Alfred Tarski ' 5 SemanticTheory

5.1 Tarski' 5 Goals
Alfred Tarski, 1902 1983 one of the great logician-mathematiciansof this , was writing mathematical papers of great importance before his century, twenty-secondbirthday , and by his twenty- eighth he had invented the first formal semanticsfor quantified predicate logic, the logic of all reasoning about mathematics! The heart of this great accomplishmentis his theory . of truth . It has beencalled the semantictheory of truth , but Tarski himself never uses this label, preferring instead ' semantic conception of truth ', which he believes is the conception of truth that is the essenceof the , , correspondence theory (Tarski 1933 153; Tarski 1969 63). So in his own theorist. I shall discussin section 5.8 whether , he is a correspondence eyes Tarski is right in thinking that his is a correspondencetheory. In the meantimeI shall follow conventional practice and call it a semantictheory . of truth . He calls truth a semanticconcept becauseit can be defined in terms of other semantic concepts especially the concept of satisfaction (about , , which more below). Semantic concepts like satisfaction definition , and , , , designation deal with the relations between expressions and objects ' , ) (Tarski 1944 17 . Indeed, one of Tarski s principal goals in analyzing truth was to securethe foundations for what he called " scientific semantics,, 2 . He wanted, in other words, to establish the study of semantics as a , scientifically respectablediscipline (Tarski 1944 36). To accomplish this, he needed or at least thought he needed to provide a guaranteethat the , , discipline of semanticsdoes not presupposethe existenceof any abstract . entities whose existenceis not already presupposedby physical science The Logical Positivist school of philosophy, which was in its heyday at the time (the late 1920s, disparagedmost abstract entities as at best useless ) " or as at worst mysterious and " metaphysical objects belief , postulations in which is hardly more intellectually respectable than belief in ghostsand goblins. Under the influence of this school, Tarski embracedthe doctrine of physicalism ( 1944 sect 21, passim; 1936 406). Physicalism can be . , , defined in a rough way as the belief that all intellectually respectable


5 Chapter

conceptscan be defined ultimately and entirely in terms of the conceptsof . , logic, mathematics and physical science (I give a more detailed explication of physicalism in sections6.6 to 6.8.) Thus, to ensure that semantics conforms to the dictates of physicalism Tarski needed to reduce all , semantic concepts to physical and/ or logicomathematical concepts His . , strategy was to define all semanticconcepts savesatisfaction in terms of , truth . Truth itself is then defined in terms of satisfaction and finally , , satisfaction is defined in terms of physical and logicomathematical concepts alone (Tarski 1933 153 194 . (I explain the concept of satisfactionin , , ) section 5.4.) This strategy if successfulalso ensuresthat two of Tarski ' s , , other goals are met. No semantic term need be taken as primitive (i.e., undefined and semantic terms do not have to be defined circularly in ), terms of one another. Tarski also wanted his theory of truth to show how the grammatical structure of a sentence affects its truth value. Readers familiar with propositional logic will realize that the truth tables are a kind of es (incomplete) graphic definition of truth that accomplish just this task for the compound sentences that logic.3 BeforeTarski , no one had done this of effectively for quantified predicate ~ ogic, so he set it as one of his goals. This logical program is nowadays often called model theory, so we may ' say that Tarski s second program, besidesphysicalism was to create a , model theory for quantified predicate logic. In section 2.1 we saw one broader philosophical program in serviceto which an answer to the metaphysical project can be put . It was the epistemologicalprogram of evaluating competing theories of justification . Now we are introduced to two other philosophical programs in serviceto which Tarski wants to usean answer to the metaphysicalproject (specifically , as I interpret him, the extensional project) : physicalism and model theory. Each of these programs puts special criteria of adequacy on answersto the extensional project above and beyond the generaldemand that any answer to that project accurately describe the extension of ' is true ' . The epistemological program, as we saw, requires that we treat beliefsas truth bearers(although not necessarilyas the only kind of truth bearer). As we shall seeshortly, the physicalist program effectivelyrequires that a theory of truth make a detour, so to speak through the concept of , satisfaction and the model theory program effectivelyrequires the useof a , technique of definition called recursion (describedin section 5.3). Accordingly " " , what counts as the correct answer to the extensional project will dependon the philosophical context. A theory that servesperfectly well for

Alfred Tarski ' s Semantic Theory


one program may fail to meet the specialadequacycriteria of another. The that anyone answer to point to be kept in mind is that it is not necessary the extensional project be useful for all of these programs. In principle, there would be no causefor objection if Tarski had created two different extensional definitions of truth , one each for his two broader programs. , Presumably he would want these two definitions to be consistent with each other, but since both theories would be extensional consistency , would only require that what appearson the right -hand side of the ' = ' symbol in one of the definitions be extensionally equivalent to what ' ' appearson the right -hand side of the = in the other. In technical lingo , " the definiens(literally , the defining part ," the part on the right -hand side of the ' = ' symbol) of one of the two definitions must be extensionally equivalent to the definiensof the other. Tarski had at least two other goals for his theory. First , he wanted it to be immune from refutation by the Liar Paradox. This paradox and Tarski 's method of avoiding it are discussed sections9.1 and 9.3. Those in details of his theory relevant primarily to his solution of the paradox will not be mentioned in this chapter. Second Tarski wanted a theory that , would meet what he called the mate~ al adequacycondition. I shall have much to say about this condition later in this chapter and in section 6.1, but a brief characterization is possiblehere. The condition assertssimply that any good theory of truth has to entail all sentences the pattern of on the following : ' The wall is red' is true = the wall is red ' Snow is ' slippery is true = snow is slippery ' Bob ran around the track ' is true = Bob ran around the track ' Grass is red' is true = grassis red For every sentenceof the language for which truth is being explicated , there is an equivalencein which the sentence mentioned on the left side is and usedon the right side. It is easyenough to formulate the pattern of all theseequivalences and Tarski does so with a formula that has variously , beencalled " form T , " " schemaT , " and " convention T " : X is true if, and only if, p. We shall call any suchequivalencewith ' p' replaced any sentence the of ( by " " ' ' to , language which the word true refers and X replaced a nameof this by sentence " eQuivalencetheform(T)." 4 ) an of

Others havethought that Tarski meant to reject other conceptionsof truth asjust plain wrong. 5. so it is sentences that he regards as truth bearers.144 ChapterS It is not hard to see why Tarski thought a minimal condition on any that adequatetheory of truth is that it entail all T -sentencesi. notice that one can accept the T the sentence the right -hand side of the ' = ' express a mind-dependent on es state of affairs. other conceptions of truth .e. .1). truth actually entail the T -sentences . conception for logic.6 What we need is a . Some ) interpreters haveattributed to Tarski the very tolerant view that his is only one of many conceptions of truth and that other conceptions (e. Unfortunately . It is hard to see . as he does in the very generouspassage which he allows that if in we were ever to coin different names for the concepts defined by the different conceptions he would not mind if ' true ' were given to someother . This is why Tarski s material adequacy condition (hereafter MAC ) requires that a theory of . meaning of true (Tarski 1933 153 . so any theory that is incompatible with them would be false .28). conception and frue were given to his (Tarski 1944 27. . what remaining value he could possibly have seenfor other conceptions .g. or names a class. And at one time he explicitly denied that he was trying to capture the ordinary ' ' . the textual evidenceis ambiguous He . . they are just different. as he would put it . theories There are passagesin which he seems to take the tolerant attitude. ' ' . -sentences insist that but (In this regard..) But a theory of truth can be mighty implausible and still be ' compatible with the T -sentences(see section 6. sentences ( instantiate convention T ) : the T -sentences obviously true (if we keepin are mind that the equivalence asserted by aT -sentenceis extensional not intensional). then. One disputed issue in Tarskian exegesis concerns his attitude toward other theories (or. of sentences.5 It is worth noting that besides thinking that the semantic conception is the preferred . never quite explicitly pronounces on the rightness or wrongnessof other . .2 Tarski' s Theory ' Tarski says that ' true expresses a property . mathematics semantics and physicalism Tarski also commended his theory to epistemologists( 1933 267). coherence pragmatic) are not necessarilywrong. But he grew bolder later in life and ) was then prepared to makejust this claim and to reject the other theories for failing to capture the ordinary meaning of truth .

