Davidson, Truth and Meaning

TruthandMeaning DONALDDAVIDSON

It is conceded by most philosophers of language,and recently evenby somelinguists, that a satisfactory theory of meaning must give an account of how the meanings of sentences depend upon the meanings of words. Unless such an accountcould be supplied for a particular language, it is argued, therc would be no explaining the fact that we can learn the language: no explaining the fact that, on mastering a finite vocabulary and a finitely stated set of rules, we are prcparedto produce and to understand any of a potential infinitude of sentences. I do not disputethesevagueclaims, in which I sense morc than a kernel of tmth.l Instead I want to ask what it is for a theory to give an account of the kind adumbrated. One proposal is to begin by assigning some entity as meaning to eachword (or other sigaificant syntactical featurc) of the sentence;thus we might assignTheaetetusto '"Theaetetus"and the property of flying to "flies" in the sentence "Theaetetusflies." The problem then ariseshow the meaning of the sentenceis generatedfrom thesemeanings.Viewingconcatenationasa significant piece of syntax, we may assignto it the relation of participating in or instantiating; however, it is obvious that we have herc the start of an infinite regr€ss. Frege sought to avoid the regressby saying that the entities corrcsponding to prcdicates(for example)are 'unsaturated'or 'incomplete' in contrast to the entities that cor-

respond to names, but this doctrine seems to label a difficulty rather than solve it. The point will emerge if we think for a moment of complex singular terms, to which Frcge's theory applies along with sentences. Considerthe expression"the fatherofAnnette"; how does the meaning of the whole depend on the meaning of the parts? The answer would seemto be that the meaning of "the father of is such that when this expression is prefixed to a singular term the result refers to the father ofthe person to whom the singular term refers. What part is played, in this account,by the unsaturated or incomplete entity for which "the father of' stands?All we can think to say is that this entity 'yields' or 'gives' the father of r as value when the argument is x, or perhaps that this entity mapspeople onto their fathers. It may not be clear whether the entity for which "the father of is said to stand performs any genuine explanatory function as long as we stick to individual exprcssions;so think insteadof the infinite classof expressionsformed by writing "the father of" zero or more times in front of "Annette."It is easyto supplya theorythat tells, for an arbitrary one of these singular terms, what it rcfers to: if the term is 'Annette" it refers to Annette, while if the term is complex, consisting of "the father of" prcfixed to a singular term t, then it rcfers to the father of the personto whom t rcfers. It is obvious that no entitv corre-

From synthc*.t7 (t95711.to4-323. copyright@ rg6z by D. Reidelpublishingcompany,Dordrecht, Holland.Reprintedby permissionof the publisher. rl4

T R U T HA N D M E A N I N C spondingto "the father of" is, or needsto be, mentionedin statingthis theory. It would be inappropriateto complain that this little theory asesthe words "the father of" in giving the referenceof expressionscontaining those words. For the task was to give the meaningof all expressionsin a certaininfinite set on the basis of the meaningof the parts; it wasnot in the bargainalsoto give the meanings of the atomicparts.On the other hand,it is now evident that a satisfactorytheory of the meanings of complex expressionsmay not require entitiesasmeaningsof all the parts.It behooves us thento rephraseour demandon a satisfactory theoryof meaningso asnot to suggestthat individual words must havemeaningsat all, in any sensethat transcendsthe fact that they have a systematiceffect on the meaningsof the sentencesin which they occur. Actually, for the caseat handwe can do betterstill in statingthe criterion of success:what we wanted.and what we got, is a theory that entailseverysentenceof the form "t refers to x" where 'r'is replacedby a structuraldescription2of a singularterm, and '-r'is replacedby that term itself. Further,our theoryaccomplishesthis without appealto any semanticalconceptsbeyond the basic "refers to."Finally, the theoryclearly suggestsan effective procedurefor determining,for any singular termin its universe,what that term refersto. A theory with such evident merits deserves widerapplication.Thedeviceproposedby Frege to this end hasa brilliant simplicity: countpredicatesasa specialcaseoffunctional expressions, andsentencesas a specialcaseofcomplex singularterms.Now, however,a difficulty looms if we want to continue in our present(implicit) courseof identifying the meaningof a singular termwith its reference.The difficulty follows uponmaking two reasonableassumptions:that : logically equivalent singular terms have the samereference;and that a singular term does : notchangeits referenceif a containedsingular the samerefer; termis replacedby anotherwith .i:ence. But now supposethat 'R'and 'S'abbrevii ate any two sentencesalike in truth value. havethe satne i Thenthefollowing four sentences fireference: #i

n Ei rtl (2) $i

G, G G,

E

.f(x=.Y.R )=i(x=x)

