bp & 5 => Bp and B^ B ~ p and

V ~p) B~p not B ~ p , not B ~ 5 => not B ~ (p & ^) B ~ 5 => not Bp W q

[Bp, B^]

[Bp V 5 and B ~ p => Bq] [Bp V 5 and B ~ 5 => Bp] (3b) B ~ ( p V ^) = > n o t Bp, n o t B5 [B~p,B'^q] (4a) Bp ^ q and Bp => n o t B ~ ^ [B^] (4b) B ~ (p -»• 5) => not B ^ p , not B^ [Bp, B ~ 5] (5a)B(i=f2 and B F f i ^ n o t B ~ F f 2 [BFt2] (5b) not B ( 3 x ) ( x + x) [B{\/x){x = x)] (6) B'ixFx=> not B~ Ft (7a) B ~ ^xFx => n o t BFt (7b) BFt and BGt => not B ~ 3x(Fx & Gx) (8a) B 3 x ( F x ) => not B V x ~ Fx (8b) B V x F x => not B 3 x ~ Fx. These are quite m i n i m a l ; i t is hard to see h o w one could ascribe behefs that exhibited widespread exceptions. T h e y do not require that one is even r u d i m e n t a r i l y proficient at m a k i n g inferences (except for those i n square brackets, w h i c h are perhaps then dubious). Each connective, apart f r o m the quantifiers, has t w o conditions parallel to i n t r o d u c t i o n and elimination rules for natural deduction except that they require one not to believe a certain t h i n g — the negation o f what the rules infer. T h e reason for including b o t h w i l l become clear w h e n we t u r n to t r u t h conditions in 8.3 ; i t secures that the functional system 72

rninimally demanding, still strong enough to impose certain is, •^.'^'jons on assigning t r u t h conditions. resti'^'j ^^^^ noted, these constraints do not i m p l y deductive closure rson's behefs. There is no constraint l i k e : ' i f x were to believe fo'' ^.^ould believe p or ^ ' ; or ' i f x were to believe V x F x , x w o u l d ^'v eFa', both o f w h i c h introduce new subject matter. N o constraints nond to certain other ways o f securing completeness, e.g. '^'"^ciative rules l i k e : ' p or q and r or 5' implies ' r and p or r and q * ^ and p or 5 and q'. W e can make sense o f z's believing the former without the latter because z may satisfy the L-constraints o n conjunction and disjunction, w h i c h do not yield the associative rule. Certain observational beliefs, the " s i m p l e a t o m i c " , are usefully demarcated at this p o i n t : they contain o n l y one occurrence o f a predicate, i.e. a property or relation-in-intension, and have the f o r m [that is F ] , [this is R to that] etc. ( ' t h a t ' and ' t h i s ' are indexicals to be accounted for i n chapter 5). Simple atomic beliefs are the foundation o f the uniqueness o f A - t y p e beliefs, those whose o n l y non-logical constituents occur i n simple atomic observational beliefs. N o w the input conditions and L-constraints together i m p l y something unique about each belief, w h i c h is not to say something is uniquely true o f each belief b u t , more weakly, they i m p l y something about each behef that they i m p l y about no other. T h e significance o f this w i l l emerge i n the next section. Let us see that the weak uniqueness property holds. A-type propositions have different degrees o f c o m p l e x i t y . Simple atomic propositions are o f degree O ; their negations, other direct t r u t h functions, and quantifications o f them are o f degree i , etc. The theory implies something unique about each belief o f degree O , and, b y the L-constraints, for any belief B o f degree n - | - i , the theory implies connections w i t h a set o f beliefs o f degree n that i t implies between no other belief and that set. Thus suppose Bp & ^ is at level i ; (2a) relates i t to B ~ p and B~q, (also at level i ) w h i c h L-constraint ( i ) relates to Bp and B^ at level O. A glance at the other constraints shows that no other belief is related, b y those indirect functional relations, to that pair o f simple atomic beliefs. I n general, the theory implies something unique about Bp & 5 ( o f degree n - | - i ) , or B 3 x ( . . . x . . . ) ^tc., i f i t implies something unique about Bp and B^ ( o f degree n), °'" B ( . . . i . . . ) etc., w h i c h is the inductive part o f demonstrating systematic uniqueness for A - t y p e beliefs. The L-constraints are not deductive; they singly estabhsh functional 73

incompatibilities between beliefs, but chains o f such functional relati^ may not do so. T h a t a certain chain o f functional relations obtaj''^ between an A - t y p e belief and a set o f observational beliefs, does n ^ i m p l y that the latter are deductive consequences o f the former ' verification conditions or anything like that. O b v i o u s l y A - t y p , propositions do not i n general i m p l y their atomic constituents, i point is simply that, i n the n e t w o r k o f functional relations implied input conditions and L-constraints, no t w o A - t y p e beliefs have the saiiig relations to a given set o f observational beliefs. The objects or indices o f behefs are fme-grained propositions. So given the L-constraints, behefs i n logically equivalent propositions can have distinct functional roles. Suppose p is A - t y p e and ^ is a logically true p r o p o s i t i o n : the belief that pScq is functionally related t o the belief that 5 i n a w a y that the belief that p is not, although p and p & ^ are logically equivalent. The i n t u i t i o n that the logical equivalence o f s and s' is not sufficient for that o f ''x beheves that s'^ and '''x believes that i ' " " is thus vindicated and not explained away. ( C f Stalnaker, 1976.) N o w fine-grained propositions determine sets o f possible worlds (sec footnote, p. 58) and therefore contain all the i n f o r m a t i o n the latter contain; b u t , as indices o f functional roles, they contain far more information — all that their structure implies.

4.6

UNIQUE FUNCTIONAL ROLES

FOR

BELIEFS

T o specify a "belief-desire t h e o r y " is n o t to specify all principles and generahzations that enable us t o ascribe beliefs to z f r o m w h a t we observe about z. T o specify the latter fully w o u l d be a monumental task, requiring more than w h a t is prima facie about beliefs and desires. In 6.3 I discuss the epistemology o f attitude-ascription, i n particular h o w the various generahzations we rely on are related to those I am suggesting i n this chapter as constitutive, as determining w h a t the functional organization must he, and not h o w we know about i t . T h e L-constraints and perceptual input conditions are m i n i m a l constraints o n z's being functionally organized so as to have beliefs w i t h certain logical structure and content. T h e y are not all o f the relevant constraints, b u t having t h e m i n hand permits raising a certain fundamental question about explicating belief-ascriptions i n terms o f functional role. T h e question concerns the relation between their conjointly i m p l y i n g something unique about each (A-type) belief and

74

, J gach behef-ascription is to be explicated as expressing a the " i ^ V ^ c t i o n a l role, uniq"^^ enterprise is the conservative explication o f attitudeTha*^ functional terms certainly suggests that c o m m o n sense gscripM^^^^^^j^g J functional theory adequate for capturing Bel and that a sufficient c o n d i t i o n o f believing p is being i n a state defined pes" folk theory. N o w , I am c o m m i t t e d t o this: the by ^ ojation o f these constraints is necessary for a functional theory uately to define belief-ascriptions. B u t , I believe, no commonsense ^ nst'raints constitute a functional theory strong enough to define Bel ''°d Des This raises a fundamental question about the project. B u t first let us see that these conditions are not sufficient. The L-constraints and input conditions do i m p l y something unique about each belief something they i m p l y about no other behef; but they do not i m p l y about each belief something uniquely true o f i t . For, a permutation o f the set o f all beliefs (i.e. types) can, for a given person, satisfy all those counterfactual constraints just as w e l l as the beliefs mentioned b y t h e m . Let r be the proposition that Jones is a heavy smoker, and p that Jones w i l l die o f l u n g cancer, and suppose that i f S m i t h believed r he w o u l d not believe ~p. N o w consider L-constraint (2 a): 'Bp & 5 =^ B ~ p ' . T h e counterfactual relation w h i c h that asserts holds between Br and B ~ p ; for i t says merely that they w o u l d not occur together. N o w B r may not satisfy the other conditions the constraints associate w i t h Bp &c q (for some specified q), but that can be managed. What is necessary, for B r to satisfy the remaining constraints o n Bp & q, is that i t satisfy, first, the counterfactual c o n d i t i o n expressed by ' B p & 5 = > N o t B~q', and secondly, the one expressed b y 'Bp & 5 ^ N o t B ~ ( p & q)'. B u t they are easily satisfied i n our example; as regards the first, we simply find some q such that i f Smith believed r he w o u l d not believe ~q; and, as for the second. Smith's belief that r presumably satisfies ' B r = > N o t B ~ r ' . Thus, as regards Smith's cognitive organization, B r satisfies the conditions the theory so far requires o f Bp & q. It w i l l be w e l l to discuss a natural response to this demonstration that the commonsense constraints do not individuate content. The counterfactual connections between Smith's belief that r and other beliefs w i l l not be as entrenched as those for his belief that p &c q. Suppose Smith learned o f an effective and available cure for l u n g cancer: that

75

m i g h t sever the connection between his beheving r and not beUevir, ~ p; but n o t h i n g similar could apply to believing p &c q and believing ~ p; new i n f o r m a t i o n does not unravel rationality. This suggests adding to the L-constraints a meta-constraint: For any q, the L-constraints w o u l d remain true o f the whole set o f z's beliefs, even i f z believed q. (This is not exactly r i g h t , for suppose q= [ ~ ( p V ~ f ) ] : believing it is inconsistent w i t h the L-constraint ( l a ) . So the meta-constraint is restricted t o propositions q such that believing q satisfies the L-constraints.) N o w this stability c o n d i t i o n on rationality may seem reasonable; but i t is too strong. For these constraints are about first-order states, and i t seems possible for a malicious brain scientist to change Smith's w i r i n g so that i f contrary to fact. Smith were i n the underlying state that for h i m has the counterfactual role associated by the overall constraints w i t h the belief that every tree has red leaves, his overall rationahty w o u l d be destroyed; he w o u l d thereby cease to be capable o f beliefs. B u t the meta-constraint is inconsistent w i t h that possibility. So i t is not easy to rule out such permutations o f the set o f our possible beliefs that satisfy the L-constraints as w e l l as those mentioned by the constraints. There are other reasons for the non-sufficiency o f the conditions.

s o f our beliefs ( w h i c h do not preserve logical equivalence) peroint^ ^ ^ given person also satisfy the L-constraints. whi'^b^"^!^ c o m m o n sense for further constraints that make the To sufficient is fruitless, for further strengthenings threaten to con'ii^"'" interpersonal ascribability. I shall n o w suggest h o w the nionsense constraints, and their unique implications (in the weak co"^ j^^yg a key role that does not require them to be sufficient, that ^^''^ exhaust the functional theory w h i c h defines Bel and Des, but that ^ertheless allows conservative functional explications o f ' believes that "•Tnd'desires that 5'. Suppose that, o n a certain well-confirmed functional theory T about Smith, generated b y theoretical psychology and neurophysiology, there J5 a striking, i f o n l y approximate, one—one projectible correlation between ordinary unproblematic well-confirmed ascriptions o f beliefs to Smith and Smith's being i n functional states defined b y T. T h a t correlation w o u l d not o n its o w n warrant identifying specific beliefs w i t h specific functional states. For consider this analogy. Suppose that, i n the demonology o f a certain people, specific demonic activities b y specified demons are alleged to explain certain observable events, and that each such type o f observable event does have a distinct unified, scientific explanation, i n terms, say, o f chemical or electromagnetic factors. That w o u l d vindicate something o f the demon theory — namely, the insight that those event-types have unified causes; b u t i t w o u l d not vindicate describing those causes as demons, for demons also must be the sorts o f things that plot and strive, etc. Evidently a reductionist vindication o f one framework i n another must not o n l y preserve ways of individuating states, etc.; i t must also vindicate the satisfaction o f the predicates o f the one framework b y the states, etc., o f the other.

(1) The L-constraints are not demanding: they are about certain states' not co-occurring. H o w could w e show that no neural states w h i c h are not beliefs satisfy the relevant conditions? I t w o u l d not be impressive to insist o n specific counterexamples. (2) There are permutations o f behefs o n w h i c h logically equivalent beliefs are interchanged, such that z's beliefs on such a permutation w o u l d satisfy the L-constraints i f z is logically astute enough, w h i c h threatens the fine-grained criterion o f individuation.

So the mere projectible correlation o f T's functional states w i t h beliefs as o r d i n a r i l y ascribed does not make those functional states o f T beliefs. Suppose T also has this feature: the L-constraints and input conditions (and all other constraints to be specified) correspond, o n that correlation, to counterfactuals i m p l i e d by T that are defining conditions o f T's functional states, and w h i c h may not exhaust T. Suppose i t turns out that for each person there is a theory T' on w h i c h such a correlation and the foregoing c o n d i t i o n h o l d . T h e n , I suggest, the functional states o f each such theory T', as applied to a person z, should be counted as z's beliefs and desires. For they satisfy the/«// complement o f rationality constraints, recognitional abihties, and so o n , imposed b y c o m m o n sense as necessary conditions o f having beliefs and desires.