theory by seeinghow it would apply to a very simple languagewith a finite number of sentencesLet us say that the languagehas only the following . five sentences and that it provides no way to produce any new sentences : The table is round. John loves Mary .->. something logically equivalent is just such a logical conjunction : following ' ' (s)[ s is true = eithers = The table is round and the table is round) ' The ' or (s = carpet is purple and the carpet is purple) = ' John loves Mary ' and John loves Mary ) or (s or (s = ' Bob is a noodle-brained fool ' and Bob is a noodle-brained fool) or (s = ' Napoleon is alive' and Napoleon is alive)] ' . Napoleon is alive. conjunction of all of the T -sentencesTarski recognizesthis fact in saying that each of these T -sentencesis a " partial definition " of truth . with ' s' ranging over sentencesLet me begin this examination of Tarski ' s . Bob is a noodle-brained fool. that would meet the latter condition is a .Alfred Tarski ' s SemanticTheory 14S formula of the form (s)(s is true = . By logical conjunction . 1933 187 . and a " " " " completedefinition would be a logical conjunction or logical product ' ' Tarski means of all of them ( 1944 16. The carpet is purple. . if we can call it that. ) to a conjunction of the T -sentencesThe . So we want a theory that entails all five of the following T -sentences : ' The table is round ' is true = the table is round ' The ' carpet is purple is true = the carpet is purple ' John loves ' Mary is true = John loves Mary ' Bob is a noodle-brained fool ' is true = Bob is a noodle-brained fool ' ' Napoleon is alive is true = Napoleon is alive One theory.

new sentences can be built from old ones Thus the language includes the among its sentences following : The table is not round. .146 Chapter 5 The preceding extensional analysis of ' is true ' is completely satisfactory from Tarski ' s point of view (on the assumption that such terms as ' fool ' and ' alive' can be reduced to physical terms) for the simple language in question. . 1933 189 . . then Bob is a noodle-brained fool. The carpet is purple. 5. The technique is used to create an extensionalanalysis . John loves Mary . ' or ' . ) of terms whose extension (the set of things to which the term applies) is . or John loves Mary and Napoleon is alive. . or John loves Mary and Napoleon is alive. then' . the definition of truth . let us now imagine a language with an infinite number of sentences It in includes all five of the sentences the last language but it also has the . and Napoleon is alive. truth -functional operators ' not ' . . and there is no need for this simple language to make useof the conceptof . which is impossible(Tarski 1933 188 . then. . All other semantic terms can now be defined in terms of truth .3 Recursion But the definition only works because languagein question has a finite the number of sentencesIf it had an infinite number. given operator can be applied to build a still bigger sentenceThis is why the languagecontains an infinite number of sentencesObviously.69. 68. then the . ) . would haveto be infinitely long. do we define truth for such a language Tarski ' s answer is ? that we must make use of a technique called " recursivedefinition " ( 1969 . satisfaction at all. and the carpet is purple. ' Note that in the last of these sentences John loves Mary ' appears as a clause more than once and the operator ' and' appearsmore than once. with which . So . can There is no limit to how many times a given atomic sentence appearas a clause in a larger sentence and there is no limit to how many times a . and ' if . If the carpet is purple.. ' and' . John loves Mary . language How . definition of truth cannot have a separateclausefor each sentencein the .

This is becausetrain segmentscome in different lengths and are made up of different kinds of railroad cars. kind of railroad car. Then one must mentally divide it into two sections along one of the . (g)(g is a train segment= g is an engine or g is a dining car. or g is a freight car. segmentsThe definition also ensuresthat only a segmentof one or more . we need a separate disjunct for ea~h. set of train segmentsall segmentsother than the five types of basic . Another point to make about the ' definition is that the term being defined ' train segment. is a five-car segment Thus the last clauseof the definition gathersinto the . railroad cars will count as a train segment We are assuredof this because ' the ' = ' is translated ' if and only if . . or g is a sitting car. object is a train segment If the object does not consist of parts coupled together. we have one disjunct for each of the most basic kinds of train segments those that are only one car long. I begin with the following recursivedefinition : of a train segment . If there were only one . or g is a sleepingcar. Instead. however because earlier clauses . ) reappears in the last clauseof the definition. the object is a train segment if not. segment And if c is a freight car. . The last clause in the definition is . in the last clause By this I mean that one could success .Alfred Tarski ' s SemanticTheory 147 of infinite.e. whether or not that . Supposethe object does consist of two or more parts coupled together. three-car train segment If de is a two-car segment then de coupled to abc . simply saysthat any coupling of train segments itself a train segment So if a is a sleepingcar and b is a dining car. But notice that we do not have to have a separatedisjunct in the definition for every train segmentthere ever was or will be. The set we are interested in is the set of all true sentences the but let us first practice the recursive technique on the terms . The definition is not viciously the cashout the reference train to circular. then one simply checksto seewhether it is one or another of the five kinds of railroad cars. then c coupled to the ab segmentis a . language ' ' ' train ' segment and train . we would need only one basis clause reading ' g is a railroad car ' . then a coupled to b is a train . not.. or a and b are train segments and g is a coupled to b) The first thing to notice about this definition is that it is disjunctive. . If it is. recurs (i. segments fully use the definition to determine for any object in the world. But sincethere are different kinds of one-car train segments .

If all the elemental parts are railroad cars . one would eventually examine each elemental part to determine whether or not it is a railroad car . The hard part is filling in the blank . Consider the following definition . . there was only one recursive clause . Thus we need only one more clause in our ( t )( t is a train = t is an engine . ' ' Now let us define train recursively . ). This is not always the case . The first clause must define the most basic type of train .148 5 Chapter couplings . which is simply an engine by itself . If there is more than one way to build up nonbasic members . No matter how many elemental parts the object originally has . In both of the examples . up by coupling train segments to the back of trains . the object is not a train segment . with its three recursive ' clauses : ( x )( x is in John Doe s family = x is John Doe . If one of the sections turns out to consist of two or more parts coupled together . followed by one or more recursive clauses defining how being other members of the set are built out of the more basic members . there will have to be a separate recursive clause for each method . Thus ( t )( t is a train = t is an engine . or a is a train segment and b is a train and t is a coupled to the back of b ) Notice again the characteristic features of a recursive definition : one or more clauses noncircularly defining the most basic members of the set defined . If any of them are not . We start with ( t )( t is a train = ). Next one must examine each of the sections following the same recipe . the object is a train 8 segment . ' or a is in Doe s family and x is married to a ' or a is in Doe s family and x is born to a ' or a is in Doe s family and x is adopted by a ) . then one must mentally divide it into two subsections . or All other trains can be built engines definition or other .

the recursivedefinition of truth is the following : ' ' (s)[ s is true = eithers = The table is round and the table is round) ' or (s = ' The carpet is purple and the carpet is purple) ' or (s = ' John loves Mary and John loves Mary ) ' or (s = ' Bob is a noodle-brained fool and Bob is a noodle-brained fool) ' or (s = ' Napoleon is alive and Napoleon is alive) ' or (s = ' not p and it is not true that p) ' or (s = ' p or q and either it is true that p or it is true that q) ' or (s = ' p and q and it is true that p and it is true that q) ' or (s = ' if p. entails all the T -sentencesfor the language Consider. then q and either it is not true that p or it is true that q)] This definition also meets the material adequacy condition. and conditionalization. understanding of recursivetechnique by defining the set of tree branches : basic branches grow directly out of a tree trunk . from them: negation. for example the sentence . ' and ' is true when both it is true that and it is true that q. Accordingly. becauseit . disjunction. then q is true when either it is false that p or it is true that q.'s Tarski SemanticTheory Alfred 149 The recursivemethod works so long as there are only a finite number of types of basic membersof the set in question and there are only a finite number of ways in which nonbasic memberscan be built up or added. branches grow out of branches To return to the problem of defining truth for a languagewith an infinite and number of sentencesI note first that there are five atomic sentences . Before going further. the reader is encouraged to test her or his . and that p p q 9 ' that ' if p. . four ways to build compound sentences ' ' conjunction. Recall that not p is true wheneverit ' is falsethat p. while other Hint . that ' p or q is true if either it is true that p or it is true that q.