il5 (3) i(x=x.S)=a(a=a) (4) S For (l) and(2) arelogically equivalent,as are (3) and (4), while (3) differs from (2) only in containing the singular term f(x=x.S)'where (2) contains 't(x=x.R)'and these refer to the samething if S and R are alike in truth value. Hence any two sentenceshave the samereferenceif they have the sametruth value.3And if the meaningof a sentenceis what it refersto, all sentencesalike in truth value must be synonymous-an intolerableresult. Apparently we must abandon the present approachas leading to a theory of meaning. This is the naturalpoint at which to turn for help to the distinction betweenmeaning and reference.The trouble,we are told, is that questions ofreference are, in general,settledby extralinguistic facts,questionsof meaningnot, and the facts can conflatethe referencesof expressions that are not synonymous.If we want a theory that gives the meaning(as distinct from reference) of eachsentence,we must start with the meaning (as distinct from reference) of the parts. Up to herewe havebeenfollowing in Frege's footsteps;thanksto him, the pathis well known and even well worn. But now, I would like to suggest, we have reached an impasse: the switch from referenceto meaning leads to no useful account of how the meanings of sentencesdependupon the meaningsof the words (or otherstructuralfeatures)thatcomposethem. Ask, for example,for the meaningof 'Theaetetus flies." A Fregeananswer might go something like this: given the meaningof "Theaetetus" as argument,the meaningof "flies" yields the meaningof "Theaetetusflies" as value.The vacuity of this answeris obvious.We wantedto know what themeaningof '"Theaetetus flies" is; it is no progressto be told that it is the meaning of '"Theaetetusflies." This much we knew before any theory was in sight. In the bogus accountjust given, talk of the structureof the sentenceandof the meaningsof wordswasidle, for it played no role in producing the given descriptionof the meaningof the sentence. The contrast here between a real and pretendedaccountwill be plainer still if we askfor a theory, analogousto the miniature theory of

ll6

referenceof singulartermsjustsketched,but different in dealingwith meaningsin placeof references.What analogydemandsis a theorythat ofthe form "s all sentences hasasconsequences meansrz" where 's'is replacedby a structural descriptionofa sentenceand'm'is replacedby a singularterm that refersto the meaningof that sentence;a theory, moreover,that providesan effective methodfor arriving at the meaningof an arbitrary sentencestructurally described. Clearly somemore articulateway of referringto meaningsthan any we have seenis essentialif thesecriteria are to be met.4Meaningsas entities, or the relatedconceptof synonymy,allow us to formulate the following rule relating sentences and their parts: sentencesare synonymous whose correspondingparts are synonymous ("corresponding"here needsspellingout of course).And meaningsasentitiesmay,in theoriessuchasFrege's,do duty,on occasionasreferences,thus losing their statusas entitiesdistinct from references.Paradoxically,the one thing meaningsdo not seem to do is oil the wheelsof a theoryof meaning-at leastas long aswe requireof sucha theorythat it nontrivially give the meaningof every sentencein the language.My objectionto meaningsin the theory of meaningis not that they are abstractor that their identity conditions are obscure,but that they haveno demonstrateduse. This is the place to scotch anotherhopeful thought.Supposewe havea satisfactorytheory of syntax for our language,consisting of an effective method of telling, for an arbitrary expression,whether or not it is independently meaningful (i.e., a sentence),and assumeas usual that this involves viewing each sentence as composed,in allowable ways, out of elementsdrawnfrom a fixed finite stockof atomic syntactical elements (roughly, words). The hopeful thought is that syntax, so conceived, will yield semanticswhen a dictionary giving the meaning of each syntactic atom is added. Hopes will be dashed,however,if semanticsis to comprisea theory of meaningin our sense, for knowledgeof the structural characteristics that make for meaningfulnessin a sentence, plus knowledgeof the meaningsof the ultimate parts, does not add up to knowledgeof what a sentencemeans.The point is easily illustrated

T R U T HA N D M E A N I N G by belief sentences.Their syntax is relatively unproblematic.Yet, adding a dictionary does not touchthe standardsemanticproblem,which is that we cannot accountfor even as much as the truth conditions of such sentenceson the basisof what we know of the meaningsof the words in them. The situation is not radically altered by refining the dictionary to indicate which meaning or meanings an ambiguous expressionbearsin eachofits possiblecontexts; the problem of belief sentencespersistsafter ambiguitiesare resolved. The fact that recursivesyntaxwith dictionary addedis not necessarilyrecursivesemanticshas been obscuredin some recent writing on linguisticsby the intrusionof semanticcriteriainto the discussionof purportedlysyntactictheories. The matter would boil down to a harmlessdifferenceover terminologyif the semanticcriteria were clear; but they are not. While there is agreementthat it is the centraltask of semantics to give the semanticinterpretation(the meaning) ofevery sentencein the language,nowhere in thelinguistic literaturewill one find, so far as I know,a straightforwardaccountof how a theory performsthis task,or how to tell when it has beenaccomplished.The contrastwith syntaxis striking. The main job of a modestsyntaxis to characterizemeaningfulness(or sentencehood). We may have as much confidencein the correctnessof sucha characterizationaswe havein the representativeness of our sample and our ability to say when particular expressionsare meaningful(sentences).What clear and analogoustask and test exist for semantics?s We decideda while back not to assumethat partsof sentenceshave meaningsexceptin the ontologicallyneutralsenseof making a systematic contribution to the meaning of the sentencesin which they occur. Since postulating meaningshas netted nothing, let us return to that insight.Onedirection in which it pointsis a certain holistic view of meaning.If sentences dependfor their meaningon their structure,and we understandthe meaningof eachitem in the structureonly as an abstractionfrom the totality of sentencesin which it features,then we can give themeaningof any sentence(or word) only by giving the meaning of every sentence(and word) in the language.Frege said that only in