(3) There are almost certainly permutations o f psychological states that satisfy just as w e l l the input conditions as the ones the theory covers. The roles o f p-attentiveness andp-openness are there supposed, together w i t h p's t r u t h and the obtaining o( 0{p, z,t), to lead to z's believing that p. B u t that states a, h and c satisfy that c o n d i t i o n does not make them p-attentiveness, p-openness and the belief that p. Thus, let a be the belief that i f p then q, and b the combined state o f p-attentiveness and p-openness, and c the belief that q. I t could happen that i f p were true and 0 ( p , z, (), and z were i n a and b, then z w o u l d be i n c. Evidently there are also other ways those functional roles could be realized. A p a r t f r o m the foregoing, one suspects there are other possible 76

i i

77

Their projectible correlation w i t h ordinary ascriptions could tn moreover, that they have the causes and effects we ordinarily ascrib ' behefs and desires, and w h e n they t u r n out not to (e.g. where a sec^° to c o m m o n sense t o cause h, T m i g h t count t h e m b o t h as the comrti effects o f c) w e can take T to correct c o m m o n sense. The t r u t h o f s^,,)^ functional theories w o u l d vindicate the supposition that there are si states as beliefs and desires. The commonsense constraints determine, then, a class o f functional theories o f individuals; they are the interpersonally ascribable core o f such theories. This is not to say that c o m m o n sense thereby dictates to theoretical psychology. The picture is not o f the former evaluating the latter, but the reverse: the concepts o f belief and desire are shown not to have objective application i f correct theoretical psychology ' theories do not vindicate t h e m by satisfying the commonsense constraints. The upshot is that i t does n o t matter i f permutations o f our ordinary assignment o f beliefs equally w e l l satisfy the constraints. For the idea is not that any state that satisfies the commonsense constraints on the belief that p is the belief that p, but rather that the belief that p must satisfy t h e m , and, i f z's functional states satisfy the correlation condition, and o n that correlation satisfy the commonsense constraints, then they are to be counted as z's beliefs and to have the corresponding contents. N o w this may be puzzhng, for i t could seem I am suggesting that c o m m o n sense b o t h does and does not supply sufficient conditions for having a given belief N o w , w h a t c o m m o n sense provides is a meta-constraint o n a functional theory T's adequately defining a unique functional c o n d i t i o n for a state t o be identified w i t h a given belief The commonsense constraints are not strong enough to define those functional states o n their o w n . W h a t then determines the meaning o f ' B e l ' and 'Des'? The p o i n t is that, because the relevant functional theory may vary i n its further details f r o m person to person, Bel and Des are, strictly, systematically ambiguous.

beliefs. W h a t is present are meta-conditions on the com'^°"*Q'f any such explication. This suggests, contrary to w h a t one sideq^^'^ that certain classical scientific reductions — e.g. o f the gas iiiigb'^ e^'P jjjtistical mechanics - count as conservative explications laws w i t h i n ^ in the sens^ ^ general p o i n t t o be made about c o m m o n sense and ^'^^'^^alism i n the philosophy o f m i n d . One model has been that, function^^^^^^ ^ ^.^^^ mental predicate P as expressing a certain m exp ^^^^^^ ^ certain entrenched subset o f beliefs i n v o l v i n g P make ^""'^"fonctional theory w i t h i n w h i c h P is defined. I am suggesting a model. Take certain commonsense conditions as necessary for P "^be satisfied, and to determine necessary and sufficient meta-conditions for r to be a functional theory w i t h i n w h i c h P can be explicated. T may then contain m u c h that is not i n c o m m o n sense, and i n general had better do so i f it is t o generate sufficient conditions for given mental states. (Some problems about whether phenomenal properties are functional properties may appear different w i t h i n this version o f functionahsm.) What m i g h t the rest o f an adequate belief-desire theory T be like? One possibility is suggested by certain recent theories about conceptual role. Hartry Field (1977) takes the conceptual role o f a sentence i n z's language o f t h o u g h t t o be determined b y its subjective conditional probabihty given each o f z's other sentences. The conceptual, and hence functional, role o f each o f z's beliefs w o u l d presumably be unique within that extended functional theory o f z. Since conditional probabilities differ among persons, such functional systems, w i t h i n w h i c h beliefs w o u l d have unique functional roles, w o u l d not be interpersonally ascribable; but the core o f such functional theories (the constraints o f rationality, etc.) w o u l d be.

H o w does this p e r m i t conservative explications o f ' believes' and ' desires' ? I f I were presented w i t h such a true theory T o f z, I w o u l d accept the predicate ' z is at M n Bel[p, z, f ] ' , that T defines, as a conservative exphcation o f ' z at f beheves that p ' . For that w o u l d by m y lights show that there really is such a state as z's belief that p. Conservative explication differs f r o m classical analysis, for not all the exphcating conditions need have been present antecedently i n our

But the general idea o f such a broader functional theory does not require the part o f the theory outside c o m m o n sense t o contain functional relations that correspond to cognitive intentional, or conceptual, relations. Moreover, as I hope is clear, m y suggestion has no implications about the overall shape o f theoretical psychology. Thus it does not i m p l y that intentional models are t o be employed i n the functional subsystems o f psychological theories. N o r does i t i m p l y that A e proper research strategy is " t o p - d o w n " - that is, proceeds f r o m general functional theorizing t o neurophysiological specifics. How i t might be discovered that something i n our functional organization corresponds to the commonsense constraints is completely open.

78

79

Theoretical psychology could vindicate commonsense psychology i f i t were nowise guided by i t , even i f i t d i d not take commonse psychology seriously. ^ There is a large gap i n the foregoing, having to do w i t h vindicatit, ascriptions o f propositions to particular functional states. This is relatej to w h a t may have appeared as a discrepancy between m y concern 1,^ 4.5 that the constraints should i m p l y something unique (weak sense) about each belief, and then p o i n t i n g o u t that they may not imply something uniquely true o f each belief that they do not add up to a functional theory w i t h i n w h i c h Bel[p, z, t] is explicated. W h y is i t desirable that commonsense psychology should imply something unique about each behef? That does not enable com monsense to distinguish Bel|p, z, t] and Bel[5, z, t], w h i c h may have to w a i t upon science. The key lies i n the general matter o f vindication. As the demon example seems to show, a mere one—one correlation between the functional states o f a well-confirmed psychological theory, and our well-confirmed belief-ascriptions, w o u l d not thereby make sense of ascribing predicates o f the f o r m 'is the behef t h a t p ' . The ideology of commonsense psychology w o u l d still need vindicating. Suppose the best w e could do, by w a y o f interpersonally ascribable commonsense constraints, were the L-constraints. Describing a functional state as the belief that rats eat cheese w o u l d then under-utihze the material i n the that-clause, for its o n l y aspect that w o u l d correspond to the constraints (the o n l y aspect o f the ascription w h i c h w o u l d be determined by interpersonally ascribable generalizations o f c o m m o n sense, and capable o f being vindicated functionally) w o u l d be the logical form o f the that-clause. N o more detailed features o f ' i s a belief that rats eat cheese' could be vindicated i n functional terms. B u t i f the commonsense constraints determine something unique about each belief, every distinguishable aspect o f a that-clause w o u l d correspond to some aspect o f commonsense constraints o n functional role. If, then, all the constraints are i n fact satisfied b y the theoretical psychology, every semantically active aspect o f a that-clause is thereby vindicated, and thereby contributes something to the satisfaction o f ' is a belief that p ' b y a given functional state. This w o u l d not be so i f those constraints did not collectively i m p l y something unique about each belief So although the constraints do not determine a unique functional role for each belief the fact that they i m p l y something about each belief w h i c h they i m p l y about no other behef means that each distinct aspect o f content encodes a distinct aspect o f functional role. Each clause ' t h a t 80

distinct necessary conditions for a functional state t o be f ''^f^^T^s 'the belief t h a t p ' . The full sufficient conditions w o u l d be aescribe^ ^ broader functional theory that defined Bel[p, z, t] i n pr"^ J w i t h the correlation c o n d i t i o n . jccordan^^ i m p o r t a n t consequence o f taking Bel's exphcation to w a i t ^^theoretical psychology, namely, i t introduces flexibility i n t o h o w u p " " ^ -.^g the commonsense constraints. Suppose the input conditions ^'^'^''discovered to be too weak b y virtue o f there being some further ^'^'^dition on f o r m i n g observational beliefs apart f r o m those i n (C) - i.e. nenness, and p-attentiveness. T o anticipate this possibility we *o'uW regard the antecedent as containing a place-holder. (This, as I shall suggest i n 4.9, is a desirable m o v e i n specifying commonsense constraints on h o w belief and desire lead to action.) B y relaxing the role o f the commonsense constraints, t o constrain functional theories rather than constitute t h e m , their f o r m can be more schematic. W h a t is essential is o n l y that i f intentional i n d i v i d u a t i o n conditions are to be vindicated i n functional terms, each aspect o f a that-clause must correspond to an interpersonally ascribable constraint o n functional role.

4.7

SYSTEMATIC

ROLES

FOR B-TYPE

BELIEFS

So far, uniqueness o f systematic role has been established o n l y for A - t y p e beliefs. N o w B-type beliefs, whose indices contain non-observational concepts, constitute a dauntingly large class; to characterize their systematic roles w o u l d be r o u g h l y the same sort o f task as the semantic description o f a w h o l e language. M y purpose here is considerably more modest — namely t o indicate i n a general w a y the f o r m o f such a characterization o f systematic role for B-type behefs. The idea is like Catnap's idea o f meaning postulates (Carnap, 1947), that is, constraints o n non-logical terms that are partially constitutive o f their meaning, as i n ' r e d is a c o l o r ' . I shall call such constraints o n beliefs " M - c o n s t r a i n t s " . W e can expect resistance to the n o t i o n o f M-constraints, b o t h o n the Quinean g r o u n d o f i m p l y i n g analyticity, and also i n connection w i t h the claim that terms do n o t necessarily change their meaning w h e n theories containing t h e m change. These raise different issues. First, no n o t i o n o f analyticity is presupposed here. Rather the idea is to account for interpersonal s y n o n y m y i n terms o f interpersonally ascribable constraints on functional roles. The M-constraints are 81

constraints o n a class o f functional theories, whose satisfaction , i n t u i t i v e l y m i n i m a l c o n d i t i o n for having such and such beliefs, -wj ' i n t u i t i v e l y m i n i m a l ' adverts to conservative explication. B u t both ^ relevant functional roles, and the methodological concept o f conserv u, explication, have their place w i t h i n a naturalistic framework. Introducing M-constraints naturally does c o m m i t me to denying th second p o i n t i n its strongest f o r m . B u t i t has a weaker, but still substantive, f o r m compatible w i t h the current p o i n t : to discard a thcor that contains a certain t e r m , i n favor o f a distinct theory that also contains i t , is not thereby to change its reference. Functional role and reference are i n general distinct, and i t is quite possible for reference to remain the same w h i l e functional role changes. The n o t i o n o f M-constraints could suggest something far too strong - that a person's beliefs constitute a theory i n w h i c h each non-observational concept can be defined a la Ramsey or Lewis (cf. 3.2). B u t i t is no more required for non-observational concepts than for logical structures that the constraints i m p l y something uniquely applicable to each. In general, the desideratum is interpersonally ascribable constraints that i m p l y something about the functional role o f each belief that they collectively i m p l y about no other, w h i c h is a m u c h weaker c o n d i t i o n than Ramsey-Lewis definability. The full functional theory adequate for defining 'Bel\p,z,t]' is not comprehensively determined b y commonsense constraints; M-constraints do not have t o guarantee that any state w h i c h satisfies them is such and such belief M-constraints are not meant to be verification conditions. The requirement o f systematic uniqueness imphes that there are functional connections, perhaps indirect, between every non-observational belief and some set o f observational beliefs, but these relations are not supposed to m i r r o r confirmational relations. Thus the belief that {p-*q) \/ rh related via L-constraints to the belief t h a t p , but the latter does not count as evidence for the former. So w i t h M-constraints, Suppose 'z believes that electrons are smaller than apples' is among them. Together w i t h other constraints, i t may establish a chain o f functional relations between some belief about electrons and the observational belief that there is an apple before me. The latter is evidently n o t thereby i m p l i e d t o be among the verification conditions o f the former.

(is next to. A n d there are relations o f contrariety (nothing syrnf"^"^^, ^jj over and green all over), o f determinable-determinate is both ^^^J^^^^ o f subkind to k i n d (iron is a metal). There must be (red f°hese' to establish m a n y distinctions in functional role among erioiiS °^jjg£5 B u t probably systematic uniqueness is not thereby ^'^^^^j because o f pairs o f concepts that exhibit symmetries - e.g. "^'^'^'^V,'' ' s o u t h ' . So the M-constraints must i m p l y interpersonally se 'nor ble functional relations o f B-type beliefs to observational behefs. asc Consider non-observational natural-kind concepts. Suppose Louis , the difference between a beech and an e l m , or a l u m i n u m and "^^V bdenum (to use examples o f Putnam's). W h a t interpersonally """ibable constraints m i g h t functionally distinguish those beliefs o f Louis's that we describe using 'beech' and those using ' e l m ' ? I suggest taking such beliefs t o be indirectly metalinguistic. Suppose we say, pointing to a tree, that Louis beheves i t to be an elm (he has taken someone else's w o r d for i t ) . O n e t h i n g w e m i g h t thereby be saying is that for some N, ( i ) Louis believes i t is o f the k i n d called N b y the experts, and (2) the k i n d called N by the experts is elm. (This allows that Louis doesn't speak Enghsh.) A variation w o u l d be to substitute for the relation among N, e/m(kind) and the experts, a causal relation among N , elm and the speakers o f Louis's language. (These are i n effect description theories that capitalize on the insights o f anti-description theorists.) N o w since the concept o f such and such hnguistic expression involves the n o t i o n o f perceptual repeatables, asymmetrical functional relations o f such beliefs and observational beliefs w o u l d thereby be secured. The p r o b l e m n o w is h o w uniqueness o f systematic role can be secured for the concepts 'is called', 'experts', 'caused'.

N a t u r a l examples o f M-constraints are those that establish entrenched properties o f relations - e.g. the transitivity o f 15 north of and the

Because a metalinguistic account is an i m p o r t a n t ingredient o f any workable theory o f interpersonally ascribable belief content, i t may be useful here t o consider what m i g h t establish the functional roles o f behefs i n v o l v i n g the semantic t e r m 'x is called y and z ' . N a t u r a l l y w e do not ascribe to all believers the mastery o f an explicative theory o f semantic concepts (e.g. as i n chapter 10) w i t h i n w h i c h that relation can be defined. B u t consider this. A m o n g English speakers there is an entrenched belief that sentences o f the f o r m ' N ' s are called N by us' are true. N o w i f w e count t h e m among the M-constraints, a unique systematic role w o u l d thereby be secured for x is called y by z, i.e. for beliefs o f that f o r m . The idea is n o t that the latter are all disquotational, but that there is no other relation whose disquotational role i n beliefs o f the former restricted class is projectibly interpersonally ascribable.

82

83

N o w these constraints are o f course not inter linguistically ascribable f they involve having metalinguistic beliefs about English i n the foj, ' N ' s are called N by us' is true. For speakers o f other langu^g^^ however, some beliefs are exactly structurally parallel t o t h e m , if^ is, they have thoughts i n v o l v i n g the concept x is called y by z. w o u l d enable us to ascribe that concept t o t h e m , i n terms o f its uniq^, systematic role, as determined b y structurally parallel conditions. ^ have, then, a broadened n o t i o n o f M-constraint: the interlinguisticalK ascribable condition is not i n terms o f the specific content o f those disquotational behefs, b u t i n terms o f a structure i n their content which seems expressible as a constraint o n functional role. I f such beliefs are metahnguistic, then the combination o f systematic! uniqueness for semantic concepts, via disquotational beliefs, and t h e ! observational status o f quotation descriptions o f linguistic expressions, encourages the idea o f M-constraints relating non-observational to observational beliefs and thereby securing systematic uniqueness for those B-type beliefs. There is a certain i m p o r t a n t class o f B-type beliefs such that, were the interpersonal ascribability o f their functional roles secured, the general topic o f interpersonal s y n o n y m y w o u l d be considerably affected. I mean beliefs that arise f r o m applying the Ramsey—Lewis m e t h o d o f theory interpretation to theories that satisfy the following condition. The O-terms include observational terms (in our sense o f figuring i n the i n p u t conditions), a certain mathematical vocabulary, the concept 'causes', and the predicates 'is a property', 'is a relation', etc. N o w suppose, w h a t seems safe, that w e can establish suitable interpersonally ascribable constraints o n these non-observational O terms. Beliefs i n such theories, Ramseyfied, are then interpersonally ascribable i n the sense that our interpersonally ascribable functional organization makes i t possible for each o f us to have t h e m . This means that, even w i t h o u t comprehensive M-constraints, " t h e o r e t i c a l " beliefs o f a large class are objectively interpersonally ascribable, i.e. their functional roles are.

indirect m o d e l , learning the meaning o f 5 requires learning * ''^'tain perceptions constitute evidence for S. B u t the connections that ''^'^'^^ account relate non-observational beliefs functionally to that on ^ ^ ^ j beliefs do not establish a priori evidential connections observati ^^^_^bservational and observational beliefs. The M '"^'^^raints, as I conceive t h e m , make i t conceptually possible to be an '^""^verification realist, i n the sense that they allo w one to think that . be true even t h o u g h we could never get evidence about p's ^uA^f^c)reover, the grounds for ascribing truth conditions t o beliefs are "xterttal to the functional system ( c f chapter 7), and so a functional theory o f belief including a special role for observational beliefs i n establishing uniqueness, is consistent w i t h a metaphysical realist c o n ception o f truth conditions. There is more than one theoreticalobservational distinction; not every one o f them is positivist or anti-realist. That some such distinction is entrenched i n commonsense psychology appears t o me philosophically unexceptionable.