this languagehas an infinite number of quantifiers and variables Because sentenceswe shall have to make useof the recursivetechnique again. This.150 ChapterS ' ( 1) Bob loves Mary . An open sentence is an expression that is . So ' Bob loves Mary . In addition to the truth -functional operators. and ' ' (5) The carpet is purple is true. there is an extra little wrinkle in this languagethat complicatesmatters. or the table is round and the carpet is purple.' The definition tells us that this sentenceis true if and only if either ' ' (2) Bob loves Mary is true. this languagehas . and (5) is true if and only if the carpet is purple. and topics. . or the table is round and the carpet is purple ' is true = Bob loves Mary. ) sentencesand quantifiers. or ' (3) The table is round and the carpet is purple ' is true. quantified predicate logic. 5. The two kinds of expressions are open . But (4) is true if and only if the table is round. In the last languagewe looked at. which is the logic of all reasoning about numbers sets angles vectors matrices and other mathematical objects . all the sentences wereeither basicsentences or built up from basic sentencesbut in a quantified languageit is possible . Specifically he wants a truth definition for the language of . . .4 Satisfaction But Tarski wants to define truth for languageseven more complex than the last one. but . . or the table is round and the carpet is purple. to build new sentences combining two expressionsneither of which is by itself a sentence( 1933 189 . . of course is one of the T -sentences . The definition further tells us that (2) is true if and only if Bob loves Mary and that (3) is true if and only if ' ' (4) The table is round is true.

. sentenceis a kind of sentenceIt is not.Alfred Tarski ' s SemanticTheory exceptthat it has a variable in grammatically completejust like a sentence one or more of the placeswhere one would expect to seea noun. thereby creating . needed for mathematics can get along with just the existential and universal quantifiers. Each of : the following is an open sentence x is a noodle-brained fool. . anything does not. x is the father of y. does not say anything. (The logic . ) expressionstill counts as an open (nongenuine sentence .) Binding the variables in an open sentence and . and either z includes y or y is a fool. . each variable in the open sentence There is only one quantifier per variable even if some of the variables appear more than once in the : . An expressionthat does not say becauseit really . (x )[ x is black and (y)(3z)(y is betweenx and z or z loves x )] Note that if even a single variable is left unbound in an expression the . x is black. a genuinesentenceis simply a matter of quantifying over . An open sentence not a sentence . sentenceis neither true nor false We can turn an open sentenceinto a by genuine sentence closing it . x is betweenSam and Mary . There are an infinite number of quantified sentencesIn itself this might ? not seemto be a problem. x is the father of John. There are two ways to do this: replace the variables with names or bind the variables with quantifiers. ' The term ' open sentence is misleading in that it implies that an open is . sentence Hence the following are all genuine sentences (x )(x is a noodle-brained fool ) (x )(3y)(x is the father of y). Why not just treat theselike other sentences own clause in the Give each noncompound quantified sentence its . of course say anything true or false Thus an open . Bob is betweeny and z. x is betweeny and z.

functional operators just as well as can genuine sentences. . can be made by affixing appropriate quanti fiers to an open sentence. (3x)(x is red) ( 3x)(3y)(x is red and y is red) (3x)(3y)(3z)(if x is red. however . This option . parts evenfor languages with only one predicateand only one kind of quantifier.1O Tarski ' s great insight is this: sincethe property of truth is not possessed we by open sentences must find some other property with the following characteristics : . by . But there are an infinite number of open sentences because open sentencescan be compounded with the truth . The possession nonpossession the property by a given quantified or of sentence completely determined by the truth -functional operators in the is sentenceand by the possession non-possession the property by the or of clausescontained within the overall sentence . And again. It can be possessed both open sentences and genuine sentences . and define the truth of compound quantified sentences recursively . Since the open sentences that are the parts of these quantified sentences have no truth value. Therein lies the dilemma that Tarski solved. one quantifier for each variable in the open sentence. It is possibleto define truth in terms of genuinesentencespossessing or not possessing property so that the resulting definition of truth will the entail all of the T -sentences . ( For an example . we cannotrecursivelydefinethe truth of suchsentences in termsof the truth valueof its parts. Quantified sentences. is itself a sentenceAnd the precedingexamplesshow that this holds . again . because some compound quantified sentences are not built up from other genuine sentences. then eithery is red and z is red or z is not red) So there are an infinite number of quantified sentencesnone of whose .152 Chapter 5 definition of truth . see the last member of the list of open sentences given above . The first two characteristics ensure that we can define the property in question recursively and thus that we can apply our definition to languages with an infinite number of sentences The third characteristic . uncompounded open ' . sincethere are an infinite number of such sentenceswe cannot give them each its own clausein the . definition of truth .) Thus all of the following are genuine sentences. is not available .

I shall define this property in a by moment.) ' It will come as no surprise now to learn that the table satisfies x is ' and it satisfies ' ' purple and x is round if and only if it satisfies x is purple ' x is round ' . it satisfies . ' ' ' if x is purple. . compareTarski 1933 195 . out let me simplify by leaving sequences genuinesentences of account and and talking only of what it means for an individual object to satisfy an ' ' . Finally . it counts as a semanticconcept. Tarski s theory is at least a quasi-realist theory. (Sincesatisfaction is a relation betweenexpressions and parts of the world . This will ) .5 that it is most usefully interpreted as a Realist theory. a relation . and falseotherwise (Tarski 1944 25. The table satisfies x is purple or x is round ' or it satisfies' x is round ' . : becomes " A sentence true if it is satisfied by all [ sequences objects of] " . ' remain a bit mysterious until I presentTarski s explication of the concept of satisfaction but first let me formalize what he has given us so far: . note how simple the resulting definition of truth is . Tarski says a relational property. So the facts of the ' matter are important . es So an object satisfies an open sentenceif and only if it possess the the predicate of the open sentence(Tarski 1933 . of course or by being constructed out of a purple-colored material. So . purple . property expressedby ' : of 190 . . sentences Satisfaction is. purple.Alfred Tarski ' s SemanticTheory 153 ensures that the definition of truth will meet the material adequacy condition (Tarski 1933 189 . It satisfies' x ' x is if and only if it satisfies purple ' ' ' is not purple if and only if it fails to satisfy x is purple .) For the moment. So the table satisfiesthe if compound open sentence and only ' ' if it is both round and purple. between a sequenceof objects on the one hand and a sentenceor open on sentence the other. (I shall argue in section 6. has all the requisite characteristics A more correct name would be ' ' satisfaction a sequenceof objects . open sentence Let x is purple be the open sentence'and let my table be ' how does the table satisfy x is purple ? By being painted the object. but first . where ' s' ranges over sentences Now what we need is a definition of satisfactionthat will ensurethat this definition of truth will entail all the T . . The essence Tarski s theory is already beginning to emerge the ) ' ' table satisfies x is purple when the table is purple. ) ' Satisfaction' is the nameTarski gaveto the property he discoveredthat . of (s)(s is true = s is satisfied by all sequences objects). then x is round if and only if it either fails to satisfy x is ' or it satisfies' x is round ' .

the sequenceNote that sequences . the chair. the objects is important when dealing with sequencesThe following sets . the table) ( the chair. of . . like sets can be infinite in size(e. Bulgaria. sequence . the table. which must love which? Theseare the sort of questions that forced Tarski to resort to sequences objects( 1933 191 . Finally . the table. the chair. there is some sequencethat has just that object in the fourth place and there is a . the table. . the . under what circumstanceswill the table and the chair satisfy ' x loves y ' ? Must they love each other.g. democracy } But the following are all different sequences : ( the table. there are no limits on what order objects can come in. . in other words. {democracy the chair.154 5 Chapter What about the open sentence' x loves y ' ? Shall we say that the table satisfiesthis when it lovesitself? Or should we say insteadthat only a set of two objectscan satisfyan open sentence with two variables? And if we say the latter. sequence natural numbers and can have abstract objects as members ) Note also that an object can appear more than once in a sequenceSo the . democracy) When you changethe order of the objectsin a sequenceyou have changed .. that hasjust that object in both the fourth and sixteenth places . ) of objects is a lot like a set of objects except that . that has just that object in the sixteenth place and there is a sequence . ( the table. democracy } . unlike a set the order of . the table. . there is some sequencethat has just that ordering. democracy ) . the chair. . democracy the table. ( democracy the chair. So for all the objects in the world and each possible ordering of them. . as funny looking sequences well. . or is it only necessary that one love the other. ) When Tarski speaksof all infinite sequenceshe means to include these . ) . and if so. the table. A sequence of . : following are also sequences ( the table. the table} { the chair. An implication of this is that for any object in the world . are all identical to eachother . theseare really not different setsbut are different ways of naming the sameset: { the table.