T R U T HA N D M E A N I N G

117

the context of a sentencedoes a word have tencesgot from schemaZ when 's' is replaced meaning;in the samevein he might haveadded by a structuraldescriptionofa sentenceofZ and 'p'by that sentence. that only in the context of the languagedoes a sentence(andthereforea word) havemeaning. Any two predicatessatisfyingthis condition This degreeof holism was alreadyimplicit in have the same extension,6so if the metalanguageis rich enough,nothingstandsin the way the suggestionthat an adequatetheoryof meaning must entall all sentencesof the form "s of putting what I am calling a theory of meaning meansm." But now, havingfound no more help into the form of an explicit definition of a predilhan in meaningsof in meaningsof sentences icate "is 7." But whether explicitly defined or words, let us ask whetherwe can get rid of the recursivelycharacterized, it is clearthat the sentroublesomesingulartermssupposedto replace tencesto which the predicate"is I" applieswill 'm' andto refer to meanings.In a way, nothing bejust the true sentences of I,, for the condition could be easier:just write "s meansthatp," and we have placed on satisfactory theories of imagine p'replaced by a sentence.Sentences, meaning is in essenceTarski's Convention Z as we have seen,cannot name meanings,and that tests the adequacyof a formal semantical definition of truth.T sentenceswith "that" prefixed are not namesat The path to this point has beentortuous,but all, unlesswe decideso. It looks as though we the conclusionmay be statedsimply: a theoryof are in trouble on another count, however,for it is reasonableto expectthat in wrestlingwith the meaning for a languageL shows "how the meaningsof sentencesdependupon the meanlogic of the apparentlynonextensional"means ings of words" if it containsa (recursive)definithat" we will encounterproblemsashard as,or perhapsidenticalwith, the problemsour theory tion of truth-in-L. And, so far at least,we have no other idea how to turn the trick. It is worth is out to solve. The only way I know to deal with this diffiemphasizingthat the conceptof truth playedno ostensiblerole in statingour original problem. culty is simple,and radical.Anxiety that we are That problem,upon refinement,led to the view enmeshedin the intensionalspringsfrom using the words "means that" as filling between that an adequatetheory of meaning must chardescriptionof sentenceand sentence,but it may acteize a predicatemeetingcertainconditions. be that the successof our venturedependsnot It was in the nature of a discovery that such a predicatewould apply exactly to the frue senon the filling but on what it fills. The theorywill havedoneits work if it provides,for every sentences.I hope that what I am doing may be tences in the languageunder study,a matching describedin part as defendingthephilosophical (to replace'p') that,in someway yet to sentence importance of Tarski's semanticalconcept of be made clear, 'gives the meaning' of s. One truth. But my defenseis only distantlyrelated,if obviouscandidatefor matchingsentenceisjust at all, to the questionwhethertheconceptThrski s itself, ifthe objectlanguageis containedin the hasshownhow to defineis the (or a) philosophmetalanguage; otherwisea translationof s in the ically interesting conception of truth, or the metalanguage. questionwhetherTarskihascastanylight on the As a final bold step, let us try treatingthe position occupiedby p'extensionordinary use of such words as "true" and ally: to implementthis, sweepawaythe obscure "truth." It is a misfornrne that dust from futile "meansthat,"providethe sentencethat replaces and confusedbattles over thesequestionshas 'p' with a propersententialconnective,and suppreventedthose with a theoretical interest in ply the descriptionthat replaces's'with its own language-philosophers, logicians, psycholopredicate.The plausibleresult is gists, and linguists alike-from recognizingin the semanticalconceptof truth (underwhatever (I) s is T if andonly if p. name) the sophisticatedand powerfirl foundaWhat we require of a theory of meaning for a tion of a competenttheory of meaning. languaget is that without appeal to any (furThere is no need to suppress,of course,the ther)semanticalnotionsit place enoughrestricobvious connection between a definition of tions on the predicate"is Z" to entail all sentruth of the kind Tarski has shown how to con-

ll8

struct,and the conceptof meaning.It is this: the definition works by giving necessaryand sufficient conditionsfor the truth of every sentence, and to give truth conditionsis a way of giving the meaningof a sentence.To know the semantic concept of truth for a languageis to know what it is for a sentence-any sentence-to be true, and this amounts,in one good sensewe can give to the phrase,to understandingthe language.This at any rate is my excusefor a feature ofthe presentdiscussionthat is apt to shock old hands: my freewheelinguse of the word "meaning,"for what I call a theory of meaning hasafter all turnedout to makeno useof meanings, whetherof sentencesor of words. Indeed since a Tarski+ypetruth definition suppliesall we have askedso far of a theory of meaning, it is clear that such a theory falls comfortably within what Quine terms the "theory of reference" as distinguished from what he terms the "theory of meaning."So much to the good for what I call a theory of meaning,and so much, perhaps,againstmy so calling it.8 A theory of meaning (in my mildly perverse sense)is an empiricaltheory and its ambitionit to account for the workings of a natural language.Like anytheory it may be testedby comparing someof its consequences with the facts. In the presentcasethis is easy,for the theoryhas beencharacterizedasissuingin an infinite flood of sentenceseachgiving the truth conditionsof a sentence;we only need to ask, in selected cases,whetherwhat the theory aversto be the truth conditionsfor a sentencereally are.A typical test case might involve deciding whether the sentence"Snow is white" is true if and only if snow is white. Not all caseswill be so simple (for reasonsto be sketched),but it is evidentthat this sort oftest doesnot invite countingnoses.A sharpconceptionofwhat constitutesa theoryin this domain fumishes an exciting context for raising deep questionsabout when a theory of languageis conect andhow it is to be tried. But the difficulties are theoretical,not practical.In application, the trouble is to get a theory that comes close to working; anyone can tell whetherit is right.eOne can seewhy this is so. The theory revealsnothing new about the conditions under which an individual sentenceis true: it does not make those conditions anv

