PART ONE
MEANING AND KNOWLEDGE
(The John Locke Lectures, 1976)
LECTURE I
The nature of truth is a very ancient problem in philosophy; but not until the present century did philosophers and logicians attempt to separate this problem from the problems of the nature of knowledge and the nature of warranted belief. That one could have a theory of truth which is neutral with respect to epistemological questions, and even with respect to the great metaphysical issue of realism versus idealism, would have seemed preposterous to a nineteenth-century philosopher. Yet that is just what the most prestigious theory of truth, Tarskiās theory,1 claims to be.
Although it requires a certain amount of sophisticated logic to present this theory properly, one of the leading ideas, the idea of ādisquotationā, is easy to explain. Take any sentence ā say, Snow is white. Put quotation marks around that sentence ā thus:
Now adjoin the words āis trueā ā thus:
āSnow is whiteā is true.
The resulting sentence is itself one which is true if and only if the original sentence is true. It is, moreover, assertible if and only if the original sentence is assertible; it is probable to degree r if and only if the original sentence is probable to degree r; etc. According to Tarski, Carnap, Quine, Ayer, and similar theorists, knowing these facts is the key to understanding the words āis trueā. In short, to understand P is true, where P is a sentence in quotes, just ādisquoteā P ā take off the quotation marks (and erase āis trueā).
E.g. what does
āSnow is whiteā is true
mean? It means
What does
āThere is a real external worldā is true
mean? It means
There is a real external world
And so on.
The claim that ādisquotationā theorists are advancing is that an answer to the question what does it mean to say that something is true need not commit itself to a view about what that something in turn means or about how that something is or is not to be verified. You can have a materialist interpretation of āSnow is whiteā; you can believe āSnow is whiteā is verifiable, or that it is only falsifiable but not verifiable; or that it is only confirmable to a degree between zero and one; or none of the foregoing; but āSnow is whiteā is still equi-assertible with āāSnow is whiteā is trueā. On this view, ātrueā is, amazingly, a philosophically neutral notion. āTrueā is just a device for āsemantic ascentā ā for āraisingā assertions from the āobject languageā to the āmeta-languageā, and the device does not commit one epistemologically or metaphysically.
I shall now sketch the second leading idea of Tarskiās theory. āTrueā is a predicate of sentences in Tarskiās theory; and these sentences have to be in some formalized language L, if the theory is to be made precise. (How to extend the theory to natural languages is today a great topic of concern among philosophers and linguists.) Now a ālanguageā in this sense has a finite number of undefined or āprimitiveā predicates. For simplicity, let us suppose our language L has only two primitive predicates ā āis the moonā and āis blueā. For predicates P, the locution
whose intimate connection with the word ātrueā can be brought out by using the phrase āis true ofā instead of ārefers toā, thus:
can also be explained by employing the idea of disquotation: if P is the predicate āis the moonā we have:
āIs the moonā refers to x if and only if is the moon.
And if P is the predicate āis blueā we have:
āIs blueā refers to x if and only if x is blue.
So the āmeta-linguisticā predication:
āIs the moonā refers to x
is equivalent to the āobject-languageā predication:
Let us say P primitively refers to x if P is a primitive predicate (in the case of our language L, āis the moonā or āis blueā) and P refers to x. Then primitive reference can be defined for our particular example L by giving a list:
Definition:
P primitively refers to x if and only if (1) P is the phrase āis the moonā and x is the moon, or (2) P is the phrase āis blueā and x is blue.
And for any particular formalized language a similar definition of primitive reference can be given, once we have been given a list of the primitive predicates of that language.
The rest of Tarskiās idea requires logic and mathematics to explain properly. I shall be very sketchy now.
The non-primitive predicates of a language are built up out of primitive ones by various devices ā truth functions and quantifiers. Suppose, for the sake of an example, that the only devices are disjunction and negation: forming the predicates āP or Qā, and ānot-Pā from the predicate P. Then we define reference as follows:
(I) If P contains zero logical connectives, P refers to x if and only if P primitively refers to x.
(II) P or Q refers to x if P refers to x or Q refers to x.
(III) Not-P refers to x if P does not refer to x.
Turning this inductive definition2 into a proper definition is where much of the technical logic comes in; suffice it to say this can be done. The result is a definition of āreferenceā for a particular language ā a definition which uses no semantical words (no words in the same family as ātrueā and ārefersā).
Finally, supposing that our simple language is so simple3 that all sentences are of the forms for every x, Px, for some x, Px, or truth-functions of these (where P is a predicate), then true would be defined as follows (of course, Tarski actually considered much richer languages):
(I) for every x, Px is true if and only if, for every x, P refers to x.
(II) for some x, Px is true if and only if, for some x, P refers to x.
(III) if p and q are sentences, p or q is true if p is true or q is true; and not-p is true if p is not true.