4.8

TRUTH-CONDITIONS,

A N D

T H E D E F I N I T I O N

OF

'BELIEVES'

A belief's functional role, and its t r u t h conditions, are different matters, except for the case o f observational behefs. N o w i n some sense this must imply that functional role does not determine t r u t h conditions — that capturing the former is n o t sufficient for capturing the latter. B u t then i f we consider ' x believes that p' to be equivalent to 'x is i n such and such functional state', that implies that functional role 15 sufficient for truth conditions, for the t r u t h conditions o f a belief are encapsulated in the proposition that is its object.

Finally, let m e compare this general approach t o systematic uniqueness to positivist theories o f meaning i n terms o f observational connections, and the generally anti-realist consequences o f positivism. T h e difference is this. O n a positivist theory, the meaning o f a theoretical statement is constituted by its a priori connections w i t h certain epistemic conditions. So on the simplest verificationist m o d e l , S is equivalent t o a statement that such and such perceptions occur; or.

The inconsistency is merely apparent. T h a t i t is one t h i n g to describe the functional role o f a state and another to describe its t r u t h conditions is compatible w i t h a certain conventional association o f t r u t h conditions and functional roles being built i n t o the meaning o f ' b e h e v e s ' . As i t happens, we choose t o index certain functional roles w i t h such packagings o f truth conditions, propositions; but as regards their adequacy i n indexing, our assignment o f t r u t h conditions to functional roles is arbitrary. T h a t arbitrary association is conventionally encapsulated i n 'x beheves that p ' , just as an arbitrary association o f numbers ^ i t h temperature states is encapsulated i n ' t h e temperature o f y is n° Fahrenheit'. Saying 'believes' i n connection w i t h a proposition is like saying 'Fahrenheit' i n connection w i t h a number. I n explicating temperature ascriptions, w e do n o t expect a justification o f all specifics

84

85

i n the choice o f numbers. Similarly w i t h ' x believes that p functional explication o f ' B e l ' does not motivate assignments propositions to functional roles, w h i c h are nevertheless built into meaning o f ' B e l ' . So there is a sense i n w h i c h functional role JQ^^ " d e t e r m i n e " t r u t h conditions — namely, conventionally; to ascribe functional state using the verb 'believes' is to do so according to * certain conventional association o f t r u t h conditions and functional roles T h a t raises the separate p r o b l e m whether that association can be showi^ on independent grounds not to be arbitrary, and there the analogy with temperature w o u l d end. So the project has t w o stages: first to say h o w each behef is associated w i t h a functional role; secondly, to explain w h y functional roles are are assigned certain extra-mental conditions. I n 7.2 w e shall see h o w to filter :er out these t w o aspects o f a behef i n separate levels o f description; b'Ut n o w , w h i l e w e retain propositions, the t w o levels are inseparable.

I shall use 'desire' and ' w a n t ' interchangeably for the and ^ ' ' ' ' ' ^ ^ a t t i t u d e ; this is not mere t e r m i n o l o g y , for i t implies that generi'-P ^ ^^jfjc pj.o_attitudes, 'likes', ' w a n t s ' , 'values', ' w o u l d hke the an ideal t h a t ' , 'is i n favor o f , 'has a yen f o r ' , ' t h i n k s i t ^ be the case t h a t ' , 'lusts after' are to be exphcated i n terms nting iu general as a unified functional state. T h e contents o f ^^"^^^ ^ matter o f their potential interaction w i t h certain beliefs '^^^leading to decisions. T h e uniqueness o f their functional roles is '"herited f r o m that o f beliefs, and so there is no need to introduce i n p u t ditions for desires; desires are not functionally specified i n terms of their causes or terminating conditions, b u t solely i n terms o f °nteraction w i t h beliefs.

The functional roles described so far do not suffice for explicating behef for beliefs must systematically interact w i t h desire to lead to decision

Certain beliefs, o f a type I shall call instrumental, have a special role: they have the f o r m ' x believes that x's d o i n g A w o u l d b r i n g about p'. It is by virtue o f their mediation that other beliefs influence action: the belief that q and the belief that [ i f q x's d o i n g A w o u l d lead to p] lead to the instrumental belief that x's d o i n g A w o u l d lead to p, w h i c h given the desire that p , sometimes leads to d o i n g A. Thus all beliefs are states o f an input—output system. Formulating the o u t p u t condition o f the behef-desire theory has t w o stages: something connecting instrumental beliefs and desires w i t h certain intentions or decisions, and something connecting i n t e n t i o n to action. It w i l l become clear that the t w o could be amalgamated, w i t h intention cancelling out, thus suiting sceptical intuitions about volitions or intentions as distinguishable states. B u t there are advantages i n introducing this third type o f propositional attitude, w h i c h anyway seems inevitable o n the f o l l o w i n g familiar grounds. Suppose the desire that q wins out i n a conflict o f desires, and an intentional b o d i l y movement results, a k i c k , say. Consider a case exactly hke i t except that something, not a mental state, prevents the kick. W e w o u l d ascribe some propositional attitude other than desire w i t h that b o d i l y movement as its intentional object; the person decided, intended, or simply tried to kick. Moreover that state w o u l d be caused by a desire that q, and the belief that that action w o u l d b r i n g about I f that is true o f a frustrated action, i t is unmotivated t o deny i t o f a successful action, to h o l d that the latter is caused simply b y the belief and the desire w i t h o u t the intermediate state. H o w should that state be described? ' T r y i n g ' is n o t the most appropriate w o r d i n the successful case (although, as A r m s t r o n g (1973) argues, this may be due m o r e to imphcatures than to the meaning o f

86

87

Finally, let me suggest an alternative w a y o f v i e w i n g the relatiion between the t w o levels. Regard the current project, w h i c h is to she low h o w ' Bel[p, z, t]' captures a certain functional state, as g i v i n g a certs tain necessary condition o f believing t h a t p . Accounting for t r u t h conditio :10ns w o u l d then complete the project o f p r o v i d i n g sufficient conditions for L a state's having a certain content, viz. being i n a functional state with such and such associated (truth) conditions. This is not m y preferred account o f the t w o levels; w h y I prefer the earlier f o r m u l a t i o n becomes clearer i n 7.3 and 7.4. Perhaps these remarks w i l l clarify the relation between the functional explication o f B e l [ p i , X j , f], and Quinean indeterminacy. I t is relative to an assignment o f propositions that ' z beheves that p ' has a unique functional role, just as i t is relative to an assignment o f numbers that ' t h e temperature o f x = «° Fahrenheit' has its distinct i m p o r t . The claim o f intrasystematic uniqueness does not i m p l y that those assignments o f propositions o f numbers to functional roles or temperatures are uniquely best. The determinacy o f truth conditions cannot be resolved merely i n terms o f the psychological roles o f beliefs; the question turns on a further interaction between functional states and extrapsychological reality.

4.9

WANTING,

WILLING AND

ACTING

' t r y i n g ' ) . ' D e c i d i n g ' isn't quite r i g h t either. T h e problem is not t l , decisions have to be conscious, but that decisions are events. Runni„^' a few miles is not a basic action that springs fully shaped f r o m an injt ^ decision, like the practiced playing o f an arpeggio; but nor is it 1^ up by a series o f decisions. Rather, one's r u n n i n g is the effect o f onemore or less continuously intending to be r u n n i n g . So a better tertn f^^ this t h i r d propositional attitude is intending. There are t w o strategies for proceeding. The first involves specifying the functional role, not o f intending i n general, w h i c h covers both ulterior intentions behind actions and intentions i n advance o f action but a k i n d o f intending I shall call willing. Its objects are basic action types — b o d i l y movements and mental acts. The general concept of intending w o u l d not then be defined directly by the functional theory but indirectly i n terms o f believing, w a n t i n g and w i l l i n g . T h e second strategy involves i n t r o d u c i n g the full n o t i o n o f intending into the functional theory. There are problems, and I shall present i t later mainly to clarify and to motivate the first strategy by contrast. Certain actions are the p r i m a r y explananda for belief- and desireascription — b o d i l y movements, and intentional mental acts, which I shall n o t discuss. Consider a non-basic action like flying to A b u Dhabi. A l t h o u g h the intention to fly to A b u D h a b i occurs i n its explanation, given enough i n f o r m a t i o n such an explanation can be factored into t w o components: ( i ) actions like boarding the plane and staying aboard, whose explanation contains the i n t e n t i o n t o fly t o A b u D h a b i , certain beliefs and intentions to do those specific actions, and (2) independent facts, like the fact that the plane flew to A b u D h a b i . This division of explanatory labor ends w i t h b o d i l y movements, basic actions that arc not thus explained by further actions and independent facts. Since the independent facts are not themselves explained by those intentions, save per accidens, the p r i m a r y explananda o f the belief—desire theory are b o d i l y movements (or certain extended b o d i l y movements, like pulHng a trigger) described i n " m o l a r " terms. The intentions that explain them are present-tensed, self-ascriptive (in the sense o f 5.3) and have those b o d i l y movements as their objects. T h e y are " w i l h n g s " . W h a t relates belief and desire to w i l h n g ? There must be some generahzation, ascribable to c o m m o n sense i f the functional analysis is to be carried t h r o u g h . Scepticism could arise arise n o w i f one has too narrow an idea o f a belief—desire—willing connection. Call a complete decision theory any generalization that predicts decisions, willings or states o f indifference f r o m the contents, and perhaps degrees o f strength. 88

f beliefs and desires. N o w c o m m o n sense contains no complete of sets o ^j^^^j.y obviously any adequate decision theory must take Jecis'^" what a person thinks might be the case. For to any into a ^ acceptable generalization about h o w unqualified belief affects other"^'*^^j^^^^ counterexamples i n w h i c h , because o f what z thinks '^^'''ht'he so, z fails to w i l l the relevant action. B u t then any decision ""'^ which says exhaustively h o w degrees o f belief and u t i l i t y theory,^^^ decision, is hard to ascribe to c o m m o n sense. Concepts o f h degrees o f belief and desire do n o t yield a complete decision theory while precise concepts o f subjective probability are theoretical. IVloreover, any particular decision theory, like m a x i m i z i n g expected utihty does not seem evidently part o f c o m m o n sense. (This is not meant to deny the usefulness i n theoretical psychology o f complete decision theories.) There is a generalization relating beliefs and desires to willings, not a complete decision theory, b u t ascribable to c o m m o n sense. T h e idea is that desires and instrumental beliefs lead to willings when no competing desires override them. That may sound vacuous; w h i l e i t is not exciting, it is not empty either. W h a t is i t for one desire to override another? Suppose z desires that q and believes that z's doing A w o u l d lead to q. Suppose also z beheves that d o i n g A w o u l d or may also lead to not-p, and that 2 desires p. N o w i f the latter pair causes z not to decide to do A, then z's desire for p has overridden z's desire for q w i t h regard to A. A concept occurs here that we have not introduced functionally, namely, that z believes that d o i n g A may lead to n o t - p ; i t w o u l d seem that degrees o f behef have to be invoked after all. (The qualification, may lead to n o t - p ' , is needed; for one may refrain f r o m an action, not because i t w o u l d interfere w i t h something else one wants, b u t because i t may.) B u t there is perhaps a simpler w a y to get w h a t we need than i n t r o d u c i n g degrees o f belief A way o f " b e h e v i n g i t may be the case that such and s u c h " is b y not believing that i t is not the case that such and such. A n d so a w a y o f " b e l i e v i n g that A may lead to n o t - p " is not believing [ n o t : d o i n g A w o u l d b r i n g about n o t - p ] . This m i n i m a l concept is perhaps all we need to define overriding i n the above context."^ So here is the belief—desire—willing c o n d i t i o n : A more cautious device is this. Instead of invoking instrumental beliefs in the definition of overriding, define ' the desire that q overrides the desire that p with respect to A' simply as 'the desire that q causes z not to w i l l to do A \t requiring it explicitly to have the negative effect via an instrumental belief.

89

I f 2 desires q, believes d o i n g A n o w leads to q, has no desire t l , overrides the desire that q w i t h respect to A, and believes ^} able to do A, then z wills to do A n o w . Some comments i n order o f increasing importance. (1) The variable ' A' is v i r t u a l l y restricted to b o d i l y movement typ^^ B u t no independent definition o f basic action is presupposed; for j „ ' bodily m o v e m e n t , i f z believes i t to be w i t h i n z's power, then, givcjj the other conditions, z wills i t . (2) Actually a further quahfication is necessary. One normally has available several perhaps t r i v i a l l y different incompatible actions that equally implement the desire, among w h i c h one is indifferent. As stated the generalization implies one does every such action; so we should have this: i f a is the set o f all actions o f w h i c h the antecedent is true (for specific q), z does some action i n a. The choice made by Buridan's ass, or a human i n the same predicament, is not predicted by commonsense psychology. W h a t w e w i l l is n o t always determined by our effective desires and beliefs — habit, conditioning and perhaps random neural firings may supplement them to determine one among several equally appropriate actions. (3) The n o t i o n 'z believes z-self able to do ^ ' occurs apparently ineliminably, w h i c h creates a complication. For ability is to be treated as functionally defined, i n the c o n d i t i o n , to be given, f r o m w i l l i n g to acting. So far I have avoided requiring reflexive attitudes i n the theory, beliefs or desires about attitudes. For even i f as some have claimed, fully rational belief requires reflexive beliefs, the most promising explicative strategy w o u l d be first to specify a non-reflexive foundational level o f cognitive and inclinational states w i t h an otherwise full range o f content. So, i n t r o d u c i n g 'beheves z-self able to do ^ ' muddies the waters. The natural m o v e is to introduce a more p r i m i t i v e attitude towards A than the belief that one is able to do A: taking-zlto-be-in-one's-current-repertoire. Beliefs about ability arise w i t h the reflexive awareness o f circumstances i n w h i c h w i l l i n g s are frustrated. The new p r i m i t i v e attitude is then functionally defined by its role in the theory.