T R U T HA N D M E A N I N G clearerthan the sentenceiself does.The work of the theory is in relating the known truth conditions of each sentence to those aspects ('words') ofthe sentencethat recurin othersentences,and can be assignedidentical roles in Empirical power in sucha theother sentences. ory dependson successin recoveringthe structure of a very complicatedability-the ability to speakand understanda language.We can tell easily enoughwhen particularpronouncements of the theorycomportwith our understandingof the language;this is consistentwith a feeble insight into the design of the machineryof our linguistic accomplishments. The remarks of the last paragraph apply directly only to the special case where it is assumedthat the languagefor which truth is being characterizedis part of the languageused and understood by the characterizer. Under thesecircumstances,the framer of a theory will as a matterof courseavail himself when he can of the builrin convenienceof a metalanguage with a sentenceguaranteedequivalentto each sentencein the object language.Still, this fact ought not to con us into thinking a theory any morecorrectthat entails"'Snow is white' is true if and only if snow is white" than one that entailsinstead: (,9) "Snowis white" is trueif andonly if grass is green, provided,of course,we are as sureof the truth of (S) as we are of that of its more celebrated predecessor.Yet (,9) may not encouragethe same confidence that a theory that entails it deservesto be called a theory of meaning. The threatenedfailure of nerve may be counteractedas follows. The grotesqueness of (S) is in itself nothingagainsta theoryof which it is a providedthe theorygivesthe corconsequence, rect resultsfor every sentence(on the basisof its structure,there being no other way). It is not easy to seehow (,9)could be party to such an enterprise,but if it were-if, that is, (.I) followed from a characterizationof the predicate "is true" that led to the invariable pairing of truths with truths and falsehoodswith falsehoods-then there would not, I think, be anything essential.to the idea of meaning that remainedto be captured.

T R U T HA N D M E A N I N G

t

What appearsto the right of thebiconditional 't in sentencesof the form is true if and only if p," when such sentencesare consdquences ofa theory of truth, plays its role in determining the meaningof r not by pretendingsynonymybut by adding one more brush-stroketo the picture which, taken as a whole, tells what there is to know of the meaning of s; this stroke is added by viroe of the fact that the sentencethat replacesp'is true ifand only ifs is. It may help to reflect that (,S)is acceptable,if it is, becausewe are independently sure of the truth of "snow is white" and "grassis green"; but in caseswhere we are unsureof the truth of a sentence,we can haveconfidencein a characterization of the truth predicate only if it pairs that sentencewith one we havegood reasonto believeequivalent.It would be ill advisedfor someonewho had any doubts aboutthe color of snowor grassto accepta tlleory that yielded (.9), evenif his doubts were of equal degree,unless he thought the color of the one was tied to the color of the other. Omnisciencecan obviously afford more bizarre theories of meaning than ignorance;but then, omnisciencehas less need of communication. It must be possible,of course,for the speaker ofone languageto constructa theoryofmeaning for the speakerof another,though in this casethe empirical test of the correctrressof the theory will no longer be trivial. As before,the aim of theorywill be an infinite conelatioh of sentences alike in ruth. But this time the theory-builder mustnot be assumedto have direct insight into Iikelyequivalences betweenhis own tongueand thealien.What he must do is find out, however hecan,what sentencesthe alieh holds true in his own tongue (or better, to what degreehe holds themtrue). The linguist then will attemptto construct a characterization of truth-for-the-alien i,which yields, so far as possible,a mapping of i'sentences held true (or false) by the alien onto sentencesheld true (or false) by the linguis;q. Supposingno perfect fit is found, the residue of held true translatedby sentences held (andvice vena) is the margin for error (foror domestic). Charity in interpreting the and thoughtsof othersis unavoidablein directionaswell: just as we must maxiagreement,or risk not making senseof

119

what the alien is talking aboui,so we must maximize the self-consistencywe attribute to him, on pain of not understanding ftin. No single principle of optimum chariryemerges;the constraints therefore determine no single theory. In a theory of radical translation (as Quine calls it) there is no complbtelydisentanglingquestions of what the alien meansfrom questionsof what he believes. lVe do not know what someone meansunlesswe know what he believes;we do not know what someonebelieves unless we know what he means.In radical translation we are able to break into this circle, if only incompletely, because we can sometimes tell that a person accedesto a sentencewe do not understand.lo In the past few pagesI have been asking how a theory of meaning that takes the form of a truth definition can be empirically tested, and haveblithely ignoredtheprior questionwhether thereis any seriouschancesucha theorycan be given for a natural language. What are the prospects for a formal semantical theory of a natural lahgUage? Very poor, according to Tarski; and I believe most logicians, philosophersof language,and linguistsagree.ltl-et me do what I can to dispel the pessimism.What I can in a gerieral and programmatic way, of course; for lidre the proof of the pudding will certainly be ih the proof of the right theorems. Tarski concludesthe first sectionof his classic essayon the concept of truth in formalized languageswith the following remarks, which he italicizes: Theverypossibilityof a consistent useof theexpression'truesentence'which is in harmonywith thelaws of logicand the spiritof everydaylanguageseemsto be very questionable,and consequentlythe same doubtanachesto thepossibilityofconstructinga correctdefnition of thisexpression.r2 I-ate in the sameessay,he returns to the subject: the conceptof truth(aswell asotherscmanticalconcepts)whenappliedto colloquiallanguagein conjunction with the normal laws of logic leads inevitablyto confusionsandcontradictions. Whoever wishes,in spite of all difficulties,to pursuethe semanticsof colloquial languagewith the help of