While I have left out the mathematics of Tarskiās work (how one turns an āinductive definitionā like the above into an āexplicitā definition of the form āsomething is true if and only if ā¦ā, where ātrueā and ārefersā do not occur in āā¦ā) and I have ignored the immense complications which arise when the language has relations ā two-place (or three-place, etc.) predicates ā I hope I have conveyed three ideas:
(1) āTruthā and āreferenceā are defined for one particular language at a time. We are not defining the relation ātrue in Lā for arbitrary L.
(2) Primitive reference is defined āby a listā; and reference and truth in general are defined by induction on the number of logical connectives in the predicate or sentence, starting with primitive reference.
(3) The āinductiveā definition by a system of clauses such as (I), (II), (III), (Iā), (IIā), (IIIā), can be turned into a bona fide āexplicit definitionā by technical devices from logic.
As a check on the correctness of what has been done, it is easy to derive the following theorem from the definition of ātrueā:
āFor some x, x is the moonā is true if and only if, for some x, x is the moon.
And in fact, one can derive from the definition of true that
(T) āPā is true if and only if P
when the dummy letter āPā is replaced by any sentence of our language L.
That this should be the case ā that the above schema (T) be one all of whose instances are consequences of the definition of ātrueā ā is Tarskiās āCriterion of Adequacyā (the famous āCriterion Tā) for definitions of āis trueā.
Notice that while the idea of disquotation may initially strike one as trivial, Tarskiās theory is obviously very non-trivial. The reason is that the idea of disquotation only tells us that the Criterion T is correct; but it does not tell us how to define ātrueā so that the Criterion T will be satisfied. Nor does disquotation by itself enable us to eliminate ātrueā from all the contexts in which it occurs. ā Snow is whiteā is true is equivalent to Snow is white; but to what sentence not containing the word ātrueā (or any other āsemanticalā term) is the following sentence equivalent: If the premisses in an inference of the form p or not-p/:.q are both true in L the conclusion is also true in L? Tarskiās method gives us an equivalent for this sentence, and for other sentences in which āis trueā occurs with variables and quantifiers, and that is what disquotation by itself does not do.
There are many problems with Tarskiās theory which bothered me for a number of years. One problem, which does not seem to me to be too serious, is that ātrueā is taken as a predicate of sentences (i.e. strings of written signs) ā strictly speaking, what is being analysed is not āis trueā, as a predicate of statements, but āexpresses a true statementā. But I take it that, although it is contrary to ordinary usage to speak of sentences as true or false, this usage is perfectly clear, and also this amount of deviation from ordinary usage is probably inevitable in any reconstruction (e.g. the use of the word āstatementā).4 A more important objection is that the theory does not allow for sentences which are neither true nor false (e.g. āThe number of trees in Canada is evenā), or sentences containing indexical words (such as ānowā, āhereā, T). In fact, the theory can only be applied in its canonical form to languages in which every predicate is well defined and non-indexical. It now seems to me that it can be modified to apply to languages of other sorts; but this is not the sort of problem I shall deal with in these lectures. I shall also not distinguish between Tarskiās original theory and a recent and very elegant variant proposed by Saul Kripke. The criticism I wish to deal with in detail ā one with which I at one time agreed ā is due to Hartry Field.5
Field concedes that Tarski did accomplish something of philosophical importance in showing how to define the semantical notions of reference and truth in terms of the semantical notion of primitive reference. But Tarski was wrong, Field contends, in thinking that he had philosophically clarified primitive reference. Here is the crucial paragraph from Fieldās article:6
Now, it would have been easy for a chemist late in the last century, to have given a Valence definitionā of the following form:
(3) (āE) (ān) (E has valence n = E is potassium and n is + 1, or ā¦ or E is sulphur and n is ā 2)
where in the blanks go a list of similar clauses, one for each element. But, though this is an extensionally correct definition of valence, it would not have been an acceptable reduction; and had it turned out that nothing else was possible ā had all efforts to explain valence in terms of the structural properties of atoms proved futile scientists would have eventually had to decide either (a) to give up valence theory, or else (b) to replace the hypothesis of physicalism by another hypothesis (chemicalism?). It is part of scientific methodology to resist doing (a) as long as the notion of valence is serving the purposes for which it was designed (i.e., as long as it is proving useful in helping us characterize chemical compounds in terms of their valences). But the methodology is not to resist (a) and (b) by giving lists like (3); the methodology is to look for a real reduction. This is a methodology that has proved extremely fruitful in science, and I think we are giving up this fruitful methodology, unless we realize that we need to add theories of primitive reference to T1 or T2 if we are to establish the notion of truth as a physicalistically acceptable notion.
What Field contends is that the Tarski definition of primitive reference (in our example, P primitively refers to x if P is the phrase āis the moonā and x is the moon or P is the phrase āis blueā and x is blue)...