a complete decision theory does n o t . B u t this is all t o the in the ^^^^j^^t is obscure to c o m m o n sense. Some have made m u c h good> ^j^gj. jesires do not unqualifiedly lead to decision, b u t are of ^''^j by other desires w i t h no apparent l i m i t t o the process being overr'^^ by our commonsense theory, d r a w i n g conclusions as radical ' " " C ' k r e d u c i b i l i t y o f psychology (Davidson, 1970), the existence o f a* ^., unacknowledged sort o f explanation (Peacocke, 1979), and * xistence o f a new f o r m o f theoretical generalization (Grice, 1975b). ^''u conclusions are, I t h i n k , based upon not seeing this as an instance f a broader and not awfully mysterious aspect o f explanation. Consider forces. Certain o f our general behefs about physical objects can be expressed thus: i f to an object i n such and such circumstances a certain force is apphed, i t moves i n a certain direction, unless a countervailing force prevents that. Again this is not t r i v i a l ; but nor does it motivate any o f the reactions just mentioned to w h a t seems the same feature o f the connection between w a n t i n g and w i l l i n g . Moreover the analogy can be made even closer; for, i t seems to me, i t is o n l y i n a now unmotivated anti-mentalism, one w h i c h overlooks the functional interpretation o f mental concepts, that i t seems w r o n g to think o f desires as like forces. (5) As I earher suggested, the perceptual input condition may contain a place-holder, for a condition supplied b y the more comprehensive functional theories that define Bel, Des, and I n t . This is i m p o r t a n t i n the current context, and also i n connection w i t h intending, as I shall explain. I f there is a place-holder in the antecedent o f the beliefdesire—willing c o n d i t i o n , and a well-confirmed theoretical psychology imphed something o f that pattern for z's states, that w o u l d tend to vindicate ascribing beliefs and desires to z. A similar point is that i f overriding' is n o t explicable i n the w a y suggested, i t may still be a basic functional relation w i t h i n the more comprehensive theory, as w i t h f-attentive', etc. Finally there is the o u t p u t c o n d i t i o n o f the belief^esire theory,

(4) Finally, w e come to the question whether the non-overriding clause makes the generalization trivial. The answer is, not at a l l ; for its f o r m is: ' i f condition C were to obtain and a certain specified condition C * d i d n o t cause n o t - £ , then C w o u l d cause £ ' . That requires something substantial in h o w first-order states interact. The condition leaves i t obscure, o f course, w h a t makes one desire override another,

which I suggest is simply this: I f z wills at f to do ^ n o w , z is able to d o A at t, and no external condition prevents z f r o m d o i n g A at (, then z does A at t. Introducing ability as distinct f r o m freedom f r o m external constraint 's motivated b y the functional theory's being o f an i n d i v i d u a l and not a composite individual-plus-external environment. So ' z is able to do A n o w ' is functionally defined: ' A b l e [ [ x does A n o w ] , z, t]' denotes the set o f states sufficient at t for z's w i l h n g A to issue i n A i n the absence

90

91

o f external impediment. N a t u r a l l y one m i g h t regard ability as expli^ii,, definable via a counterfactual f r o m w i l l i n g to acting, but this functio ^ definition prevents circularity. I n general, i f i t is appropriate to cxpp*^ F via its functional occurrence i n ' i f z were F and G, z w o u l d be ;:^> and i f F occurs nowhere else i n the theory, then ' Fx' w i l l natural] appear definable as ' i f x were G, x w o u l d be H' especially i f the truth o f that counterfactual is independently evidence for being i n functional state F. I f w a n t i n g and w i l l i n g were our only inclinational and conative concepts, i f we lacked the general concept o f intention, then, to account for strategic behavior, we w o u l d have to rely on certain derived wants W h e n n o r m a l l y w e w o u l d count z to intend to do ^ as a means to B we w o u l d have to count z as w a n t i n g to do /4. That seems plausible; it does not i m p l y the implausible stronger derived-want principle that, i i f one wants B and believes that doing ^ is a means to B, then one wants A at least to some small degree. A l l that is required is that in those cases in w h i c h we count z as intending to do a certain action we should count z as w a n t i n g to do that action. W h i l e w a n t i n g can do some o f the explanatory w o r k o f intending, something is evidently left out. B u t 'intends' is not easy to account for. Perhaps i f we could define 'prefers' independently, i n terms of some n o t i o n o f ' o v e r r i d i n g ' related to, but obviously not the same as, that already defined, 'intends' could be defined thus: z intends to do A iff z wants to do ^4, 2 prefers doing A to not doing A, z believes that z w i l l do A, and that behef is non-inferentially caused by that want and that preference ( c f Grice, 1971). N a t u r a l l y w e should w a n t some explanation o f w h y this construction has a unified role i n commonsense psychology. C o u l d w e introduce intending i n general as a basic functional concept instead o f w i l l i n g ? This w o u l d have to cover intending long i n advance o f the appropriate action. The condition w h i c h parallels the belief—desire—willing condition w o u l d be: I f z beheves that doing A leads to q, and z wants q, and no other desire overrides the desire that q w i t h respect to A, and z beheves that z is able to do A, then z intends to do A. This condition is not plausible, for i t amounts to saying that i f one believes one's desires o n balance w i l l be fulfilled b y doing A then one forms the intention to do A. B u t there are reasons for doubting this. First, a person's current desires about the future may favor a given future action, w h i l e the person still adopts a wait-and-see attitude about the 92

In the w i l l i n g c o n d i t i o n , that is ruled out since, i f the action Jecisi""^ immediately, someone w h o d i d n ' t w i l l i t has not a w a n t but j5 iesit^ ^^^^ ^ velleity, or really prefers inaction or another action, some' g f not f o r m i n g intentions towards more distant actions That ^^^^jg^tious. Secondly, the antecedent o f that condition doesn't seel".* that z believes that z w i l l do A. B u t m y intuitions about '^^'^'nding rather favor taking intending to do A to i m p l y believing that '"'^^"will or m i g h t do A. This may partially explain the difference between wanting and intending. B u t not f u l l y : for, to make a w e l l Icnown point, z can want what z beheves w i l l result f r o m A and, l^nowing this, predict that he w i l l decide to do A, w i t h o u t yet deciding to do A- So we add that the belief is non-inferential and caused b y the relevant wants. B u t this is not sufficient because o f deviant ways i n which the non-inferential belief may be caused (e.g. b y a brain manipulation). Thus there are problems about explicating intending w i t h i n c o m monsense psychology. B u t again a more comprehensive theoretical psychology may come to the rescue. For i f functional states o f 2 are suitably correlated w i t h commonsense bases for ascribing intentions to 2, and satisfy conditions something like the ones discussed, then no doubt we w o u l d count them as 2's intentions. Thus c o m m o n sense need not provide conditions o n intending that yield a functional theory sufficient for explicating i t . There is a m o r a l here regarding deviant causal chains ( c f Peacocke, 1979)- Causal conditions that are supplied b y c o m m o n sense i n explication, say, o f m e m o r y or intentional action, etc., may be satisfied in odd ways, so that the concepts to be explicated are not satisfied. Deviant causal chains threaten the sufficiency o f causal explications. N o w the earlier suggestion about the relation between commonsense constraints and theoretical psychological theories removes this general threat to the conservative explication o f mental and semantic concepts w i t h i n a physicahst framework. That c o m m o n sense does not supply: everything is compatible w i t h non-ehminative explication.

4.10

T H E

CLASS

OF

B E L I E F - D E S I R E

THEORIES

The propositional variables that prefix conditions hke the L-constraints are so far unrestricted; they range over the infinite set o f all propositions. B u t on t w o plausible assumptions, that leads to an unacceptable result. Bel does not have the n u l l set i n its range; but then. 93

the behef-desire theory is true o f 2 o n l y i f for epery proposition p is a first-order state that belongs to Bel[p, z, t], that has the function ' role o f the behef that p. O n its o w n this does n o t entail infinitely mat,^' first-order states o f z, for Bel m i g h t map infinitely many propositio ^ onto finitely many states; finitely many first-order states could reaH infinitely m a n y functional states. B u t i t w o u l d be amazing i f that wcre so, for any pair o f distinct beliefs o f z w o u l d i n fact always have different potential effects and causes; different first-order states would need to be i n v o l v e d . Indeed, one w a y for a creature to realize the belief-desire theory w o u l d be via first-order states w h i c h are structurally isomorphic to propositions and w h i c h then constitute something like the "language o f t h o u g h t " that Fodor (1975), H a r m a n (1973), and Field (1978) have argued for. B u t then i f the theory quantifies over infinitely m a n y propositions, any creature w i t h behefs has infinitely many possible first-order states w i t h the relevant causal roles. This means, o f course, not infinitely m a n y state-tokens, but infinitely many structurally possible states. B u t that, I take i t , is not o n , given the finite nature o f the h u m a n central nervous system. Another reason for not a l l o w i n g unrestricted generalization over propositions is conceptual gaps - z's lacking a certain concept, o f a quark, say. There may not be possible states o f 2's neurophysiology that are n o w uniquely suited as the physical basis for that concept, w a i t i n g to be triggered by a suitable learning process. I t seems more likely that w h a t neurophysiological state o f z has the functional role o f the belief that p depends upon h o w z acquired its concepts; i f p contains a concept not already acquired by 2, there may be no possible state-type o f 2's w h i c h n o w has the second-order property o f being the belief that p. A n analogous p o i n t applies i£ p ranges over singular propositions w h i c h contain individuals as constituents. I f H a r r y has had no cognitive contact w i t h U l u g h Beg, no state o f Harry's is set aside b y nature to be Harry's internal name o f U l u g h B e g ; w h a t state has that role w o u l d depend o n h o w he achieved cognitive rapport w i t h U l u g h Beg.

concepts b y predication and iterations o f truth-functional from ' ficational operations, up to a certain level o f complexity. Thus j i i d qii^"^^j^^ ^ behef-desire theory o f 2 to have propositional variables '"^ed to a finite propositional field. O r , the commonsense belief— restricte^^^^^ prefixed b y an existential quantifier over '''^^"^ositional fields. For any such field, Bel then has o n l y finitely m a n y f st-order states as values. c npose a certain person 2 lacks one, isolated, recognitional ability is in the scope o f ' O ' i n the perceptual input condition. Since that ^' dition is part o f the very fabric o f the belief-desire theory, i t w o u l d then follow that 2 does not realize the theory and hence has no beliefs nd desires. B u t this has more bark than bite, for a very similar set o f functional conditions 15 true o f 2. Let us describe the general features of the "behef-desire t h e o r y " , abstracting f r o m the concepts i t requires of believers. It employs propositional indices, certain cognitive states that are subject t o L-constraints and, for a subset o f those indices, certain inclinational states that interact w i t h cognitive states to lead to certain conative states (wilhngs). N o w our imagined 2 reahzes a functional theory that resembles i n these and many more detailed ways the most complete possible version. A l t h o u g h , i n a certain strict sense, the ' B e l ' we apply to 2 and the one apphed to the fully i n f o r m e d believer is ambiguous, our ordinary 'beheves', applied interpersonally, reflects the shared features. There is no reason not to extend the point to members o f radically different cultures — t o Kalahari bushmen or members o f the Golden Horde. Functional theories that have those structural features were realized b y t h e m , and i t is thus that they had beliefs and desires, w h i l e lacking our recognitional abilities and M-constraints, although doubtless not our m i n i m a l logic.

B u t our commonsense belief—desire theory does not seem to preclude that, for each person, o n l y finitely many propositions are w i t h i n that person's current cognitive and inchnational range, namely those whose constituent concepts and logical operations the person has mastered, up to a certain level of complexity. Such a set o f propositions, a propositional field, w i l l satisfy certain conditions: a sufficient variety o f constituent concepts, and membership by all propositions resulting

W e can envisage theories that we w o u l d w a n t to count as about propositional attitudes b u t do not have the sort o f structure I have specified. A n y variant o f decision theory w o u l d be such, one w h i c h included m a x i m i z i n g expected u t i h t y or a m i n i m a x principle. The indexical account o f the role o f propositions i n the behef-desire theory can be carried over t o such a theory. Instead o f defining a distinct systematic role for each belief and desire tout court, the theory w o u l d generate a unique systematic role, for each proposition and each degree o f belief, for the belief in that proposition to that degree. N o such theory could generate a bona fide theory o f propositional attitudes unless i t included an input law w i t h a sufficient range o f observational concepts

94

95

to insure the requisite uniqueness o f systematic role. I n partici, theory whose o n l y specification o f i n p u t is i n terms o f senso * stimulation is doomed to failure, unless i t attributes to the believ;?' rather elaborate theories (M-constraints) that relate sensory stimulation t o propositions about the w o r l d . Short o f requiring such an epistctj, ological extravaganza, there is, I t h i n k , no hope o f explaining propositional content i n terms o f a sensational theory o f i n p ^ combined w i t h a decision theory. Ascription o f content is at that! m i d d l i n g level o f functional organization w h i c h presupposes quite contingent and culture-bound,^ even i f interpersonally ascribable recognitional abilities. ' A general problem for functionalism has been raised by N e d B l o c k (1978). Specifying input and output, in commonsense conditions, is bound to be too restricted to human features, too chauvinistic, as B l o c k puts it. In other species, pain may not be caused by factors that cause human pain; and evidently pain and other mental states will cause different bodily movements in other species than in us. Hence commonsense functional theories w i l l not capture the fully general concepts o f non-species-specific mental states. As regards input, let us keep in mind the distinction between those general features o f functional theories that generate the most general concept of " belief" and the specific theory that defines beliefs with the contents that we are used to ascribing. O u r normal ascription o f specific content is chauvinistic, in that it presupposes (among other things) a certain range o f recognitional abilities, that is, o f relevant input conditions. As regards output, the condition goes from 'x intends to do / I ' to ' x does A', but there is no chauvinistic restriction to human actions A. Notice also that the problem o f output for states like pain w o u l d vanish if they were treated as causes of desires, and only indirectly then as causes o f actions.

5

Beliefs about particulars need an account i n functional terms o f certain beliefs that are about ticulars " i n d e x i c a l l y " and not under descriptions o f t h e m , for example beliefs about perceived particulars, tensed beliefs, and beliefs bout oneself There is a topic i n the literature closely related to this topic, w h i c h needs at the outset to be f i r m l y distinguished f r o m i t . I mean that o(de reoccurrences o f singular terms ( ' I ' , ' n o w ' , ' t h a t ' , ' h e ' , 'the F', 'N') i n that-clauses i n English, what they contribute to belief-ascriptions. I am not denying that i t is an i m p o r t a n t topic: i f m y theory o f beliefs about particulars fitted no good theory o f de re terms in that-clauses, there w o u l d be a p r o b l e m . Suppose all behefs about

particulars

were about

them

under

descriptions; one still could ascribe such a behef (/e re, w i t h o u t using a description, simply b y saying what or w h o i t is about. So i f y o u believe that the man standing i n the corner is a botanist, I can say y o u beheve Watson is a botanist, either w i t h no i m p l i c a t i o n about your conception o f Watson, or w i t h an i m p l i c a t i o n o n l y about the general sort o f description i n v o l v e d . In beliefs about perceived particulars, I shall suggest, there is something w h i c h , w h i l e not a description, functions like one, as the mode of presentation o f y under w h i c h x believes y to be F. I t corresponds to a functional fact about the belief B u t , as on the description theory, It is not intended that whenever a belief about a perceived particular IS ascribed, a specific mode o f presentation is mentioned. Y o u may mention the object, etc., the belief is about, w i t h o u t any implication about the mode o f presentation, or w i t h indication o n l y o f its general type. The topic I am pursuing is then not h o w the singular terms we use i n ascribing beliefs constrain the modes o f presentation i n v o l v e d . • am concerned w i t h what figures i n the content o f such beliefs, and not h o w our ascriptions determine, indirectly, classes o f such c o n tents. The theory o f the latter w h i c h I find most attractive is Stephen Schiffer's (1977, 1978) (adjusted to include, as modes o f presentation.