120 exactmethodswill be drivenfint to undertakethe thanklesstaskof a reformof this language.He will to defineits structure,to overcome find it necessary the ambiguityof the termswhich occurin it, and intoa seriesof languages finallyto splitthelanguage of greaterandgreaterextent,eachof which standsin the samerelationto the nextin which a formalized It may,however languagestandsto its metalanguage. be doubtedwhetherthe languageof everydaylife, 'rationalized'in this way,wouldstill preafterbeing andwhetherit would not rather serveits naturalness featuresof the formalized takeon the characrcristic languages.l3 Tlvo themesemerge:that the universal character of natural languagesleadsto contradiction (the semantic paradoxes),and that natural languagesaretoo confusedand amorphousto permit the direct application of formal methods. The first point deservesa seriousanswer,and I wish I had one.As it is, I will say only why I think we are justified in carrying on without havingdisinfectedthis particularsourceofconceptual anxiety. The semantic paradoxes arise when the range of the quantifiers in the object languageis too generousin certainways.But it is not really clear how unfair to Urdu or to Hindi it would be to view the rangeoftheir quantifiers as insufficientto yield an explicit definition of 'true-in-Urdu'or 'true-in-Hindi'. Or, to put the matter in another,if not more serious way, there may in the natureof the casealways be somethe languageof thing we graspin understanding (the of truth) that we cannot concept another communicateto him. In any case,most of the problemsof generalphilosophicalinterestarise within a fragment of the relevant natural languagethat may be conceivedascontainingvery little set theory. Of coursethesecommentsdo not meet the claim that natural languages are universal.But it seemsto me this claim, now that we know such universalityleads to paradox, is susPect. Tarski'ssecondpoint is thatwe would haveto reform a naturallanguageout of all recognition beforewe could apply formal semanticalmethods.If this is true,it is fatal to my project,for the task of a theory of meaningas I conceiveit is not to change,improve or reform a language, but to describeand understandit. Let us look at the positive side.Thrski has shown the way to

TRUTHAND MEANINd; giving a theoryfor interpretedformat tanguageS' of variouskinds; pick one as much like English as possible.Since this new languagehasbeen' explainedin EnglishandcontainsmuchEnglish we not only may,but I think must,view it aspart of Englishfor thosewho understandit. For this fragment of English we have, ex hypothesi,a theory of the required sort. Not only that, but in interpretingthis adjunctof English in old English we necessarilygave hints connectingold and new. Wherever there are sentencesof old English with the sametruth conditionsas sentencesin the adjunctwe may extendthe theory to cover them.Much of what is calledfor is just to mechanizeas far aspossiblewhat we now do by art when we put ordinaryEnglish into oneor anothercanonicalnotation.The point is not that canonical notation is better than the rough original idiom. but rather that if we know what idiom the canonicalnotation is canonicalpr, we haveas good a theory for the idiom asfor its kept companion. Philosophershavelong beenat the hard work of applying theory to ordinary languageby the device of matchingsentencesin the vernacular with sentencesfor which they have a theory. Frege'smassivecontribution was to show how "all," "some," "every," "each," "none," and associatedpronouns, in some of their uses, could be tamed;for the first time, it was possible to dream of a formal semanticsfor a significant part of a natural language.This dream came true in a sharp way with the work of Tarski. It would be a shameto miss the fact that as a result of these two magnificent achievements,Frege'sand Tarski's, we have gaineda deep insight into the structure of our mother tongues. Philosophersof a logical bent have tended to start where the theory was and work out towards the complicationsof natural language.Contemporarylinguists,with an aim that cannot easily be seento be different, start with the ordinary and work toward a general theory. If either party is successful,there must be a meeting.Recentwork by Chomsky and others is doing much to bring the complexitiesof natural languages within the scope of serious semantic theory. To give an example: suppose successin giving the nuth conditionsfor some significant range of sentencesin the active