. there is also no limit to how many variablesit can have. In general for any object. . . It does not ma~ what the other members of the are. . there is another sequencethat is just like this one except that it has the table in the fifth place. they are irrelevant. open sentencewith more than one variable is satisfied Since there is no can limit to how big a compoundedopen sentence be. the chair. (Hereafter when I use the word ' sequence I . . . there is a . . satisfyingan open sentenceThe open sentence XI is purple is satisfiedby an infinite sequenceof objects if and only if the first member of the ter sequenceis purple. Because variable in the open sentence the is the first variable. Thus an infinite sequence satisfy' if XI3 is purple. then x210vesX735 X46is will or fails to satisfy ' X13is purple ' or it round ' if and only if either the sequence does satisfy ' X2 loves X735or X46is round ' . . An infinite sequence the first member of the sequence loves the second but unlessthe second . So the is if original open sentence satisfied by the sequence and only if either the 13th object in the sequence not purple or the 2nd object loves the 73Sth is ' or the 46th object is round. (The sameimplication holds. only the first memberof the sequence matters. our recursivedefinition of satisfaction One other complication is necessary before we can understand how an . sequence just like this one except that it has just that object in the fifth . of course for all the other placesin the .) So much for satisfactionas applied to open sentencesHow . democracy the table. . and z with the variables XI ' X2 X3 X4 . . refer always to an infinite sequence .' Alfred Tarski s SemanticTheory 155 Another implication is that for a given sequencesay .) sequence The fact that this is so will turn out to be highly significant for . satisfying an open sentence we talk of an infinite sequenceof objects ' ' . The truth -functional operators will continue to work as they have. ( the table. and it doesnot matter whether theseother objectssatisfy' XI sequence is purple ' . y. the sequence will not satisfy ' X2 loves XI ' . An infinite ' ' sequencesatisfies X2 is purple if and only if the second member of the is satisfies' XI loves X2' if and only if sequence purple. ) . place. So we must replaceour x . Bulgaria. a quantified sentence We know that if the definition of ? satisfy sequence truth is to entail all the T -sentencesthen a universally quantified sentence . And there is another just like this one exceptthat it has Downing Street in the fifth place. also loves the first . Now instead of talking of a single object ' ' ' . It will satisfy the latter if and ' ' ' ' only if either it satisfies X2 loves X735 or it satisfies X46is round . then doesa .

had better turn out to be true when and only when say (X4 everything in the world is round. this conclusion applies to existentially quantified ' sentencesas well and to nonquantified sentenceslike ' Ralph is short . so far as universally . .) 2. S must satisfy the open sentencethat would be created by deleting the ' ' quantifier. the condition that must be met for S to ' ' )(X4 is round ) is exactly the samecondition that any sequence satisfy (X4 must meet to satisfy this sentencenamely. So in this caseit must satisfy X4 is round . sequencethen everything in the world is round. (X4 As it will turn out . . sentenceas a sentencethat is satisfied by somesequence The resulting -sentences definition would entail all the T corresponding to each universally . the two definitions amount to exactly the samething. . there is some sequence just like S except that it hasjust that object in the fourth place. This sameopen sentence must also be satisfiedby everysequence that is like S except that it has a different object in the fourth place. So. As noted above.156 5 Chapter ' ' )(X4is round) . ' ' universally quantified sentencesuch as (X4 )(X4 is round ) : 1. quantified sentenceof the language It would entail. says in effect that everything in the world other than the object in the fourth place of S must be round. works perfectly Sincea definition of truth as satisfactionby somesequence ? well. everything must be round. So if . why did Tarski define truth as satisfaction by all sequences The answeris that. And what goesforsequenceS goesfor any other sequence well. just Recall from above that for every object in the world . sincecondition (2) tells us that all of thesesequences must satisfy ' X4is ' the condition round . Thus ' (X4 )(X4 is round ) ' is satisfiedby asequenceSif and only if everything in the world is round. Betweenthem ( 1) and (2) are saying that every object must be round. Tarski ensures this by setting two conditions that must be met for a sequencecall it sequenceS to satisfy a . for example ' )(X4 is round ) is true = everything is round. quantified sentencesare concerned we might just as well define a true . . (Thus whatever objectS has in the fourth place must be round. Thus. then eo ipso the sole condition for the satisfaction of this sentence by all the other . as a practical matter. And if it is satisfied by some . the sole condition for S' s satisfying this sentenceis met. as The universally quantified senten is not satisfied by some sequence ~ unless everything in the world is round.

there is somesequence just like S exceptthat it hasjust that object in the kth place. The > (3xt)4 is satisfiedby a sequence > condition ensuresthat this will be so: following ' ' An expressionof the form (3xt)4 is satisfied by asequenceS if and > differing from S in at most the kth place satisfies only if somesequence 4. Notice here too that if > satisfiesthe existential claim. ' turn out that ' (3xt)4 is true when and only when something 4 s. First . in which case none of the other sequences either (Tarski 1933 194 . So if one sequence sentence that could only be because not ff one fails to satisfy the . Thus it matters not whether we define truth as . The samepoint . or satisfaction by some sequence as satisfaction by all sequences Sometechnical points before we go on. > So if the definition of truth must entail all the T sentencesit had better . we can further sequences generalizethe condition so that it applies to any universally quantified follows the quantifier . Third . differing from S in at most the fourth place. one sequence . is . all open sentencesThen any exHression the form (xJ4> is satisfied over differing from S > by asequenceSif and only if 4 is satisfiedby all sequences II in at most the kth place. just means satisfied by some sequenceit had better turn out that ' ' when and only when something 4 s. Let 4 range sentenceno matter what open sentence > . And since > > ' true ' . is to be satisfiedby them all. Second we can generalizethe new combined condition so that it doesnot matter which variable appears . ) satisfy the sentence : see applies to all other genuinesentencesto be satisfied by one sequence . in the sentence Let k be any integer. Then any expression of the form ' is round) ' is satisfied by asequenceS if and only if the open (Xt)(Xt sentencecreated by deleting the universal quantifier is satisfied by all differing from S in at most the kth place. sequences also met. So the condition will be met when and only when somethingin the world 4 s. conditions ( 1) and (2) can be combined into a single condition requiring that the open sentence created by deleting the universal quantifier be satisfied by all sequences . > But as we saw above. would everything is round. has this form : 12 for The T -sentence an existentially quantified sentence ' ' > (3xt)4 is true = (3xt)4 . they all do. for every object. as we shall . ' ' of .' Alfred Tarski s SemanticTheory 157 satisfiesthe sentencethey all do.