96

97

not just descriptions, as on his account, but also the non-descriptive ones to be introduced). The idea is this. Take a sentence S of the form believes that . . . y . . . where the singular term y occurs de re (i.e. refers to a real object the belief is about). 5 implicitly refers to types of modes of presentation: ' there is a mode of presentation m, of the contextually relevant type, which determines y, and x beheves that .. .m...'. With that interpretation of de re singular terms in that-clauses in mind, let us turn to modes of presentation and the aspects of functional role they correspond to. 5.1

BELIEFS

ABOUT

PERCEIVED

PARTICULARS

The form 'Ft' appears in the L-constraints on quantification and identity; the question is what it is for a belief to have this form — to be about a particular. There are really two questions: how to represent such beliefs antecedently to a functional explication, and how then to explicate their content in terms of functional role. The clearest theory of beliefs about particulars has been the description theory, that beliefs are about particular things outside ourselves by containing definite descriptions true of those things. So my belief that Machu Picchu is in Peru is about those buildings by containing a description — say, ' the Andean city called ' Machu Picchu''; and my belief that that vase contains carnations is about the vase because my behef contains a description like ' the vase which is causing such and such sensations in me'. This theory, the hete noire of the "new theory of reference ", must, to be interesting, allow (for reasons to be given) that some beliefs, about oneself or the present time or one's mental states, are about them by virtue of some non-descriptive relation to them. My concern in this section is beliefs about perceived particulars. They include beliefs one forms about things while perceiving them, and a lot more. I look out of the window, spot a goat, and form the behef that the goat is capering. If I later discover that my chrysanthemums are no more, I may infer that the goat has eaten them. The latter belief is about the goat by virtue of my earlier perception, and not formed by perceiving the goat eat the chrysanthemums. So beliefs about perceivfed particulars are about them by virtue of perceptions that may or may not be causes of the rest of the belief's content. My reason for singling out these beliefs is that they are the least plausible candidates for the description theory of beliefs about things outside oneself for reasons I turn to later. As regards other beliefs about particulars, say. 98

^jgining proper names, I am more inclined towards the description '•''^^ Standard alternative to the description theory of sentences or rartces of the form 'Ft' is that they express singular propositions, " positions with built-in particulars. So ' that cat is fat' would express P jQposition that has as one constituent the property fatness and as ^neither the referent, say a certain cat. Transferred to beliefs about rticulars, this theory would treat them as relations to such singular Propositions, which may be represented as '[Fa]', or, in general, ' [FOi' • • •' °n] '• That is not just an intension, a function from possible vvorlds to truth-values; it has structure — think of it as a pair of a simple or complex entity F and the n-tuple not B^p'. What could nial^' the belief that a is F and the belief that a is not F doxasticall incompatible? The answer has been that they are about a by virtue of the same perception of a; that perceptual event-token is built mto prepositional content so that it satisfies the L-constraints. This is not to say that when we ascribe such beliefs we always refer direct'y (Q particular perceptual events, for we may be implicitly existentially quantifying over them. So we have a way to represent the content of these beliefs antecedently to their functional explication; now the question is how to specify unique functional roles for them. The strategy is this: first to describe the functional role of a certain subclass of beliefs about perceived particulars — namely, the observational ones; and secondly to describe the functional roles of the rest of them in terms of a certain functional interaction with the observational ones. The L-constraints do not on their own determine unique functional roles for beliefs; perceptual input conditions are also needed, that is, something that correlates aspects of the content of observational beliefs with aspects of the perceptual circumstances that normally give rise to them. That evidently is necessary in capturing the functional role of beliefs of the form [Fa^,.. .,a„/ei,.. .,e„]. Earlier I pointed out that it is not required that every observational belief-to^en should arise because of perceptual circumstances of the relevant type; I may believe there is an apple in the bowl because I was told so and not because I observed it. The point is that the belief has that content because it is a state that would arise were I perceptually related to that fact in the right way. Something like that is true of beliefs of the form [Faj,..., ct„/ei,.. .,€„]. But what functional conditions on those behefs do the perceptual event-tokens ei,...,e„ correspond to? They arc different from the input conditions corresponding to F, for the presence of perceptual event-tokens does not mean that such beliefs would occur i f somehow, those event-tokens reoccurred in conjunction with the right observational fact Fxi,...,x„. Take the goat again. My belief 102

hat he ate my chrysanthemums (where the belief did not arise « [jter) is not to be characterized as a state that would occur on ^ JO recur and be relevantly involved in my seeing the goat ^ ^^flowers; e is over and done with. It is in the index of that belief '''wather, because that behefs (future) tokens would have it, that (^^^j) event, among their causes. (P** as regards the predicative part of the belief i.e. Fxj,.. .,x„, the ' tual input conditions of 4.4 apply. A behef has the form fFoT, • • •' • specific Fand variable Q j , . . . , a„, C i , . . . , e„) J,„Ji/we have: For all ai,...,oc„, i f it were the case that F a i , . . . , a „ , and 0{z [P'^i'- • •' '•ben> for some e^,.. .,e„, it would be the case that'z believes [Fa^,.. .,a„/e^,.. .,e„]. (The 'etc.' stands for the conditions ofp-attentiveness andp-openness; ^lIlder

cf pp- 68-69.)

As regards the causal condition relevant to e j , . . . , e„, it is notoriously difficult to find a commonsense causal condition that is non-circularly sufficient in such cases; but the handy methodology of 4.6 once again comes to the rescue. I f a theoretical, non-commonsense, functional theory of z (which has the commonsense constraints as its core, and defines Bel, etc.) describes a causal relation, or class thereof, between perceptual events and belief-tokens, which intuitively are about their objects via those perceptual events, then those causal relations figure in the conservative explication of Bel[Fai,..., a „ / e i , . . . , e„], z,t]. Naturally those relations will amount to z's remembering the objects Ci.••.,«„.

The theory could seem to be a theory of reference, of the relation by virtue of which those beliefs are about a^,.. .,a„. But what has been said about the functional role of the behef that [Fa/e] leaves a quite idle, the functional role being determined simply by F and e. But on normal perceptions e, several particulars are perceived. So a refinement IS needed. When I perceive and a2 together, and form distinct beliefs about them, the perceptual input has suitably distinct aspects, each counting as an "event", a triple consisting of an n-tuple of particulars (e.g. a, and my eye), a relation, and a time. Thus a, is in part determined by building it into e,, and e, is individuated to accommodate pre-explicative individuation of such beliefs. But that is lot sufficient to uniquely determine reference; thus it does not select the goat rather than the goat's visible surface as part of e. As regards Junctional role, this indeterminacy, I suggest, must stand. What picks 103

out the goat rather than its surface, as the referent, hes in the tli of truth conditions (8.3) and reference (9.9); given the satisfa^ conditions of' F', the selection of the goat rather than its surface foil'^"^ from the general maximizing of reliability for such beliefs. What about non-observational beliefs about perceived particular ) What about their functional roles gives them subjects in common ^ observational beliefs? A certain L-constraint plays a crucial role here namely, 'BFt and BGf =>Not B ^ ^xFx 8c Gx': i f one believes r ' F and believes (is G one does not beheve that nothing is both F anj G. Its importance lies in securing that beliefs whose indices share j subject index themselves share a subject; in particular, it implies that a non-observational behef [Ga/e] interacts with the observational belief [Fa/e] in a way that beliefs which do not share a subject with the latter belief do not. Let us now turn to the description theory's account of beliefs about perceived particulars. Often my only individuating conceptions of perceived objects are of their being related uniquely to me, or to things I am directly aware of (The question of beliefs about oneself is taken up in 5.3.) Moreover those conceptions often involve psychological and not spatiotemporal relations. Thus suppose I attend to a quite unprepossessing star, randomly chosen; there are no spatial relations, of which I have suitable information, that relate it and no other star to me. One description that involves a psychological relation might be this: 'the star I am attending to'. That does not yield a persuasive description theory: for what makes it true of an object that I am attending to it? A description account requires yet a further psychological relation, in a new description; while any other account prompts an unanswerable reply: if attending to an object doesn't have to be via some description of it, why does a belief about it have to be so ? The only suitable psychological relation, I think, is something like this: ' the object that is causing that aspect of my sensations (sense data, sensory field)'. How is that aspect to be captured? Not in purely quahtative terms, for wouldn't it sometimes be necessary to refer, say, to the left side of one's visual field? That can't be done purely qualitatively. Descriptive theorists thus, with Russell, have to allow aspects of sensations to be objects of beliefs directly, not under descriptions. But i f beliefs can non-descriptively be about sense data, why not non-descriptively about physical objects? Metaphors of "directness" will not contribute much to the functional theory of content. 104

there is another problem about description theory: it is B^' jhat we form beliefs about details of our sensations when jpip'*^"^^ apparently exclusive devotion to external goings on. One ^''^'^ ders how we could be as successful as we are in registering the scene i f our thoughts were as full as that description theory passing '^'^'perhaps a description theorist need not be dissuaded, for the model objection implicitly suggests of beliefs about the sensory manifold ' be the wrong one. We can of course abstract from the external "^rrelates of our experiences and be psychophenomenological; but erhaps that is not what the description theorist needs. Another conception of our sensations is as appearances, whose psychological status is not being registered. If a child puts pressure on an eyeball and sees a blurred mess, he can be quite aware of how it looks, without conceptualizing it either in psychological terms, or as features of the external scene. That awareness of how it looks is perhaps a constant ingredient of perception and may be all the description theorist needs. But still the theory implies more structure in the content of beliefs than the perceptual index theory, without apparently any persuasive reasons for doing so; the perceptual index theory satisfies the Fregean requirement, and is less intellectualist. Although inclined against the description theory for beliefs about perceived particulars, I find it more plausible for other beliefs; for example, those naturally expressed by using names, where metalinguistic descriptions seem generally available. Also, some behefs are about particulars under complex dossiers, which may be mixed bags of perceptual indices and descriptions, and their functional roles can be viewed as amalgams of the functional roles associated with the dossier's components. The point against the description theory of beliefs about perceived particulars has to do with psychological plausibility; it requires too much content. But the current anti-description-theory animus rests on less mundane grounds; commenting on them may serve to clarify what I take to be the real issue, by contrast. First there are singular terms in modal statements. Many have been •m pressed by the fact that Aristotle could have turned out differently. Could have lacked all his known biographical properties. This 15 an objection to a description theory that takes names in modal contexts to be equivalent to definite descriptions that have narrower scope than the modality. It is of course false that ' Aristotle might not have been 105

a philosopher' is equivalent to ' Possibly: the philosopher who did and such was no philosopher'. But what an absurd theory; evid proper names normally occur with wider scope than the modalitig^^'^ whose syntactic contexts they occur.^ The description theorist sho^ ^ hold that proper names differ from overt descriptions in this respe^f where often overt descriptions in modal contexts are scope-ambiguQ^ proper names are most naturally read (usually) as having widest scop ^! that is simply part of a sensible description theory. The rigid designator theory of names is naturally associated with particularist and "haecceitist" metaphysics: ordinary objects are ontologically basic, not constructions out of properties, and inhabit many possible worlds rather than merely having qualitative counterparts there. This metaphysical view I subscribe to, but it is quite compatible with the description theory. It has been a mistake, beginning with Russell and continuing through Kripke, to assimilate semantic and psychological questions, on which the description theory should turn to metaphysical questions about whether objects are ontologically basic, or constructions out of properties, etc. Russell associated the description theory with the theory that objects are logical constructions, as I think did Strawson (1959) on the negative side, in his two-pronged semantic and metaphysical critique of Russell. In 'Naming and Necessity', Kripke associates his anti-description semantic theory with the metaphysical view that modal facts about objects are not about bundles of properties, that in assessing a modal statement containing a name, we carry the object and not its contingent real world properties through other possible worlds. But the two theories are quite distinct. Secondly, metaphysics has perhaps come into the picture with regard to the nature of reference rather than referents. Some (not Kripke) may have been moved by the following line. " On the description theory, the relation between belief and object is via the instantiation by the object of a description. But that relation is platonic and non-natural, not one that belongs in a naturahstic account of mind-world relations. But if we make the relation causal we naturalize it." This is confused. (I have encountered mixed opinions on whether it is a straw man.) Some beliefs involve descriptions: does that mean that there is no naturahstic account ^ I do not mean to beg questions about whether proper names have scope. O n the rigid designator theory they have no scope, since that w o u l d imply that they must be evaluated at a certain stage in the evaluation o f the truth conditions o f the whole in w h i c h they occur; whereas on the rigid designator theory they can be evaluated at any stage in the overall evaluation. M y point can be put in a w a y that everyone can agree to: they are existentially generalizable from outside, and so if they are held to be equivalent to expressions that do have scope they must be assigned widest scope.