A N DM E A N I N G TRUTH voice. Then with a formal procedurefor transforming each such sentenceinto a corresponding sentencein the passivevoice, the theoryof truth could be extendedin an obvious way to this new setof sentences.la Oneproblemtouchedon in passingby Tarski doesnot, at leastin all its manifestations,haveto be solvedto get aheadwith theory: theexistence in naturallanguagesof "ambiguousterms."As long as ambiguity does not affect grammatical form, andcanbe translated,ambiguityfor ambiguity, into the metalanguage,a truth definition will not tell us any lies. The trouble,for systematic semantics,with the phrase"believesthat" in English is not its vagueness,ambiguity, or unsuitabilityfor incorporationin a seriousscience:let our metalanguagebe English, and all theseproblemswill be translatedwithout loss or gaininto themetalanguage. But thecentralproblem of the logical grammar of "believesthat" will remainto hauntus. The exampleis suitedto illustrating another, and related,point, for the discussionof belief hasbeenplaguedby failureto observe sentences a fundamentaldistinctionbetweentasks:uncovering the logical grammiu or form of sentences (which is in the provinceof a theoryof meaning as I construeit), and the analysisof individual wordsor expressions(which aretreatedasprimitive by thetheory).ThusCamap,in the first edition of Meaning and Necessiry,suggestedwe render"John believesthat the earthis round" as "John responds affirmatively to "the earth is round'as an English sentence."He gavethis up whenMatespointedout thatJohnmight respond affirmativelyto one sentenceand not to another no matterhow close in meaning.But there is a confusion here from the start. The semantic structureof a belief sentence,accordingto this ideaof Carnap's,is given by a three-placepredicatewith placesreservedfor expressions referring to a person,a sentence,anda language.It is a different sort of problem entirely to attempt an analysisof this predicate,perhapsalong betpvioristic lines. Not least among the merits of Tarski'sconceptionof a theory of truth is that the purity of methodit demandsof us follows from theformulationof the problem itself, not from theself-imposedrestraintof someadventitious puritanism. Philosophical

'l2l

I think it is hard to exaggeratethe advantages to philosophy of languageof bearing in mind this distinction between questions of logical form or grammar, and the analysisof individual concepts.Another examplemay help advertise the point. If we supposequestionsof logical grammar settled,sentenceslike "Bardot is good" raiseno specialproblemsfor a truth definition.The deep differencesbetweendescriptiveand evaluative (emotive, expressive,etc.) terms do not show here. Even if we hold there is some important sensein which moral or evaluativesentences do not have a truth value (for example, because they cannotbe 'verified'), we ought not to boggle at "'Bardot is good' is true if andonly if Bardot is good"; in a theory of truth, this consequence should follow with the rest, keeping track, as must be done,of the semanticlocation of suchsentencesin the languageas a wholeof their relation to generalizations,their role in such compound sentencesas "Bardot is good and Bardot is foolish," and so on. What is special to evaluativewords is simply not touched: the mystery is transferredfrom the word "good" in the object languageto its Eanslationin the metalanguage. But "good" as it featuresin "Bardot is a good actress"is anothermatter.The problem is not that the translationof this sentenceis not in the metalanguage-letus supposeit is. The problem is to frame a truth definition suchthat "'Bardot is a good actress'is true if andonly if Bardotis a good actress"-and all other sentenceslike itare consequences.Obviously "good actress" doesnot mean"good and an actress."We might think oftaking "is a good actress"as an unanalyzed predicate.This would obliterateall connection between"is a good acfress"and "is a good mother,"andit would give us no excuseto think of "good," in these uses, as a word or semanticelement.But worse, it would bar us from framing a truth definition at all, for there is no end to the predicateswe would have to treat as logically simple (and hence accomodatein separateclausesin thedefinitionofsatisfaction): "is a good companionto dogs," "is a good 28year-old conversationalist,"and so forth. The problemis not peculiarto the case:it is the problem of attributiveadjectivesgenerally.

T R U T HA N D M E A N I N G . It is consistentwith the attitude taken here to deem it usually a strategic error to undertake philosophicalanalysisof words or expressions which is not precededby or at any rate accompaniedby the attempt to get the logical grammar snaight.For how can we haveanyconfidencein our analysesof words like "right," "ought," "can," and "obliged," or the phraseswe use to talk ofactions, events,and causes,when we do not know what (logical, semantical)parts of speechwe haveto dealwith? I would say much the sameaboutstudiesof the 'logic'of theseand other words,andthe sentencescontainingthem. Whether the effort and ingenuity that has gone into the study of deontic logics, modal logics, imperativeand erotetic logics has been largely futile or not cannot be known until we have acceptablesemanticanalysesof the sentences such systemspurport to treat. Philosophersand logicians sometimestalk or work asif they were free to choosebetween,say,the truth-functional conditional and others, or free to introduce nontruth-functional sententialoperatorslike "Let it be the casethat" or "It ought to be the casethat." But in fact the decision is crucial. When we depart from idioms we can accomodate in a truth definition, we lapse into (or create) languagefor which we haveno coherentsemantical account-that is, no accountat all of how such talk can be integrated into the languageas a whole. To retum to our main theme: we have recognized that a theory of the kind proposedleaves the whole matter of what individual words mean exactly where it was. Even when the metalanguageis differentfrom the object language,the theory exertsno pressurefor improvement,clarification or analysisof individual words,except when, by accident of vocabulary, sraighrforward translation fails. Just as synonomy, as between expressions,goes generally untreated, so also synonomyof sentences,and analyticity. Even such sentencesas 'A vixen is a female fox" bearno specialtag unlessit is our pleasure to provide it. A truth definition doesnot distinguish between analytic sentencesand others, exceptfor sentencesthat owe their truth to the presencealoneofthe constantsthatgive the theory its grip on structure: the theory entails not only that these sentencesare Eue but that they will remaintrue underall significantrewritings