genuinesentences the languageare quantified sentencesAlso following Tarski ( 1933 168 I shall have only two truth -functional operators: ' not ' . ' ' and sufficient that the table be colored satisfy x is purple . and ' or ' . then' can be definedin terms of . In the ' ' " ' ' ' following definition . means not ' it is not the casethat . There is no point to complicating the languagewith any others. and S does not satisfy c /> / or (J = "" or c ' and eitherS satisfies' " or S satisfiesc /> / or (J = ' (Xi) c '. " . and the kth object in S is red) or (J = ' Xi loves x) ' for somek andj . . The reasonthesemany different clausesare required is that the language and sufficient conditions for an object's satisfying a given open necessary sentence different from the conditions necessary are and sufficient for that to satisfyany other open sentenceAs I noted above for the table to . objects in our language at least to begin with. ).' . . object . rangesover integers The -. . that is. ' becauseand' and ' if . and the kth object in S loves the jth object in S) or (J = ' -. 5. (J . the kth place satisfiesc / ' ' and some or (J = (3Xt) c that differs from S in at most the /> sequence 13 kth place satisfiesc )] . is (s)(s is true = s is satisfied by all [ or some sequences ] ). . c '. . /> And the definition of truth .158 Chapter The language for which I shall define ' satisfaction' has just two ' ' ' ' predicates is red and loves . The language also has the infinite set of variables described above. and c all range over both genuinesentences /> ' ' like ' k ' ' ' ' ' and open sentencesj . This means that the only in . it is necessary . . once again.s Names and Natural Languages One outstanding characteristicof Tarski ' s account is that the definition of satisfactionmust have a separateclausefor eachpredicatein the language for which we are defining truth . and then another clauseexplicateswhat it to meansfor a sequence satisfy ' loves'. So the official definition of or satisfaction is: (J (J is satisfied by an infinite sequenceS )[ = (J = ' Xi is red' for some k. and every sequence that differs from S in at most /> . Following Tarski. . I shall allow no namesof . and so on for every predicatein the to . one clauseexplicateswhat it means for a sequence satisfy ' is red' . not ' and ' or ' .

Thus. it is necessary '" 14 So for Tarski . languageshave an infinite number of predicates Think of the series of ' is the first in line ' ' is the secondin line ' ' is the third in line '. and someonemight believethat Ralph is secondin line and so forth . . but natural . . We . predicates Of course all lines are finite. typically usea variable like p to range over all predicates Such schemas . It would not mean that such predicates do not exist. series of clauses one for each stand .557823rd . . for all the predicatesin the seriesgesturedto above. predicatesin the language Such a schemaclause (for a language whose are all one-place) might look like this: predicates ' ' (z = PXt . for some k.'s Alfred Tarski SemanticTheory 159 ' ' and sufficient purple. But important that the definition apply to any for any predicate in the language there is a potential belief in which just . but for the table to satisfy x is round . reasoning which are the languagesthat interested Tarski . model and formal semanticsalso tend to hide this fact by replacing all the theory. the 36th in line . On the other ' ' ' ' hand. Textbooks in metalogic. ) . the 1. as a spacesaving read it not as a real clause in the definition but only -in for a whole. but this only meanswe would eventually get . ' satisfaction of " x is that the table be round. They wish to define truth ultimately for beliefs and it is . descriptions like ' the first in line . not a genuinelycompleteclause Strictly speaking it is senseless are to . and the kth member of S is an element of the set denoted by ' P ' ) ' ' But a schemais an open sentence no quantifier binds the P ) and is thus ( . Someonemight somedayhave the belief that Ralph is first in line. etc. only works for languageswith a finite number of predicates (Compare Putnam 1978 10) This is no hindrance for the languagesof mathematical . . purple ' satisfaction of " x is round " ' . . possibly infinite . epistemologists belief someonemight have. . This turns out to be a different concept from ' fact was partly hidden in Tarski s presentation of this theory becausehe used as his example languagea languageof set theory that has only one ' ' . predicate is included in ( 1933 165 169 . to predicatesthat cannot be truly predicated of anything. without modification Tarski 's technique of defining satisfaction and hencetruth is not applicable . clauses for the various predicates with a single schemaclause for all ' ' . . to a natural language1SThis is an especiallydisappointing result for . that predicatefigures. of the language predicate The implication of the fact that the definition of satisfactionmust havea ' separateclausefor each predicate in the languageis that Tarski s method .

as Donald Davidson believes it is possible in principle to change (and complicate) . So if. to the language and thereby addedsix unquantified but genuinesentences' Ralph is red' . ' Ralph loves Mary ' and ' Ralph Mary loves Ralph' .160 Chapter 5 in line '. This issuewill be discussedin detail in section 8.1. Tarski doesnot try to define satisfactionor truth for any languageswith . we first take ' n" and ' nil' as variables ranging over namesin the languageand then add this: ' ' ' ' (z = n is red for some n . names Let us seewhat would happen if we tried. Supposewe added a ' ' ' ' . are built up from a finite stock of words. Now we must change the definition to ensure that it will entail the following T -sentences : ' ' is true = Ralph is red Ralph is red ' ' is true = Mary is red Mary is red ' ' Mary loves Mary is true = Mary loves Mary ' ' Mary loves Ralph is true = Mary loves Ralph ' ' Ralph loves Mary is true = Ralph loves Mary ' ' Ralph loves Ralph is true = Ralph loves Ralph To accomplish this. etc. and the object denoted by ' n" loves the object denoted by ' n"' ) This has become the standard way of adding names to a language for which truth is being defined. . and the object denoted by ' n" is red) or (z = ' n' loves n'" for some n' and some n". ' Mary is red'. the definition of satisfactionso as to handle all forms of adverbial or other predicate modification .16 To salvagethe project of defining all semanticterms ultimately in terms of nonsemantic terms. : ' loves Mary '. couple of names say Ralph and Mary . but it would be unacceptableto Tarski as it stands because it makes ' satisfaction' defined partly in terms of the semantic concept of denotation. ' Mary loves Ralph '. we would have to supplementthe definition of ' satisfaction' with a definition of ' denotation ' having a separateclausefor each primitive name in the language : (xXy)[ x denotesy = (x = ' Ralph' and y = Ralph or x = ' Mary ' and y = Mary )] . then it may be possible in principle to create a roughly Tarski-like definition of truth for natural languagesafter all.

Instead of Ralph is red the languagewill have a ' sentencereading (3xXx Ralphizes and x is red) ' (Quine 1970 25). would have to add a clauseto the definition . the very fact that there is such a sentencewill . he thought. He gave severalreasonsfor this. Ralphizes rneans especially since we are not allowed to cash in its rneaning in terrns of denotation or any other sernanticconcept. would consider this a friendly arnendrnent He insists that all tenns that . .1. We can also ' ' ' ' is is say This sentence false . in conjunction with a . and ' That sentence true ' . have at least (see section 8. One of thern is that . 9. So on Tarski 's .21. language This result is general Not only does each primitive predicate and each primitive name require a separate clause so too does every . rneanings and one rnust wonder how well we really understand what ' ' .68). O.3 how Tarski ensuresthat the artificial languagesfor which he definestruth will be sernanticallyopen( 1933 164. .Alfred Tarski ' s SemanticTheory 161 There are a variety of other ways one could createa roughly Tarski -like set of definitions for languageswith names but all of them require that we . narnes like ' Ralph ' and ' Mary '. . ) that natural languagesare. where this refersto the very sentence which it appears But as we shall seein section in . . exarnple beginning is plural . ' That verb is unclear' ' The noun at the ' . But the rnost irnportant reasonis .3) one clause each for every name in the . V. The fact that each predicate is listed separately in the definition of satisfaction rneans that languages with different predicates will have different definitions of satisfaction and hence of truth . 1944 20. Quine suggests that we can havea languagewithout narnesbut with all the advantagesof one that does have narnes Instead of introducing . term of any kind in the language primitive W. 65. on the pattern of the clause for ' is red" for each such new predicate It is questionablewhether Tarski . Tarski did not believe it was possible to define truth for natural . theory of truth for the language entail a contradiction. and 1969 . . . there is no systematicway of detennining for a natural languagewhich of its expressionscount as grarnrnatically cornplete sentences and which do not (Tarski 1969 65. for . hopelessly paradoxical because " " " " . We . Tarski 1933 164 . we introduce the predicates ' Ral' ' ' " ' phizes and Maryizes . on the right side of the equivalencesign be absolutely clear in their appear . languages as distinct frorn the artificial languagesof logic and rnathernat ics (Tarski 1969 68). We shall see in section 9. they are what he calls universal or sernanticallyclosed languages A closed language is one that has the power to describe the sernantically sernanticcharacteristics of its own elernents We can say in English.