106

1 tion to the objects that satisfy the descriptions? That is J their re jgnying that there is a naturalistic theory of a behefs tanta"!"""*^ f an object, for the instantiation of a description is just being ^'^"^ jicates' being true uniquely of the object. The upshot would certai" P^j^-j^ have a naturalistic theory of reference, we would be that' account of satisfaction, which makes the former rather ,,,ve no sucn ,„,nterestmg. 5.2

THE TENSES

OF

BELIEFS

theory of content evidently must account for tenses. The literature ^ 'yerificationism and on behaviorism, both pro and con, has quite °"htly stressed the importance of the question what makes certain bliefs past-tcmed. As with all questions about content on my theory, this involves both functional role and truth conditions, but here I shall discuss just functional role - of present and past tenses (the account of past-tensed beliefs yielding an account of future-tensed behefs). If my general thesis about content and functional role is correct, the "belief-desire theory" must assign to the belief that p was the case a unique systematic role. A present-tensed behef that p can be represented initially as [Now p], where p captures the rest of the propositional index including perceptual events - that is, a belief like [Now ¥0.^^0.2/e^^, 6 2 ] is subsumed under the form [Nowp\. What entity this denotes can be discussed when we have the functional role. I shall take other tensed beliefs to employ quantification over times, a relation 'earher-than', and the indexical 'now'. The belief that p was the case is the behef that some occurrence of p is earlier than now; the belief that p will be the case is the behef that now is earlier than some occurrence of p. The present tense is the key to the past and future: let us then begin with its functional role. The natural starting place is present-tensed observational beliefs, and their input conditions, and a first approximation is this: i f f is true at t and 0{z,p, t) etc., then z beheves [Now p] at t. But that is not adequate for assigning a unique interpersonally ascribable systematic role to present-tensed beliefs; it is true of them but also of others. Suppose I see Harold falling off his horse, and thereby believe that he is now falling. At the next moment I cease to believe tbat Harold is now falling: he has fallen. But surely I have also acquired another belief that I do not cease to believe once Harold has fallen. A 107

rather dull example is that Harold is not among the unfallen. But tK is a more interesting possibility. It seems that beliefs can be ab "^^' particular times in much the same way as they are about percei*^^' particulars, indexically and not under descriptions. This is esper-^i^^ interesting in connection with the present tense; for it would mean th y a belief's being indexically about the present time does not eo / account for its being present-tensed. So let us consider this oi^^" indexical reference to times. From beliefs that are indexically about perceived particulars it is small step to beliefs that are indexically about perceived events. The sani considerations against the description theory apply: the requirei^ psychological reflexiveness is equally implausible. Now, what is an event? I doubt there are pressing reasons to take events to be anything more basic than triples of particulars (including spatial positions) properties and times. ^ But reference to the particulars and property involved often will not individuate an event, and so, it seems, indexical reference to events presupposes indexical reference to times. A belief's indexical reference to a time can be captured by the device suggested for indexing beliefs about perceived particulars - namely, a certain perceptual event. So the belief that the object a, as seen on e, was F at time t, as presented on perception e', is represented thus: [Fnot B [ ~ 3xFx & Gx]'. °(3[Now p] is not observational, then, for any observational q, if z believes Now p and z believes Now q, then z doesn't believe that p and are never true together. No aspect of propositional content other than the present tense is interpersonally ascribable as interacting with present-tensed observational beliefs thus.^ I turn to the question of what entity ' [Now p]' stands for. If only the functional role of present-tensed beliefs were the issue, all we would need is the information contained in p together with something to mark the functional role of the present tense. Then building the present time (into the content of ^'s belief at t that Now p would be redundant. Now, p is (apart from perceptual indices) a propositional function with a time variable. So we could take present-tensed beliefs to be relations, not to full propositions (singular or general), but to propositional functions, i f all that mattered were functional role; that the belief is present-tensed would then be marked by its object being such an entity. (I must emphasize that that would be so only by virtue of the relevant formulation of the conditions of the belief-desire theory, and therefore of the definition of'Bel'.) There would not, on that way of marking the present tense, be two belief-relations, one for tensed beliefs and one for tenseless beliefs.^ There would be one relation, defined by its role in the behef-desire theory; some of its relata would be propositional functions and some full propositions. But the content of a belief is a matter not just of functional role but ' O n e quahfication: i f ]) is logically true, and simple, the belief that p satisfies the constraint regardless o f its tense. There are various ways o f eliminating this, but it is unimportant: what does it matter i f every logically true belief has the present tense? Other tenses can be embedded. * This is a mistake I made elsewhere (1976a) with regard to self-ascriptive beliefs. In suggesting that their object is a certain sort o f one-place propositional function (see the following section) I took that to mean that there is a special self-ascriptive belief-relation.

109

also truth conditions. Naturally, z's belief at t that Now-p is true if only if it is the case that p at and so we can build ( into its inj"^'^ [p at (], which is a singular proposition. (How truth conditions c into the picture is the topic of 7.3 and 7.4.) ^ Might [p at t] exhaust the content of present-tensed beliefs? isj^^. one accepts that beliefs can indexically be about times under perceptu'[ indices without being present-tensed. Otherwise, that singular prop^ sition would be adequate, for the only beliefs whose indices had th~ form [p at t] would be present-tensed beliefs, and thus' [Now p]' coui^ variably denote such singular propositions. But, I take it, some belief are indexically about times via perceptual events; thus a belief's having the present time in its content is not sufficient for it to be present-tensed So we should add a futher index that marks the present tense; why not the word 'now' itself? ' [ N o w p ] ' can then variably denote Not B(a, ~ 3xFx & Gx). Beliefs about 115

one's mental states, about one's past, etc., are beliefs about one virtue of this functional relation to observational self-ascrintivp -ascnptive i'^, b c ky, (Once again, this remark must be understood in theehght light of of,.,;^f» There are familiar questions about the self that this account do answer. It does not say what a person is, which may appear odd it purports to account for the content of self-ascriptive beliefs; surgj^^ it seems, that must involve the concept of a person. But it is not cl ^' that a self-ascriber must believe himself to be any certain sort of thj^^"^ (It is not even clear that a self-ascriber must attribute thoughts to hiniself which would mean a rather limited sort of self-ascription; but it wonlj provide a basis on which to explicate the more familiar sort.) TWQ self-ascribers need not agree on the rational grounds for self-ascription Suppose two cobblers each have memories of former princely existences where the rest of the story goes as usual, with interesting causal connections established between the earlier princes and the later cobblers, etc. Now suppose that both are apprised of the facts, exactly analogous in the two cases, and that one self-ascribes the quasiremembered princely facts and the other does not. (It is hard to say who pulls off the more difficult feat.) They then don't agree on what a person is, but each has self-ascriptive beliefs. T

1

1

-

1

.

1

There is also the matter of truth conditions for is the same person as y'. That must be part of a general account of the truth conditions of self-ascriptive beliefs about the past; but it is not something we need in describing a systematic role for self-ascription. Indeed, the concept of self-ascription could be used in the analysis of'x is a person' without circularity. For reasons much like the ones about present-tensed beliefs, the behever ought to be built into the content of self-ascriptions. It is not that the use of B* would leave us without an exact way of giving their truth conditions, but simply that, if content in general encapsulates truth conditions as well as functional role, we may as well standardize here. Suppose, again, that singular propositions were not involved in the content of any other behefs. Then 'B(a, [Fcx])' would be a perfect way to represent self-ascription. Not better than B*, of course; in no sense would there be only one relation involved on the former and two on the latter. It simply would not matter whether we indexed the functional role as 'B(a, F)' or as 'B(a, [Fa])'; on one the single belief-relation would have in its extension some pairs of persons and one-place propositional functions, and on the latter some pairs of persons and singular propositions. But if singular propositions are part ii6

„t of some non-self-ascriptive beliefs (on the perceptual of f " rv^ we do need something more in the content of self,nde'' ^'''Sefs an arbitrary marker like the word ' I ' . Harold's belief ascriP"''^ is now driving this car as perceived under e is then captured at by

Bel[Harold, t, [Harold drives this car at f / ' I ' , e, now]]. t)Cl[i^

vstematic role has been described. "'*^s I earlier remarked, tensed beliefs can intuitively be individuated ccording to either functional role or truth conditions. They give different results. So if Harold and Herbert each believes he is getting bald, in a sense they believe the same thing; and if Herbert also beheves about Harold that he is getting bald, then in another sense they believe the same thing. Now the interesting thing is that Harold and Herbert cannot both have one behef with the same content in both senses; only Harold can have a belief that is both self-ascriptive and about Harold. Given how the functional role is established, that is quite unproblematic. If there were some single "proposition" the behef in which counted as Harold's self-ascriptive behef that he is getting bald, then there would be a mystery — something Harold could never communicate. But there is no such proposition. There are two ways of individuating Harold's belief hence two ways to give its "content", and each of these may separately apply to pairs of beliefs of Harold's and Herbert's. But they must be distinct pairs; no pair of beliefs of Harold's and Herbert's can be counted as the same in both dimensions. Self-ascription is as Httle a puzzle as self-shaving.

117

6

Objectively determinate beliefs and our knowledge of them 6.1

FUNCTIONAL

ROLE

AND THE DETERMINACY

OF CONTENT

Quine's celebrated thesis of the indeterminacy of translation (IQQQ chapter 2) has a natural extension to the ascription of beliefs. I f thg translation of Pierre's utterances into mine is always underdetermined by the facts, inevitably any attribution of beliefs to Pierre is also underdetermined. The foregoing account of belief content in terms of functional role prompts the question whether it is in conflict with the indeterminacy thesis, and also what light it may throw from its independent perspective. There are two levels to our discussion of indeterminacy. There b the question whether belief-ascriptions are underdetermined by the totality of physical fact and theory. Now on my explication of belief-ascriptions, the answer will be straightforwardly no. But there is a further possible indeterminacy - an arbitrariness in the assignment of propositions to functional roles, one that ought to be disconcerting to a believer in objective intentionality. But I shall argue in chapter 8 that when we consider natural constraints on assigning truth conditions that arbitrariness is largely eliminated. The upshot will be that there really is very httle indeterminacy. There is a certain facile argument for indeterminacy which gets a quick reply. For any stretch of x's behavior, there are always several incompatible ascriptions of beliefs and desires which would equally well explain that behavior. Therefore, the argument goes, it is arbitrary which of those behef and desire combinations we ascribe; hence indeterminacy. The gap in the argument is clear enough: it assumes the only constraints on ascribing attitudes are those germane to rationalizing behavior. But first, attitude-ascriptions make claims about functional organization that go beyond the mere systematizing o f behavior; non-behavioral evidence for how x is internally functionally organized can be further evidence for some ascriptions over others. 118

the argument ignores perceptual input connections, which Secon'^^^' counted in the evidence. Indeterminacy does not follow jlso a""^ 'fj-otn the existence of incompatible rationalizations of merely ^ \ l,ehavio^-^^.^^j^_^j^j.jptions underdetermined by physical fact and •> Consider the truth conditions my account gives to ' x believes theo''>'_^^ f' - viz. 'x is in (one of the states in) Bel[p, x, t\ A necessary '^^'^X on is that x's states are functionally organized in the way eo" '^^^j Now whether x has such a functional organization is ^redetermined by physical fact and theory, assuming the latter ^ies all relevant counterfactuals about first order states. It is also 'iTc^sary that x be in a first order state with a certain functional role, the one which the relevant psychological theory (cf 4.6) uniquely associates with the behef that p. But i f it is physically determinate whether x realizes such a functional system, it is physically determinate whether x is in a given state of that system. Propositions are assigned to underlying states as indices of functional roles. Even if that assignment is arbitrary, it is still physically determinate whether x is in a state with the functional role with which p is associated by the definition of Bel.^ Hence the truth or falsity of 'x is in Bel[p, X, t]' is determined by the totality of physical fact and theory. But, surely that other assignments of propositions would be equally adequate in indexing those functional roles somehow affects determinacy. Suppose we change the perceptual input condition systematically as follows. So far it has this form: If it were true that p and 0{z,p, t) etc., then it would be the case that z believes that p at t. The proposition p indexes the functional state whose input condition it figures in. That is eminently natural, but i f what matters is simply to index that functional role, other propositions are equally appropriate. So i f / is a one-one function, other than identity, from observational propositions to observational propositions, a new input condition could be: If it were true that p and 0{z,p, t) etc., then it would be the case that 2 "believes" thaty(/i) at t. Of course, when this is coupled with L-constraints, M-constraints, and the belief-desirf T. Naturally this would count as evidence only within a well-devej theory of Pierre's beliefs, desires and dispositions thereto. It has not been my purpose to argue that there is no epistemolo indeterminacy, no set of competing principles of ascription that en well accords with the perceptual-behavioral data. (There are of no two competing sets equally compatible with the total empirical da "^^^ Rather I have argued that certain well-known reasons for thinking th ere is are not persuasive. In the absence of more telling arguments th following attitude seems to recommend itself Here we have th*^ standard scheme, which asserts that all members of a certain group have a certain functional organization and that certain generalizations are true of their functional states. The evidence for that theory is impressive for it successfully systematizes a vast quantity of perceptual-behavioral facts, including linguistic behavior. So there is no special problem about how the principles in ( i ) - ( 6 ) can be highly contingent, as my theory requires, and still have a firm epistemological status.

I

That our interpretative task, with a newly encountered individual of our own society, is epistemologically on a par with interpreting a newly encountered Alpha-Centaurian, is a view that has achieved a certain standing. T h e commonsense dismissive reaction turns out after all to be correct; it took rather sophisticated bhnkers to have held it. Evidence that many members of a natural kind instantiate a certain functional system, and certain generalizations about its functional states (the contingent principles), will under certain conditions be evidence that a newly encountered member of that natural kind satisfies them. So we are not faced with the problem of radical interpretation whenever we acquire a new colleague; that would be as though the question of the new colleague's having a heart and kidneys were on a par with some Alpha-Centaurian's having them.

does not follow that the functional states with which behefs (') -figd have defining connections with overt language. The ''^^d"systeni may, without adverse epistemological repercussions, stand^f^^j .^^^ components; one asserts the existence of functional be ^^^^ j^pjg defining characterization contains nothing about language; *^*'^^'ther contains generalizations connecting, in a structured way, the "jstic behavior to the functional states. The conjunction of the two ''"^hat the Hnguistic evidence supports. T ) ihefine-grainedness of beliefs, which may suggest that they must linguistic because their fine-grained structure is like that of sentences, been captured independently of connections with language. For the f,ne-grainedness of beliefs is a matter of the fine-grainedness of their functional roles. In 4.5 I pointed out that for each logico-structural difference between propositions there is a difference in the functional roles of the indexed beliefs. (3) It is naturally of considerable importance why linguistic evidence seems so essential to a determinate assignment of fine-grained abstract beliefs; to that I now turn. 6.4

LINGUISTIC EVIDENCE AND

FINE-GRAINED

BELIEFS

That linguistic evidence has a special status in attitude-ascription has been registered. But since the functional roles of attitudes do not involve connections with overt language, this must appear somewhat puzzling, for that special evidential status could suggest that relations to language are partially constitutive of fine-grained belief. First let me make it clear just what the evidential status of language is here, and then show how the functional theory straighforwardly explains that.

Some remarks, finally, are in order about linguistic evidence for behefs, which are, I am claiming, non-hnguistic functional states. N o doubt i f hnguistic performance were subtracted from our total evidence, the remaining perceptual—behavioral facts would not provide a basis for ascribing anything like the full complement of abstract, fine-grained beliefs (Davidson's terms) we typically ascribe. It is not easy to imagine non-linguistic or non-communicative behavior, however complicated, that would determinately provide a basis for such ascription. What is the significance of this? There are several points to be made.