of their nonlogical parts. A notion of logical truth thus given limited application, related notions of logical equivalenceand entailment will tag along. It is hard to imagine how a theory of meaningcould fail to reada logic into is objectlanguageto this degree;and to the extent that it does, our intuitions of logical ruth, equivalence,andentailmentmay be calledupon in constructingand testingthe theory. I tum now to one more, and very large, fly in tJle ointment: the fact that the same sentence may at one time or in one mouth be true and at anothertime or in anothermouth be false.Both logicians and those critical of formal methods here seemlargely (thoughby no meansuniversally) agreedthat formal semanticsand logic are incompetentto deal with the disturbances causedby demonstratives. Logicianshaveoften reactedby downgradingnatural languageand trying to showhow to get along without demonstratives; their critics react by downgrading logic and formal sbmantics.None of this can makeme happy:clearly,demonstratives cannot be eliminatedfrom a natural languagewithout loss or radical change,so thereis no choicebut to accommodatetheoryto them. No logical errors result if we simply treat demonstrativesas constantsl5;neither do any problemsarisefor giving a semantictruth definition. "'I am wise' is true if and only if I am wise," with its bland ignoring of the demonstrative elementin "I" comesoff the assemblyline alongwith "'Socratesis wise' is true if andonly if Socratesis wise" with irs bland indifference to the demonstrativeelementin "is wise" (the tense). What suffersin this treatmentof demonstrativesis not the definitionof a truth predicate,but the plausibility of the claim that what has been definedis truth.For this claim is acceptableonly if the speakerand circumstancesof utteranceof each sentencementioned in the definition is matched by the speaker and circumstancesof utteranceof the truth definition itself. It could also be fairly pointed out that part of understandingdemonstratives is knowing therulesby which they adjust their reference to circumstance;assimilatingdemonstrativesto constant termsobliteratesthis feature.Thesecomplaints can be met, I think, though only by a fairly farreachingrevision in the theory of truth. I shall

123

T R U T HA N D M E A N I N G barely suggesthow this could be done,but bare suggestionis all that is needed:the idea is technically trivial, and quite in line with work being doneon the logic of the tenses.16 We could take truth to be a property, not of sentences,but of utterances,or speechacts,or times,andpersons; orderedtriplesof sentences, but it is simplestjust to view truth as a relation betweena sentence,a p€rson,and a time. Under such treatment, ordinary logic as now read applies as usual, but only to sets of sentences relativized to the samespeakerand time; further logical relations between sentencesspoken at different times and by different speakersmay be articulatedby new axioms.Suchis not my concern. The theory of meaning undergoesa systematicbut not puzzling change:corresponding to each expressionwith a demonsfrativeelement there must in the theory be a phrase that relates the ruth conditions of sentencesin which the expressionoccursto changingtimes and speakers.Thus the theory will entail sentenceslike the following: "I am tired" is true as(potentially)spokenby p at t if andonly ifp is tired at t. '"Thatbook was stolen" is true as (potentially) spokenby p at t if and only if the book demonstratedby p at f is stolenprior to /.17 Plainly, this course does not show how to eliminate demonstratives;for example, there is no suggestionthat "the book demonsnatedby the speaker"can be substitutedubiquitouslyfor "that book" sclva veritate. The fact that demonstratives are amenable to formal treatment ought greatly to improve hopes for a serious semanticsof natural language,for it is likely that many outstanding puzzles, such as the analysisof quotationsor sentencesaboutpropositionalattitudes,can be solvedif we recognize a concealeddemonstrativeconstruction. Now that we have relativized truth to times andspeakers,it is appropriate to glance back at the problem of empirically testing a theory of meaningfor an alien tongue.The essenceofthe methodwas,it will be remembered,to correlate held-truesentenceswith held-truesentences by wayof a truth definition, and within the bounds of intelligible error. Now the picture must be elaboratedto allow for the fact that sentences aretrue, and held true, only relative to a speaker ill {F

s

and a time. The real task is therefore to translate each sentenceby another that is true for the same speakersat the same times. Sentences obviouslyyield a very senwith demonstratives sitivetestof thecorrectnessof a theoryofmeaning, and constitutethe most direct link between languageand the recurrentmacroscopicobjects of humaninterestand attention.l8 ln this paperI haveassumedthat the speakers of a language can effectively determine the meaningor meaningsof an arbitrary expression (if it has a meaning),and that it is the cenfral task of a theory of meaning to show how this is possible.I havearguedthat a characterizationof a truth predicate describesthe required kind of structure, and provides a clear and testablecriterion of an adequatesemantics for a natural language.No doubt there are other reasonable demandsthat may be put on a theory of meaning. But a theory that doesno more than define truth for a languagecomesfar closer to constituting a completetheoryof meaningthansuperficial analysismight suggest;so, at least,I have urged. Since I think there is no alternative,I have taken an optimistic and programmatic view of the possibilities for a formal characterizationof a truth predicate for a natural language. But it mustbe allowedthat a staggeringlist of diffrcultiesandconundrumsremains.To namea few: we do not know the logical form of counterfactual nor of sentences about or subjunctivesentences, probabilitiesandaboutcausalrelations;we have no good idea what the logical role ofadverbsis, nor the role of attributive adjectives;we haveno theory for mass terms like "fire," "water," and "snow," nor for sentencesabout belief, perception, and intention, nor for verbs of action that imply purpose.And finally, there are all the sentencesthat seemnot to havetruth valuesat all: the imperatives,optatives,interrogatives,and a hostmore.A comprehensivetheory of meaning for a natural languagemust cope successfully with eachof theseproblems. ACKNOWLEDGMENTS An earlier version of this paper was read at the Eastem Division meeting of the American Philosophical Association in Decernber, 1966; the main theme traces back to an unpublished paper delivered to the Pacific Division of the American Philosophical Association in 1953. hesent

124 formulations owe much to John Wallace, with whom I have discussed these matters since 1962. My research was supported by the National Science Foundation.