. neither a model theorist per senor a physicalist per sehave any particular need to allow beliefs as truth bearers Neither an epistemologist nor a . So there is not one theory of truth . as noted above contemporary textbooks .3. An epistemologistmust acknowledgebeliefsas (one kind of) truth bearer but . of ' true-in -L2 ' .6 Programs and Special Adequacy Conditions I noted in section 5. model theorist per se has need for a theory that reducestruth to physical . the language call it Ll ' for which . however make explicit in the . (s)(s is true in L2 = s is satisfied in L2 by all sequences ).1 that each of the different programs in whoseservice an answer to the metaphysical project can be put imposes special restrictions on what will count as a good answerto the project. of metalogic or model theory no longer avoid using the semantic concept of denotation in the definition of satisfaction and henceof truth .162 Chapter 5 systemthere is not anyone theory of truth . no sentence token true in one languageis false or meaningless in another.) It does not seem that a physicalist per se must make use of recursive . 5. 1944 14. but no one answerto the project hasto be relevant to every suchprogram and thus no one theory of truth need meet all of these special restrictions. truth is being defined: (s)(s is true in L1 = s is satisfied in L1 by all sequences ) For a different language call it L2 ' the definition of truth is . [ 1933 153. I argued in section 2 that Tarski is . . but for the wrong reason He claimed that this limitation results from the alleged fact that a sentence token can be true in one language and false or meaninglessin another .6 . 1969 64] . definition of truth the fact that the definition applies to only one language by explicitly naming in the definition . mistaken in this last claim . we really do not evenhave two theories of truth here What we have is one theory of ' true-in -L1 ' and one . In the end. Indeed. language (Tarski acknowledged this. Tarski has defined not truth but theory truth -in -this-languageand truth -in -that-language The question of whether . . phenomena and so there is no need to avoid defining truth in terms of other semantic conceptsl ? (Thus. there is a different one for every .) We can. this characteristic is a serious objectionto Tarski 's work is taken up in section 6. .

the facts) must mirror the structure of the statements(sentences that we use ) to refer to (describe them? Indeed.lo (b)[ b is true < (3B)(3x)(3y)(3R)(b = ( B. it has beensaid that there are no disjunctive facts . it is not at all clear . x . s ranges over all statements and t ranges over all sentences however . Why must we assumethat the states of affairs (and. techniques Does an epistemologist Would . As we saw in section4.2 that the formulation of Russells theory. :. of thesesorts. But the techniques ' alluded to in note 3 of chapter 4 allow us to extend Russells theory. either there are no disjunctive. ' We saw in section 4. conditional facts or negativefacts.Alfred Tarski ' s SemanticTheory 163 ' . without useof recursion. conditional . all disjunctive and conditional statementswill come out false and so too will all negations . or negativestatesof affairs. :. Either way. if they obtain. . or there are but they never obtain. I have defined a fact as a state of affairs . it is not clear what relevancethis has. . and that statesof affairs (over which x ranges and thus types ) of statesof affairs (r ) can be as complex as you please This is controversial. compounded quantificationally complex. R. But this view was unconvincing for two reasons First . So on this view. and negativestatementsmust be taken as referring to disjunctive. ? .10 (s)[ s is true < (3x)( 3r)( 3t)(t & & is used to make s) (s refers to x ) & (t describesr ) (x is of type r ) & (x obtains)] . . the implication is that on Austin ' $ theory of truth . for example Russells and 's nonrecursivetheoriesof truth servethe 's Austin epistemologist program of evaluating theories of justification (the specialobjections made to each of them in the last chapter aside ? ) In Austin 's caseI think the answeris yes This is clear if we keepin mind . It appears . then. Second even if there are no facts . and/ or syntactically complex they may be. conditional . that neither a correspondenceas-correlation nor acorrespondence as-congruencetheory need make use of recursion to be applicable to all ' statementsor beliefs So Russells and Austin ' s theoriescan be usedin the . preciselybecause is it not the casethat disjunctive. to beliefsof any degreeof complexity. it is part of Austin ' s point that ) statementsare not isomorphic to statesof affairs. that in the formula for his theory. conditional . y) & xRy)] . that obtains. why there cannot be facts of thesesorts. only accounts for beliefs about two-place relations.7. and negativestatesof affairs. .

164 Chapter epistemologicalprogram of evaluating theories of justification ." 18 ) an of The reasonthis caveatis neededis that the MAC would not be credible as a criterion of adequacyfor a theory of truth unlessit is specifiedthat the sentenceof which truth is being predicated be the same sentencethat assertsthat the state of affairs p obtains. and only if. But if the T -sentences would be ridiculous when X and pare not coordinated in this way. because neither makesit possibleto seehow the truth value of a statement(belief) is affectedby the quantifiers contained therein. 5. Which state of affairs? The state of affairs of the table' s being purple.1). X is true if and only if p. this just meansthat p 's coordinationwith X is an additional necessary condition to X 's being true (in addition . specificallyhis schemaT . and thus it would not itself meet the MAC . But notice that neither of them could be put to use in Tarski s program of constructing a model theory for quantified predicatelogic (see section 5. of course Thus for any given true sentence the fact . . or at least the fact that they are not recursive poses no special problems in this ' regard. with his definition of truth . language which the word true refers and X replaced a nameof this by sentence " equivalence theform ( T). so it would not entail the equivalences matching its own form. Moreover. if it obtains. So the instantiation conditions for schemaT specify the way in which X and p must be coordinated. that is. We can seewhy this is so by recalling the instantiation conditions that Tarski embeddedin his text after the displayed schema : X is true if. p. Schema T cannot be a definition of truth .7 The Essence of Tarski ' s Notion of Truth Tarski often complained that his critics confusedhis MAC . that makesit true is the fact that the sentence es express . because has unbound variablesand thus is an open sentenceThis means it . makes ' the table is purple ' true. that it does not really entail anything. We shall call any suchequivalencewith ' p' replaced any sentence the of by " ( " ' ' to . to . So Tarski wants his theory to say that there is one particular state of affairs that . simply binding the variables in the schemawould not produce a definition of truth consistent with :rarski ' s intentions.

section 5. ' The table is ' purple meansthat the table is purple. . " ' puts it this way: Supposethat P is the proposition There are six protons . In other words. it doeS19 The extensional facts are expressedby what are called . however formulated. The easiestway to get this point into a quantified formula is to make it explicit . A definition of truth consisting of just a quantified version of schemaT would not expressTarski ' s insistencethat the X and the p be coordinated in this way. does not entail the T sentences itself. v . M -sentences any sentences are instantiating the following schema : X meansthat p. Arthur Pap. and Keith Lehrero have all made the point that the coordination of truth bearer to the obtaining state of affairs is an necessary condition for truth . wherep is replacedby a sentence and X by the name of that very sentence . ' Bob is a noodle-brained fool ' meansthat Bob is a noodle-brained fool. This is not a coincidence but I postpone further discussionof this matter until . This theory looks strikingly similar in structure to the of ' formula for Russells theory and the simpler formulations of Austin 's. Donald Davidson A. M -sentencesexamplesof which are the following : . (The attentive reader ought have a sense deja vu here. sentencesit declares to be true. O 'Connor. where y rangesover states of affairs.) This theory. One way of doing that would be with this formula : (X ) { X is true = (3p)[ (X says that p) and p] } Those who blanch at quantification over propositional variables might prefer es (X ) { X is true = (3y)[ (X express y) and y obtains] } .Alfred Tarski ' s ~ mantic Theory 165 ' p s obtaining). Quine. ' Snow is white' meansthat snow is white. D . O 'Connor . As Davidson says " What Convention T . and the trite .8. w . English speaker is true if and only if grassis green reveal is that the truth of an utterance depends on just two things: what the words as spoken mean and how the world is arranged" (Davidson 1986 309). like ' " Grass is Green" spoken by an " . but in conjunction with certain undisputed extensional by facts. N . . Prior . O. J.