Some do not regard language as having in principle a special evidential role. Robert Stalnaker writes: " Imagine that we discovered living creatures — perhaps on some other planet — who did not communicate, but whose behavior was predictable, for the most part, on the hypothesis that they engaged in highly sophisticated theoretical deliberation." (Stalnaker, 1976, p. 82.) (I shall take ' communicate' here to cover all overt linguistic behavior, including working out theories On paper, etc.) I must agree that if we discovered such taciturn creatures Isuilding bridges, airplanes, and so on, we would indeed think they engage in 'theoretical deliberation'; we would, as Stalnaker says, be tempted to attribute to these creatures not just belief and desire analogues but beliefs and desires themselves". That the temptation

138

139

makes sense prima facie, supports a conception of attitudes inde of connections with overt language. (Stalnaker draws a ^'^'^^l conclusion: " we would n o t . . . have any reason to hypothesize t h ' ^ ^ ' ^ ^ ^ thought in a mental language or in any language at a l l " . I agree'^^^1 the language of thought hypothesis does not follow from a fine-Jra '''^M belief-desire theory (cf 9.1), but I do not see that this examnle cu"^^** that.) ^ "'^^s Stalnaker goes on to speak of "the behefs that our hypoth attributed to these creatures". This is quite a further step: it is one th^^'^ to say that their non-linguistic behavior, etc., is evidence that they hay^ theoretical beliefs; it is quite another claim that it supports ascripti,,,,^ of specific theoretical beliefs. (The difference is between: (a) we h.,v^ evidence that they have some beliefs or other and (b) there are certain beliefs i j , . . . , i„ such that we have evidence that they have 61,...,/, ) So the question is: can totally non-linguistic evidence support ascriptions of specific fine-grained beliefs ? Naturally it can, when the creatures belong to groups about whose behefs we already know a great deal. But we are talking about Alpha-Centaurians. It is doubtful that such technological feats would be explainable only by ascribing a certain specific theory to the Alpha-Centaurians; incompatible theories could dictate the same technological solution to a technical problem. (This differs from Quine's point and my discussion of it in 6.1; lingustic evidence \\as included there.) Given only the non-linguistic behavioral and environmental evidence (without further evidence about the AlphaCentaurians' internal organization), it is difficult to see how we could ascribe to them specific theoretical beliefs as a better explanation of their feats than any other possible ascription. But then we have a problem. I f the functional roles of beliefs involve no special connections with language, why does the ordinary evidence have to include something about it? This is important, for it might suggest that the earlier conclusion, that the Alpha-Centaurians have theoretical beliefs, was not justified, based on too weak an analogy with our language-using fellows when they display such technological skills. The problem is not that there is no conceivable non-linguistic evidence for ascribing fine-grained beliefs to new creatures; a theoretical psychology-cum-neurophysiology might uncover a certain fine-grained functional organization. T h e question is why, in the ordinary evidence, language has an ineliminable role - why beUef-ascription is radically 140

> with respect to the non-Hnguistic behavioral-perceptual ^.viden'^^' ^ ^ j . depends on these facts: (i) linguistic behavior is My ^"J^isomorphically to the functional structure of beliefs; (2) no stru'^^'f^j^jvior is similarly structured; (3) nothing else in ordinary other be ^^^.^ behavior warrants such specific hypotheses about func"''"'1 c

S

n

i

"

^

^

.

,

.

r

ards ordinary linguistic dispositions, the syntactic structure ot

tential utterances corresponds significantly to the structure of the ^"tions; and we explain that dispositional structure in terms of the '''^^Ture of underlying functional states. N o w I take it as evident that „ linguistic behavior of ours has similar structure, at least not in no non-""& way that shows up in our commonsense reasoning. * " w h y then can we not find evidence about internal functional structure in behavioral-perceptual facts that are not isomorphically structured? Quine has claimed that the general underdetermination of theory by observational evidence implies the indeterminacy of our ascriptions of theoretical beliefs to Pierre. That is correct provided we throw out our structured, linguistic, evidence. This will sound odd unless one keeps in mind that we are speaking of ascribing functional states and not truth conditions. The reason we have already seen; given a stretch of Alpha-Centaurian perceptual-behavioral data, however impressive their technological products, a sufficiently ingenious theoryconstructor among us might suggest several incompatible theories whose apphcation by the Alpha-Centaurians would lead to the same non-linguistic behavior. The general underdetermination of theory by observational evidence thus explains the indeterminacy of ascriptions of certain fine-grained beliefs with respect to ordinary non-linguistic evidence. There is nothing constitutive of belief-ascription in this; it is an empirical fact about the evidence normally available to us. That brain scientists could discover on the basis of non-ordinary evidence that the correct explanation of Alpha-Centaurians' abilities to build airplanes and bridges is certain theoretical dehberations - internal events i n volving certain functional states - shows that our ordinary non-linguistic evidence is simply deficient, crucially more rudimentary than the scientists' evidence (and than the hnguistic evidence). This no more needs philosophical explanation than our inability to deduce the nervous system of a newt from its observable behavior.

I';

141

7

Intentionality without intensions ^ 7.1

ATTITUDES

AND

PROPOSITIONS

T o understand the role of propositions is to be able to dispense with them; that is a guiding idea of this chapter. Beliefs so far are, in on^ sense, relations to propositions; but that must be treated cautiously f i , relation is abstract, and the proposition serves as an extrinsic indicator of functional role; that beliefs are relational does not imply that their relata are in the believer's natural environment. ' Beliefs a r e . . . ' admits o f rather different completions; the following are each appropriate: (a) beliefs are relations to propositions; (b) beliefs are functional states; (c) beliefs, for a given person and time, are physical states with certain functional roles. O n the last two, that beliefs are relations to propositions rather drops out of the picture. This is not unlike the case of temperatures, which are in one sense extrinsic relations to numbers, and in another intrinsic states of molecule collections. ' Temperature' could denote the complex states of affairs expressed by 'the temperature in degrees Celsius of % = « ' , or it could denote properties of x by virtue of which (given our indexing system) such relational states of affairs obtain. Another feature of temperature is relevant: we do not expect the underlying states of molecule collections themselves to be relational m a way that parallels the relation to numbers. N o underlying entity corresponds to the number, and no underlying relation corresponds to 'x has temperature « ' . Similarly, while my explication of belief does not preclude the underlying state from being, in some interesting sense, relational, it does not imply it either. One might discern the following structure in the development of the analytical theory of belief From a platonist conception o f belief, based on the notion of grasping a proposition, there emerged the more naturalistic idea of beliefs as relations to sentences, he\ie£-ascriptions being taken to describe the underlying relation indirectly, via the underlying sentence's meaning. So ^x believes that would mean: (3/) B*(x, / ) & R ( / , that-i). 142

y rnore basic than the relation expressed by the ascription - call ^^^^ ^ •' the sentence 5' of one's own language; R is some relation or 'means the same as' to a proposition, sentence, or like Then it became apparent that R, while not as overtly drawn utte""*"^^' g^tiquated psychology as " g r a s p i n g " a proposition, froin j^^^ jg still difficult to account for naturalistically.^ The next ne^et eliminate meaning and to concentrate on B * as the only '"Tbehef-relation. This preserves the intuition that belief is relational gyoiding propositions and meaning. But the cost is high, for it * '. full classical intentionahty, on which the ordinary structure "^f rationalizing action appears to depend, (cf 2.4). " Now the first-order states that realize the functional roles of beliefs d desires might be language-relative internal linguistic states, and erhaps that would interpret B * after all. But the relation x's being in state y does not vindicate the underlying-relation theory. For on this theory there is not just one relation; there are believing*, desiring*, intending*. N o doubt they can be reintroduced by definition, given my theory, but their philosophical point, as primitive terms of a basic sentential attitude theory, is then lost. B * is not drawn upon by the functional theory of chapter 4. The proposition does not give the meaning of a sentence which the speaker accepts (that is, in some independent sense of ' meaning'); it rather indicates a functional state, and that state's truth conditions. T h e sentential replacements of propositions, to be introduced in 7.2, are the ascriber's sentences; again, their function is not to translate the believer's sentences but to index functional roles, and to provide a peg on which to hang truth conditions. The ontological issue, of meanings as abstract entities, has not been the main point in scepticism about intentionality; the usual substitution of sentences or utterances merely exchanges propositions for the equally problematic concept of equivalence in meaning. But not only is eliminating propositions not sufficient for a naturalistic account, in a certain notable way it is, oddly enough, not necessary either. For the functional theory yields necessary and sufficient physical

conditions for

z's

I'eheving that p. Even though propositions are not in the domain of natural science, this relation to a proposition is determined entirely by ' T h e r e is the possibihty of expUcating that relation in terms of equivalence in functional folo. B u t that does not lend support to the B * pattern o f explication. As I pointed out in 2.3, "equivalence in functional r o l e " is too short; some functional system has to be specified to generate a useful equivalence relation. B u t to specify that (as I claim to have done in chapter 4) is to explicate belief-ascriptions without bringing in B * .

the physical facts. H o w can that be? Consider again the Uf, .



functional theory o f 6.2, on which all instances o f ' t h e belief t]|^*'"l« are treated as semantically simple. T o each sentence 'x beheves t h ' '

'

there corresponds a preposterously unmanageable sentence t h a t ^ j ' ' ' not quantify over propositions and is equivalent to 'x believes th apart from the latter's reference to a proposition. They agree exacti ^ x's functional state; relative to the ontology, each entails the oth^"* When a naturalistic account o f truth conditions is also provided t^ contribution o f propositions is fully naturalized, in a way that permit"^ their elimination. Ontological scruples can then be humored.

^

In 6.2 I argued that, given certain desiderata, indices cannot b ehminated. But if there were no alternative to propositions, the mi^^ theory o f intentionality would (as I see it) be only partially illuminating leaving an annoying opacity in an otherwise limpid physicalism. It that, and not the absence o f a naturahstic account o f the relation to the proposition, which motivates finding surrogates for propositions. The motivation is ontological, but depends upon a certain further concern. I f the concern were just to find necessary and sufficient conditions for

believes that..

the unwieldy theory of 6.3 would

do, and propositions could be viewed simply as convenient^cfiOMs. But we are after a theory o f intentional structure — that is, those general interrelations among beliefs and desires that depend upon logical form, the mastery o f which in large part constitutes our mastery of attitudes. T h e importance o f purifying

attitude-ascriptions

o f reference to

propositions lies in this: only then is it clear that intensionality is irrelevant to intentionality. Propositions have two overlapping functions: to index functional roles, and to encapsulate

truth and fulfillment

conditions. I say

" o v e r l a p p i n g " because the truth and fulfillment conditions of some attitudes figure in their functional roles. In the next section I discuss how to replace propositions in the functional theory, and defer the general question o f truth conditions to the subsequent section.

P

d beliefs, and among beliefs, desires and willings. O n the other uth conditions are invoked on the periphery

o f the functional

han'^' ^"^.^ ji^e input and output conditions. That exhausts the contheorY' ^ ^ ••

propositions to the functional theory. I f non-intensional

like sentences, and their non-intensional properties, can suitably

"^""ate propositions and their intensional properties, then, as regards onal role, intensionality will have been eliminated, while the full 1 ssical criteria o f individuating beliefs interpersonally will have been preserved. The adequacy conditions o f this elimination are as follows. First, the peW deintensionalized theory should be satisfied by 2- at r i f and only if the propositional theory is satisfied; furthermore, for each p,

'x

believes that p at t' ought to be satisfied by ^ at f just in case its replacement is thus satisfied. Let us call this conjunction the condition offunctional equivalence. Secondly we want something more, since the unwieldy theory of 6.2 satisfies that condition. That theory fails to treat 'x believes that p' as semantically complex, and therefore to preserve the generalizable structure in interactions among attitudes, and between them and perception and action. The new theory must parallel the old theory's ascriptions o f intentionality. Call equivalence in intentional

this the condition

of

structure.

The new theory substitutes sentences for propositions, but, unlike the semantic relation theories o f 2.3, no pragmatic semantic concepts arc here invoked — that is, nothing about what the sentence means in the ascriber's language or mouth. That is how the reconstruction escapes a circular interaction of theory o f meaning and rational psychology. No doubt the claim that this can be done is surprising. H o w can belief-ascriptions assert relations to the ascriber's sentences except insofar as the latter have meaning? It is necessary to take a somewhat abstract view, to escape from the grip o f the semantic-relation model (which is quite intuitive). In place o f something equally intuitive, let one's guide be the adequacy of the two stages o f explication: first to

7.2

PROPOSITIONS

ELIMINATED

FROM

THE

FUNCTIONAL

THEORY

T w o features of propositions are exploited in their employment a* indices, namely their interrelations o f logical structure and, for some propositions, their truth conditions. The former are deployed in the internal constraints, those that generate functional connections between 144

reconstruct

the

functional

theory

non-intensionally,

substituting

sentences for propositions subject to the requirements o f functional equivalence and equivalence in intentional structure, and secondly, to reconstruct truth conditions using only a non-intensional apparatus. I am supposing all propositions to be first-order expressible; for those 'who demur, this may be viewed as a simphfying strategy that leaves open the possibility of supplementing the L-constraints. Let L be 145

sider the internal constraints, say the L-constraints and the

quantificational English, adequate for expressing all relevant tions and propositional functions. A one-one mapping from * ^P^skg

''''^^V^re-willing condition. T h e new

tions to the sentences o f L that express them will be useful - so 1 ^'^^^ . . . through . suitable ' contrUs take taL L to be rid• o f- synonymy and. ambiguity,

beli^^"

and supplementation. O u r propositions are fme-grained, so t|!°'^

form*-

suit*''

theory

must contain a

for the sentences of L , to permit generalizing syntactic p^g^

r(Jx)...x...'^,

etc. W e replace internal constraints

^

structure imitates the syntactic structure o f the regimented L. {}-{^ ^'"^ forth I shall simply say ' propositions' and ' sentences' when it is ^j*"^' that propositional functions and open sentences are to be inciudgj?' Although' expresses' is a semantic term, to be avoided if uncxplai^jj think o f it as defined abstractly for L as in a formal semantics. But

T merly the constraints were expressed in terms of relations o f logical °ructure among propositions, they are now in terms o f syntactic

'

converse is more to the point: let 'p's expression in L = 5' capture th converse function from propositions to sentences, similarly abstractl defined. The sentence to be substituted for a proposition, in the new ! theory, is then its expression in L. (This notion is not part of the new theory; I use it to avoid tedious description of the new theory.)

relations among sentences. Xhe L-constraints impose conditions on what underlying states are candidates for being z\s at (: they must be counterfactually related as the L-constraints say such and such beliefs are related. N o w consider some instantiations o f the old and new L-constraints: Bel [[roses are red and violets are blue], Jones, t],

In the propositional theory, some indices contain singular proposi-

Bel[[not: roses are red], Jones, t],

tions, for example, in beliefs about perceived particulars, tensed beliefs

Bel'[['roses are red and violets are blue', Jones, t], and

and self ascriptive beliefs (cf chapter 5). In replacements for such indices in the new

theory, it will help to have initally taken singular

propositions to be pairs o f n-ary propositional functions and «-tuples of particulars; the former can then simply be replaced by open sentences. Let us call the new indices, with sentences or open sentences in place o f propositions or propositional functions, the sentential analogues o f the latter. N o w the sentential analogue of p is just its expression in L. In what follows there is no harm in thinking o f all old and new indices as simply propositions and sentences; nothing turns on the more complex structures o f indices. The intended effect o f the replacement is that the new belief^desirc theory defines new functions B e l ' , Des', Int' which map sentential indices, persons and times onto (sets of) underlying states, according to their functional roles. Naturally the new theory and the old theory should conjointly entail, whenever 5 is p's expression in L, that B e r [ 5 , z, t] — Bel[p, z, t]; similarly for Des' and Int'. I f that is so, and the new theory restricts itself to the ground covered by the old theory, the conditions o f functional equivalence and equivalence in intentional structure will be met.^ Let us see that there is no problem for the new theory's appropriately defining B e l ' , Des', and Int'. ^ T h e t w o theories are to be logically equivalent relative to their ontologies. T h a t is, Q captures the ontological c o m m i t m e n t to sentences and propositions and T is the old theory and T ' the n e w theory, then what ought to be true is: • ( Q ( T is true o f ^ at ( i f f T' is true of z at ())•

146

Bel'[['not: roses are r e d ' , Jones, t]. If a and P are first-order states that are candidates for being denoted by the first and the second, respectively, and the third and the fourth, respectively, then the old and new L-constraints impose exactly the same condition: if Jones were in a at f then Jones would not be in P at t. O f course it is the complex counterfactual position o f a and ^ in a vast network o f states that determines whether they can be those beliefs. But every counterfactual condition imposed on the belief that p and the behef that q by the old theory, is imposed by the new theory on the b e l i e f that s and the b e l i e f that r, when s and r are the expressions o(p and q in L. In general, the internal constraints of the two theories do not distinguish the behef that p from the b e l i e f that 5, i f s is the expression o( p in L — do not, that is, distinguish Be\\p, z, t] from Bel'[5, 2 , i\. That purely syntactic relations serve just as well in indexing internal functional connections among beliefs and desires as relations o f logical structure among propositions is not surprising. For what matters in capturing functional connections is their structure; among

i f certain relations

propositions mirror those connections, so do isomorphic

relations among sentences, on a certain one—one correspondence with propositions. This is the first step in deintensionalizing the belief-desire theory. As soon as belief-contents are seen as, in part, functional roles. 147

uction. A l l we need is a Tarski-type predicate T ' such that ' T ' i '

intensions become irrelevant to that extent. But to vindicate that we need to reconstruct the input and output conditions.