NOTES l. Elsewhere I have urged that it is a necessary condi tion, ifa languageis to be leamable, that it have only a finite number of semantical primitives: see "Theories of Meaning and Leamable Languages," in Proceedings of the 1964 Intentational Congress for Logic, Metlndolog and Philosophy of Science (North-Holland Publishing Company, Amsterdam: l96s), pp. 383-394. 'structural 2. A description'ofan exprcssion describes the exprcssion as a concatenation of elements drawn from a fixed finite list (for example of words or letters). 3. The argument is essentially Frcge's. See A. Church, Intmduction to Mathematical logic, vol.I (hinceton: 1956), pp. 24-25. It is perhaps worth mentioning that the argument does not depend on any particular identification of the entities to which sentences are supposed to refer. 4. It may be thought that Church, in'A Formulation of the Logic of Sense and Denotation," in Structure, Method and Meaning: Essays in Honor of H. M. Shcffer,Henle, Kallen and L:nger, eds. (LiberalArts hess, NewYork: l95l), pp.3-24, has given a theory of meaning that makes essential use of meanings as entities. But this is not the case: Church's logics of sense and denotation are interpreted as being about meanings, but they do not mention expressions and so cannot of course be theories of meaning in the sensenow under discussion. 5. For a recent and insmrcdve statement of the role of semanticsin linguistics, see Noam Chomsky,'"Ibpics in the Theory of Generative Grammar," in Current Trends in Linguistics, Thomas A. Sebeok, ed., vol. III (Ihe Hague: 1966). In this article, Chomsky (1) emphasizes the central importance of semantics in linguistic theory (2) argues for the superiority of transformational grarnmars over phrase structur€ grammars largely on the grounds that, although phrase structure gmmmani may be adequate to define sentencehood for (at least) some natural languages, they are inadequate as a foundation for semantics, and (3) comments repeatedly on fte 'rather primitive state' of the concepts of semantics and remarks that the notion of semantic interpretation "still resists any deep analysis". 6. Assuming, of course, that the extension of these prcdicates is limited to the sentences of Z. 7. Alfred Tarski, 'The Concept of Truth in Formalized Languages," in Logic, Semantics, Metamathematics (Oxford: 1956), pp. 152-278. 8. But Quine may be quoted in support of my usage: ". . . in point of meaning. . . a word may be said to be determined to whatever extent the truth or falsehood of its contexts is determined." '"Truth by Convention," first published in 1936;now inThe Waysof

TRUTH AND MEANINC Paradox (NewYork 1966), p. 82. Since a truth definition determines the truth value of every sentence in the object language (relative to a sentence in the metalanguage), it determines the meaning of every word and sentence.This would seem to justify the title Theory of Meaning. 9. To give a single example: it is clearly a count in favor of a theory that it entails "'Snow is white' is true if and only if snow is white." But to contrive a theory t$at entails this (and works for all rclated sentences) is not trivial. I do not know a theory that succeeds 'mass with this very case (the problem of terms'). 10. This sketch of how a theory of meaning for an alien tongue can be tested obviously owes its inspiration to Quine's accountofradical translation in chapterII of Wotd and Object (New York: I 960). In suggesting that an acceptable theory of radical translation take the form of a recursive characterization of truth, I go beyond anything explicit in Quine. Toward the end of this paper, in the discussion of demonstratives, another strong point of agreement will turn up. I l. So far as I am aware, there has been very little discussion of whether a formal truth definition can be given for a natural language. But in a more general vein, several people have urged that the concepts of formal semantics be applied to natural language. See,for example,the contributions ofYehoshua BarHillel and Evert Beth ta The Philosophy of Rudolph Camap, Paul A. Schilpp, ed., (La Salle, I1l.: 1963), and Bar-Hillel's "Logical Syntax and Semantics," Language 30,230-237. 12. Tarski,ibid., p. 165. 13. [bid,.,p.267. 14. The rapprochement I prospectively imagine between bansformational grarnmar and a sound theory of meaning has been much advanced by a recent change in the conception of transformational grammar described by Chomsky in the article referred to above (note 5). The structures generated by the phrase-structurc part of the gralnmar, it has been rcalized for some time, are those suited to semantic interpretation; but this view is inconsistent with the idea, held by Chomsky until recently, that recursive operations are introduced only by the transformation rules. Chomsky now believes the phrase-stnrcturc rules are recursive. Since languages to which formal semantic methods directly and naturally apply are ones for which a (recursive) phrase-structuregrammar is appropriate,it is clear that Chomsky's present picture of the relation between the structures generated by the phrase-structure part ofthe grammar, and the sentencesofthe language, is very much like the picturc many logicians and philosophers have had of the relation betwe€n the richer formalized languages and ordinary language. (ln these remarks I am indebted to Bruce Vermazen.) 15. Quine has good things to say about this in Methods of logb (NewYork: 1950). See 8. 16. For an upto-date bibliography, and discussion, see A. N. hior, Past, Present, and Future (Oxford: 196il.