this . the application of the expressionneed not show up on all correct lists of the necessary and sufficient conditions for the application of the expression . Prior . further that. It just so happensthat in the actual world ' There are six protons in the carbon atom ' doesmeanin English that there are six protons in the carbon atom. and indeed there are six protons in the carbon atom. . . that the words used in stating P should have the meaningsthat they have in English and that the syntax of English is such that those words.22 But this fact plus the precedingequivalenceentails (by ordinary propositional logic) that the following is also true: ' There are six ' protons in the carbon atom is true = there are six protons in the carbon atom This.' What is then required for P to be true further than that there are in fact six protons in a carbon atom? And if nothing further is required how does asserting that P is true differ from asserting that P? . O 'Connor . There doesappear to be somethingmore required for the truth of P. are is just becausein the actual world . of course is just one of the T -sentencesThe sameargument could be ." 21 But Quine. in that order. which is the only world with which claims of extensional equivalenceare concerned English sentences have . madefor all the other T -sentencesSo the T -sentences true. Again. Pap. and Lehrer do not intend to imply that T -sentences false Recall from section 1. namely. Thus it just so happensthat in the actual world ' There are six ' protons in the carbon atom is true wheneverthere are six protons in the carbon atom. the meaningsthat they do. Davidson. be more than one correct extensionalanalysisof an expression and recall . as odd as it sounds this entails that a necessary condition for .166 ChapterS in the carbon atom. John Etchemendyhasmade the samepoint in an enlightening way: Thereis a substantive difference between propositions the . . expressed example : by the followingtwo sentences .3 that there can are . for . What is implied by O 'Connor ' s remark is that one correct extensional "' " ' analysis of There are six protons in the carbon atom is true is ' There are six ' protons in the carbon atom is true = ' There are six protons in the carbon atom ' meansin English that there are six protons in the carbon atom. convey the information content of p .

unlike those versionsof the correspondence whosevagueness theory he complains about. is . world ' Snow is white ' can be false when snow is white.23 We encounteredabove a theory of truth that bearsa striking structural ' similarity to the formula for Russells theory and the simple formulation of Austin 's: (x ) { x is true = (3p)[ (x says that p) and p] } . Moreover . is true. is . His reasonsfor not endorsing them are instructive: I meana definitionwhichwecanexpress the followingwords in : is ( 1) a truesentence onewhich that says thestateof affairsissoandso andthestate . snowmightnot havebeen ' Snowis white' grass ' white though . Tarski does . say if ' snow had meantvanillaice cream(Etchemendy . of affairsindeed soandso . this reason states affairs the of described vary can ' :' independentlySnowis white mighthavebeen . " ' Snow is white' " is might have beenfalse another way of saying that there is somepossible world where ' Snow is white ' is false When he saysthe two statesof affairs . . or in the lesstechnically controversial version . Thefirst of these makes claimthat depends on thecolorof snow thesecond a only . the actual. ( 1) is not something handed down to him by other . ) . can vary independently this meansthat they vary across . wasstill true. 1988 ) ' Note the modal word ' might ' in Etchemendys remark. es For easeof referencelet me call this theoryS . he says that sentences conforming to the T schemaare " completely in accordancewith " it ( 1933 157 .. Indeed. for instance .'s Alfred Tarski Semantic Theory 167 (a) Snow is white ' ' (b) Snow is white. false ' snow hadmeant ' if. thoughsnowwasstill white. claim on the other hand depends both the color of snowand the . To makethis intentionmoredefinite and to giveit a correctfonn. . (x ) { x is true = (3y)[ (x express y) and y obtains] } .. on . He does not say that it is an incorrect analysis of truth . 61 . possibleworlds He does not mean that in this. And conversely . ) The complaints here about indefinitenessand incorrect form are the only explicit complaints Tarski has about ( 1). of meaning ' Snowis white' For thesentence . precisely the taskof a semantical definition (Tarski 1933 155 . briefly consider two theories like S. .

and p " ( 1933 159 . x is a true sentenceif and only if. next considerstreating the quotation marks as a function from sentences to namesof sentencesUnder such an interpretation the " p " in the single . p is one word . for a certain p. so such a naming function would introduce a seman tical concept into the definiensof the theory and 24 thereby spoil his plans. than the letter t is a proper part of the p proper part of the word true. Tarski . What the formula is really saying as Tarski points out ( 1933 . .168 Chapter 5 . We ought to take this as a striking pieceof evidencethat it would capture the essenceif not all the details or the proper form . he does complain that naming is a semantical notion . 160 is that the letter " p " is the only true sentencewhich is absurd. marks in the formula is a proper part of the expression" ' p " ' . Second it doesnot show how the in the definiensfor . If we eliminate those features of his theory and technique relevant only to theseprograms. So the " p " that appears between the single quotation marks in the last ' " quotation cannot be the variable : p that appears elsewhere in the formula. ). .) . While Tarski does not decisively reject such a theory. . structuresof sentences affectstheir truth values But beyond his physicalist and model-theoretic ambitions. ' So it seems that the ways in which Tarski s final theory deviatesfrom an S-like theory are motivated entirely by his physicalist and model-theoretic ambitions. it has the semanticterm ' saysthat complaints ' is true ' . quotation and it is the same variable as the " p " that appears elsewherein the formula. hid the essence his notion So of truth by tacking onto it features motivated solely by ambitions that . like Russelland Austin. ) '" " ' problems concerning the interpretation of the symbol p . He rejects this becauseof certain . Hence on the ordinary " ' '" ' ' " ' '" interpretation. essence Tarski ' s idea of truth per se of of it is that Tarski. The p inside p is no more a '" " . the theory and the recursivedefinition of satisfaction on which it rests would turn into something more or lesslike the formal version of S. x is identical with ' p' . Tarski has no other genuinereasonfor not embracing( I ). of philosophers It is his own statementof the essence his own intentions. a name. truth per se (The ' per se' will be explained below. This is what is meant in sayingthat ( I ) capturesthe . First . Although he does not make them explicit here Tarski has two main ' about the form of ( I ). The other theory like S in structure that Tarski considersis this: " For all x . grammatical . The ordinary interpretation of single quotation marks is that they form the name of some expression that appears in between them. of his conception of .

for Tarski ' s definition satisfies only sentences the MAC . capturesthe translinguistic concept of of truth . no sentence is can . and the latter at least seems presuppose to that every declarative sentencein the actual world is either true or false Again. The reasontheoryS doesnot entail this principle is that it holds that there are two conditions necessary the truth of a given sentence for : the sentence must expressa certain state of affairs. theoryS allows that there are two ways a sentence might fail to be true. Or it may fail to express any state of affairs. Two other points needto be madeabout theoryS . are not strictly necessaryto the fulfillment of the ' ' . . . in which case it does not say anything. world . then. it is this program that requires him to define truth . be neither. and that state of affairs must obtain. But the definition of truth with which Tarski ended up does at least appear to entail the principle of bivalence at least when the principle is taken to be a thesis concerning . Thus. about this. Sinceit was his physicalist program . separatelyfor each language Second theoryS does not entail the principle of bivalence The latter . holds that every declarativesentence either true or false. and the whole theory must show how the truth value of a compound sentenceis affectedby the truth values of its clauses) . the actual. so it does not say anything true or anything false so it is neither . First .Alfred Tarski ' s Semantic Theory 169 though closely related. he simply imposesspecial restrictions on the formulation of the theory : the ' R ' of schema(C) is not to be a semanticrelation. it doesnot have to relativize truth to particular languages This is because unlike the . in which caseit is false(and thus not true). commitment. if he is so committed. to the principle of bivalence(taken to concern only sentences about the actual world ) is a side effect of the extra that forced him to abandon theoryS .5] . . Tarski ' s . two ways it can be not true. basedon schema(C) [ seesection 4.eliminating semanticterms in favor of logicomathematical and physical terms. and hence Tarski 's theory. It may expressa state of affairs that fails to obtain. metaphysicalproject as such (Tarski s casedoesn t parallel that of Russell and Austin in one respect Instead of adding doctrinal matter to a theory . (Seesection 6.1 programmatic goals for a discussion of whether the MAC . theory with which Tarski eventually ended up. theoryS does not need a separateclause defining truth (or some other concept. really does entail the principle of bivalence) .that forced Tarski to reject theoryS . true nor false (and thus it is not true). which express ( ) the essence Tarski 's notion of truth . like satisfaction ) es separatelyfor eachsentence or predicate . Thus theoryS.