^'"V,

A n important point needs emphasis. L is not the language of believer, nor are the syntactic relations involved supposed to h'*^^

con''"^ is true' are strongly equivalent whenever 5 is p's expression in ^^'pccaiis^ L may be part o f the metalanguage, the homophonic T J. but we could always construct a non-homophonic predicate

among underlying states. L is our artifact; with it we index funct '^^^

* ' with the same result. This point will become clearer when the role

states via the ordinary syntax o f L. T o what extent the first-ortl

T is revealed more fully. " por observational beliefs with singular propositions in their indices, is not exactly ' t r u e ' that needs replication, but 'true o f or 'satisfies'.

structure o f the underlying states resembles that syntactic structure a further question. I f those states are in some sense linguistic, then speak o f syntax at that level is also appropriate (cf 9.1).

need a relation Sat such that, necessarily, the propositional function

The perceptual input conditions of chapter 4 rely on the concept of

is true o f n-tuple x i f and only if, where s is p's expression in L ,

truth, for they say that i f ;j is a true observational proposition and z

Sat(^. *^)- ''^gain a suitable Tarski-type Sat relation on L is easily

is related at ( to ;j by O , etc., z at ( believes that p. What role does

forthcoming.

truth play here ? There is a correlation between certain perceptible states

Let us now look at the new input conditions. Assuming there was

of affairs and certain functional states whereby one o f the defining

no problem originally in defining the class o f observation propositions

conditions o f the latter is that, i f a certain o f the former obtains

in terms o f their constituents (without using the concept of belief), there

etc., z moves into that functional state. A proposition is an economical

will be no problem in syntactically defining the class o f observational

index for an observational belief; via its truth conditions, it looks

sentential indices. In the original theory

outward to the perceptible state of affairs that is functionally tied to the belief, and, via its logical structure it looks inward to the other beliefs and desires whose functional relations to the belief are definitive. H o w can we duplicate the proposition's contribution to the input law? What counts as success is quite simple. I f the new theory is logically equivalent to the old (relative to their ontologies), there must

'0(z,

we have

the relation

p, t)', a necessary condition of which is that p be observational.

So similarly for observationality in the new ' 0 ' ( z , s, f)'.^ The new conditions are as follows; where formerly we had ' I f p is true and 0{z, p, t) etc., z is in Bel[p, z, t]', where 'p' ranges over general propositions, we n o w have ' I f T5 and 0 ' ( 2 , 5, t) etc., z is in Bel'[5, z, t]'.

be a (non-intensional) predicate T of sentences such that, i f s is p'i

When, in the former condition, 'p' ranges over singular propositions,

expression in L , then necessarily p is true i f and only i f Ts. Suppose

which have the form ' (senten^^^ analogues o f singular propositions). The meaning o f that inde'^'^'*^ sentence is given by the open sentence ' x beat y h a n d i l y ' o f the conten' language, together w i t h something that represents the contributions of 'she' and ' h i m ' ( c f Loar, 1976a, pp. 153—5, and Loar, 1976b, p. The details are not m y p o i n t here; but the " m e a n i n g " o f that senten w i l l contain at least the predicates ' m a l e ' , 'female' o f the content language. This " m e a n i n g " is an ordered n-tuple o f these and other appropriate elements. The language assigns such sentences satisfaction conditions rather than fully determinate t r u t h conditions; this again is b y virtue o f the special status o f T , and its related satisfaction predicate in connection w i t h the content language.

'language' means 'such and such set o f sentences', and n o t 'such and such association o f meanings w i t h sentences'; a theory o f meaning for the language (first sense) o f a population is i n part an identification o f the language (second sense) they speak or think i n . A theory o f meaning would also include z grammar for the language ( c f 9.9; 10.4). This abstract characterization is hardly enough; for a theory o f meaning must also say what empirical connections w i t h propositional attitudes are captured by a language, that is, by that function f r o m sentences to meanings. Thus a theory o f meaning for z's language o f thought associates sentences w i t h the functional roles o f their internalizations, w h i l e one for a pubhc language associates sentences w i t h classes o f complex communicative intentions that they are conventional devices for expressing.

Moods other than the indicative are represented i n " m e a n i n g s " bv something that represents propositional attitudes other than belief The meaning then is, say, an «-tuple, o f a marker o f m o o d , the relevant open or closed sentence o f the content language, and so on. W h a t it is about the pragmatics o f conventional language use these " m e a n i n g s " capture w i l l , I hope, become clearer i n 10.4.

John M c D o w e l l writes: " a theory o f meaning for a language should be a systematic portrayal o f the capacity possessed b y someone w h o understands i t [ m y i t a l i c s ] " ( M c D o w e l l , 1980). M y reaction to this is mixed. It fits quite w e l l a theory o f meaning for a language o f thought. For the non-cognitive understanding o f a sentence is just a matter o f that sentence's having acquired the right functional role; hence a theory that mapped sentences onto those functional roles w o u l d indeed be " a systematic portrayal o f the capacity possessed by someone w h o understands" those sentences. Also, the other non-cognitive model o f language learning (described on p. 215) on w h i c h i t consists i n acquiring a disposition to produce certain beliefs i n response to certain sounds, w o u l d also make that characterization apt. (That m o d e l i n fact resembles M c D o w e l l ' s account o f that nexus between sentence and behef w h i c h is constitutive o f meaning (1980, ibid.).)

Finally, i n f o r m a t i o n about a sentence s grammatical structure is not part o f its meaning in the current sense (for a different v i e w , see D . Lewis, 1972a). T h a t is represented i n the grammar o f the language, w h i c h says h o w sentential structure and constituents contribute to meaning and t r u t h conditions. As regards the i n t u i t i o n that a sentence's meaning is a holistic fact about i t - i.e. a matter o f its place i n the whole language, rather than its direct connection to something like a propositional attitude — that i n t u i t i o n has i n effect been accommodated on my account at a more basic level, namely ( i ) i n the determination o f a propositional attitude's content by its place i n the overall belief—desire functional organization, and (2) i n the i m p l i c i t holistic considerations inevitably appealed to i n determining t r u t h conditions for beliefs on the basis o f reliability.

9.8

WHAT

IS A

THEORY

OF

MEANING?

W h i l e an answer is close to the surface i n the foregoing, i t may be useful to present i t as such. I mean a theory o f meaning for a specific language, as spoken b y a certain population, or as thought in by a certain person — that is, a comprehensive systematic description o f the semantic properties o f the language, relative to that person or population. Here 222

I f " m e a n i n g " is a matter o f sentences being conventional devices for m a k i n g k n o w n communicative intentions, a theory o f meaning for a (public) language simply says w h i c h sentences are conventionally associated w i t h w h i c h intentions. " Understanding " that language then consists i n knowing o f each sentence what i t means, and so the correct theory o f meaning w i l l indeed capture what one understands. B u t , i n that sense, any theory o f a subject X is a theory o f what one knows in k n o w i n g the facts about X: a theory o f n i t r o g e n - f i x i n g bacteria is in effect a theory o f what one knows i f one knows the facts about nitrogen-fixing bacteria. If, i n other words, understanding consists i n "propositional k n o w l e d g e " o f meanings, i t is misleading to b r i n g i t into an account o f a theory o f meaning. So, rather than characterize theories o f meaning i n general as theories o f understanding, i t is better 223

to characterize sentences w i t h (depending o n , standing is not tantamount to

them as theories o f certain possible connection^ propositional attitudes. This covers the possibi] e.g., the " t h i n k i n g - i n - E n g l i s h " hypothesis) that u n j otiose, i.e., that the obtaining o f those connection understanding.

It w o u l d be w e l l , finally, t o clarify the connection between thcor o f meaning and theory o f t r u t h . Let us revert to the simplified langua / whose sentences are indicative, unambiguous and non-indexical j \ theory o f meaning for i t associates each sentence s w i t h something, say o f the f o r m 'x beheves t h a t . . . ' , or, to accommodate an undiscussed possibility ( M c D o w e l l , 1980), something o f the f o r m ' x says that. ' M c D o w e l l has suggested that a t r u t h theory has the function of associating sentences s w i t h statement-contents by virtue of i m p l y i n g things o f the f o r m ' 5 is true i f f . . . ' . The truth theory is a systematic w a y o f associating sentences w i t h what they say. B u t there is a p r o b l e m . I assume 'x says t h a t . . . ' is relational; ' t h a t . . . ' denotes something. B u t then w h a t captures the association of sentences w i t h contents is a function, and ' s is true i f f . . . ' does not express a function, or any relation; ' t r u e ' is a one-place predicate, and ' . . . ' is occupied by sentences and not names. Perhaps we should move up a linguistic level, so that what associates sentences w i t h contents does so by relating sentences 5 to equivalences o f the f o r m ' 5 is true i f f . . . ' , via the meta-truth-theoretic statement that for all 5 and p, s is relevantly associated w i t h saying (believing) that p just i n case T implies that .s is true i f f p. B u t that is rather baroque. W h a t is wanted is a function that maps each s o n t o the right p. While the meta-truth-theoretic statement does express such a function, to say it involves ' t r u e ' or any predicate o f sentences is beside the point. It is better simply to take a theory o f meaning for L to incorporate that function-in-extension, that language, w h i c h maps each s o n t o (in our simplified language) the sentence ' . . . ' w h i c h indexes the belief s expresses. W h a t then is the relation between a theory o f meaning and a theory o f truth? The answer has already i n effect been given. A correct theory o f meaning for P or z says w h i c h language, w h i c h function o f that k i n d , is the language o f z or P. A language determines its o w n truth conditions, not by mapping sentences o n t o t r u t h conditions, but by mapping sentences o n t o the sentential indices for behefs. Sentential indices, relative to Bel, determine t r u t h conditions: for, i n the sense explained i n 7.3, i t is trivial that the behef that 5 is true i f f T i , (although 224

by no means trivial that the t r u t h theory captures something interesting). 5Q a theory o f meaning for a language determines a theory o f truth for that language, and, i t goes w i t h o u t saying, is not exhausted b y i t .

9.9

THE

DETERMINACY

OF R E F E R E N C E

]\Iothing so far has been said about the meaning and reference o f terms or subsentential expressions, or the logical f o r m or semantic structure o f sentences. Indeed as regards referential semantics, those matters are largely underdetermined by considerations about truth conditions and sentence meaning; the references o f names and predicates are not determined by the t r u t h conditions o f the sentences they occur i n . That observation (which I shall explain in more detail) is the basis o f Quine's thesis o f the inscrutability of reference (1969). Quine's thesis, t h o u g h , is far stronger than the underdetermination o f references b y t r u t h conditions; for he also holds that there are no further aspects o f physical fact and theory that determine reference. I f that were true, then, given the physicalist premise, there w o u l d be no " f a c t o f the m a t t e r " about whether the Enghsh w o r d ' d o g ' refers to dogs, etc. Let us then take the claim that reference is inscrutable to be this: the references o f the terms o f L are underdetermined b y the conjunction o f physical fact and theory and any assignment o f t r u t h conditions to the sentences o f L. N o w I shall be arguing that the claim o f inscrutability is basically not true. The quahfication is due to a vagueness i n the main issue, about what is required for a discourse, w h i c h is not explicitly physicalist (e.g. the semantic) to be determined by physical fact and theory. Before turning t o reference, i t may be o f interest to note a level o f linguistic description that deserves t o be called semantic, involves the characterization o f subsentential expressions and o f sentential structure, is determinate w i t h respect t o properties o f linguistic meaning that have already been introduced, and is (then) not about reference. There is a distinction between functional semantics and referential semantics. For, a certain sentence's conventionally expressing a certain belief i n P, or having the internal functional role for z o f a certain belief does not presuppose anything about t r u t h conditions or reference, given h o w behefs are individuated b y functional role. N o w , their fine-grained functional roles are a physically determinate fact about beliefs ( c f 6.1). N o w i t is pretty clear, w i t h o u t going into the details, that i f the sentence 'les rats mangent le fromage' and 'les chats mangent le fromage' have the functional roles, i n the language o f thought o f French speakers, 225

o f the beliefs that rats eat cheese and that cats eat cheese, then we assign well-defined distinct functional properties t o the internal pK 'les rats' and 'les chats', properties derivative f r o m the functional rol^^ o f all sentences i n w h i c h those phrases occur. The form o f such functional semantics is r o u g h l y this: the functional " m e a n i n g " of-^ certain w o r d , syntactic structure, etc., E o f a given object language wih be captured simply b y that w o r d , syntactic structure, etc., o f our conten language c o m m o n to the indices o f all beliefs i n whose object language realizations or expressions E occurs. The functional-semantic grammar o f L is then a function f r o m words, structures, etc., o f L to words structures, etc., o f our content language, i n the ideal case o f perfect isomorphism, and, otherwise, some more roundabout characterization o f their contributions to functional role. Considerations o f reliabihty and simplicity motivate at most a restriction o f suitable t r u t h predicates for beliefs to those strongly equivalent t o T . B u t this leaves great leeway i n h o w a truth predicate assigns references and satisfaction conditions. Quine's w e l l k n o w n examples (in Q u i n e , 1969) illustrate the p o i n t w e l l . I f i n w r i t i n g out a t r u t h theory, we stipulate that 'x is a. r a b b i t ' is satisfied by y i f f y is a time-slice o f a rabbit, and make suitable adjustments in the satisfaction conditions o f all other predicates (e.g. so that 'x is one year o l d ' is satisfied by y i f f y is a time-slice o f something that is one year old), as w e l l as i n the satisfaction conditions for identity (so that ' x j = X 2 ' is satisfied by < y i , y j ) i f f yi and y j are time-slices o f the same four-dimensional rabbit), then given standard assignments to the logical connectives, etc., beliefs w i l l have t r u t h conditions, under that t r u t h theory, strongly equivalent t o those they have under T . There being t w o one-year-old rabbits behind that bush is the same thing as there being t w o time-slices o f one-year-old rabbits behind that time-slice o f a bush.

our standard reference scheme, ^Fx^,.. .,x^. Given a few refinejiients, the resultant t r u t h predicate w i l l be strongly equivalent to T : for example, the