26 Karl Popper..28 gives a no to (2). . is that it is. for example with the utilitarian conception ( true . J. JAyer answer it yes the 1 side with the yea-sayerson question ( 1) because following remarks Tarski do not leave room for doubt that he thought of his theory as a by ' " theory : We regard the truth of a sentenceas its correspondence correspondence " aims to with reality ' " ( 1936 404). Wilfred Sel .corresponding with ' ' . and Mark Platts answer them in the affirmative. . concernedexclusivelywith grasping the intentions which are contained in the so-called classical conception of truth (' true . We could possiblyusefor the correspondence : the same purpose followingphrase . whether.. to) with in consists its agreement (or correspondence reality Thetruthof a sentence uponthe latter fonnulationthe tenn (For a theoryof truth whichis to be based ' ' . . or of whatis notthat it is. . his theory is a correspondence theory. Mackle.8 Is the Semantic Theory a Correspondence Theory ? There is ongoing dispute in Tarskian interpretation about 1. an is A sentence true ifit designates existingstateof affairs (Tarski's italics 29 ) Later he confirms that the difference between the latter two italicized formulas in the precedingquotation and his theory is simply that they are vaguer than his.) theory has beensuggested. Elsewherehe says that he .in a certain respectuseful' ) " ( 1933 153 .170 5 Chapter 5. ) to Weshouldlike our definitionto dojusticeto theintuitionswhichadhere the in Aristotelian classical conception truth. L. reality ) in contrast. 27 Gerald Vision lars. and Herbert Keuth defendnegativeanswersto these questions2s while Donald Davidson. or of whatis not that it is not is true of this .intuitionswhichfind their expression of ' : the well knownwordsof Aristotles Metaphysics To sayof whatis that it is not. while Hartry Field and A. isfalse whileto sayof what . . whether he intended his theory to be a correspondence theory of truth 2. Wecouldperhaps by express conception means the familiarfonnula: . All of this es establish that Tarski thought of himself as offering a more precise . SusanHaack. aside from his intentions. " " catch hold of an old notion ( 1944 13). What old notion ? I shall be . not that they say something different.

a correspondencetheory is any theory that instantiates schema(C) by specifyingwhat sort of truth bearer t rangesover. ambiguous and evendoubtfully coherent (Haack 1978 100 112 . still lessis it to ) or the theory. replaces the ' iff' with someequivalencesymbol appropriate to the project at hand. But it is clear that he regards this outcome as only further evidenceof the vagueness the traditional of . theoryS (section 5. ) impreciseis not at all the samething as charging it with dubious coherence (senselessness with ambiguity (more than one meaning). modem correspondenceformulations and hence that his is a correspondence . Haack believes that Tarski meant to reject the " old notion " as " " . correspondenceformulation . which he is sure he has captured. . . not as relevant evidenceto the question of whether the semantic and correspondence conceptions are the same conception of truth (Tarski 1944 32). which express the essenceof Tarski ' s notion of truth per se es . . Does Tarski ' s theory instantiate schema(C) ? Well. . . Indeed. But Tarski does not make either charge3o To call a theory . . . But Tarski might well have failed in his own intentions. and replacesthe ' R ' with some relation appropriate to connecting truth bearersof the specifiedkind with statesof affairs. why would Tarski expressthe desire to capture reject " the actual " meaning of a conception he thought to be meaninglessor ? multiply sensed At one place Tarski concedesthat he is lessthan certain he hascaptured the intuitions behind the correspondence theory. .32). so my affirmative answerto question ( 1) doesnot entail an affirmative answerto ' question (2). Is Tarski s theory a correspondencetheory? As we saw in section4. theory ( 1944 31. precisely because is lessthan certain that it is intended to expressthe Aristotelian he conception.7). certainly does Specifically it makesR the semanticrelation of expressing or saying But. the essence the correspondence of theory is captured in schema c: (t){t is true iff (3x)[(tRx) and (x obtains } )] (C) Accordingly.5. Haack also notes that Tarski sayshe is not surprised to learn that most nonphilosophers responding to a questionnaire rejected the claim that truth is correspondence with reality while most acceptedthe claim that ' It ' is true if and is snowing only if it is snowing. of course it was preciselyfor this reason that Tarski was .'s Alfred Tarski Semantic Theory 171 correspondencetheory. But even here Tarski seems to think it is at leastprobable that the latter is expressedalbeit vaguely by .

affairs. and that is preciselywhat Tarski cannot allow. (This last will becomemore evident in section 6. where I point out that if point confronted with a Tarski -like definition of truth for a languagethat we do not understand we cannottell that it assignsto eachsentence .172 ChapterS forced to reject theoryS . definitions) are placed in. in the . But I think that in the end it is unlikely we will ever establishthat Tarski ' s definition instantiates even this narrower version of the schema because . the definition. kind of semanticrelation. . it does not . if the languageis one we understand we . in the caseof simple languages without quantifiers or open sentenceslist eachsentence the languagein of . by the sentence but this is not made explicit in his definition. Of course we . when the truth bearer is a sentenceit would seemthat R has to be some . Does the theory he actually ended up with instantiate schema(C) ? Strictly .) have quantifiers and open sentences he must separately list . We should perhaps allow that a theory counts as a correspondence theory if it instantiates a language relative version of schema(C) : (t ) { t is true-in -L iff (3x)[ (tRx ) and (x obtains)] } . theory dependson whether we are talking about his notion of truth per se . it . can recognize that he has assignedto each predicate just the property namedby the predicate but this semantic relation is not made explicit in .3. sincea Tarskian definition of truth is relativized to a particular language while schema(C) is not. Hence Tarski has to define truths so as not to make explicit the relation between a true sentenceand an obtaining state of affairs. just the state of affairs expressedby that sentence For more complex languagesthat . definition of satisfaction for the language each predicate of the language . The former is a corre- . And this in turn meansthat he must. be thought that this feature of Tarski ' s work should not be allowed might to affect what generic category his definition (strictly . . and assignit a property. where t is a sentence statement utterance or some other sentencelike . question separatelyand simply assignit to a state of recognizethat he assignsto eachsentence just the state of affairs expressed . given his theoryS extra physicalist and model-theoretic ambitions. or about the theory to which he was forced to resort. So whether we classify Tarski 's theory of truth as a correspondence . item and the ' R ' holds the place of some language relative linguistic relation appropriate to connecting the linguistic item to a state of affairs. given his ambition to eliminate semanticconceptsfrom the definiensof a definition of truth . Again. Still .

Alfred Tarski 's SemanticTheory 173 spondence theory . is not a competitor with the correspondence theory in the sense that it conflicts with the latter . however . and it is the truth of beliefs that are of ultimate concern to epistemologists Another . is not relativized to particular languages and is unsuitable for the physicalist and model . definiens This in turn effectively requires that he define truth in tenDs of satisfaction which is itself separately defined for every predicate of the . . for any given language cannot explicitly capture the essence of . He wants to reduce all semanticconceptsto physical and logico-mathematical conceptsand. 5. Tarski ' s theory should be thought of as a correspondence theory depends on exactly what we are referring to with the phrase ' Tarski 's theory ' : theoryS or the theory with which Tarski finally ended up. I turn now to some of the major objections that have been made to Tarski 's theory. we saw that the question of whether . his own notion of truth per se Finally . Thus there must be a separate definition of truth for each language and the worry arises that he can .9 Chapter Summary ' Understanding Tarski s theory requires an understanding of the broader programs in serviceto which he wished to put it . but it is very bothersomefor those who would put the metaphysicalproject to work in the serviceof epistemology . Tarski was not bothered by this. but the latter is not . sideeffectof his banishmentof semantictenDSfrom the definiensis that his definition .theoretic programs . They are quite compatible . by . doing so. a correspondence theory . The differences between them are simply that theoryS . For no languagewith only a finite number of predicatescould ever express every belief that someoneor other might have in the future. Even the theory he ended up with . languagefor which truth is being defined. to make semanticsa scienceThis goal forcesTarski to reject any definition of truth in which an unreduced semantic tenD appears in the . define truth only for artificial languageswith a finite number of predicates .

Sign up to vote on this title
UsefulNot useful

Master Your Semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master Your Semester with a Special Offer from Scribd & The New York Times

Cancel anytime.