PHL313: Notes on Davidson, "Truth and Meaning"

Some remarks on Davidson, "Truth and Meaning"


We're going to focus on two kinds of theories of meaning: truth-based theories (including modal or possible-worlds theories) and use theories.

There are other kinds of theories. Examples include ideational theories, or reference theories, or verificationism.

Let me make a distinction between bottom-up theories, which explain meaning by beginning first with an explanation of the meaning of small parts of a language (usually words) and then other meanings are composed of those parts; and top-down theories, which usually begin first with an explanation of the meaning of large utterances (usually sentences) and then explain the meaning of the parts of language in terms of how they play a role in these larger utterances.

A bottom-up ideational theory would typically explain meaning with the hypothesis that words are labels for ideas. A top-down ideational theory would explain meaning typically with the hypothesis that sentences name propositions. Either approach is widely considered unhelpful because it seems then that we have merely pushed the problem of explaining the meaning of language back to being a problem of explaining the meaning of thoughts or propositions.

Verificationism, as we discussed, is largely an epistemlogical view. Applied to meaning, it holds that utterances have a meaning either because they are pure logic or they are determined by the kinds of effects they could possibly have if true. Utterances that we could never confirm as true or false are meaningless. (Actually, there are some details I'm skipping here, but they won't change our conclusion.) There are many problems with this view, one of which is that it appears too demanding. The very claim itself, for example, fails the verificationist criterion.

We saw in our discussion of reference why reference alone is not sufficient for meaning. The only solutions we were offered for Frege's two puzzles, and Russell's two puzzles, were both dependent upon terms having not only reference but also meaning. Kripke does not deny this -- he merely proposes that reference is not itself determined by meaning. Thus a top-down theory (e.g., sentences refer to states of affairs, or Frege claims they refer to the true or the false), or a bottom up theory (e.g., words point at things or relations between things) are both insufficient.

Truth-based theories are built on the idea that the truth conditions of a claim determine its meaning; thus, to know what a sentence means is to know what it is for that sentence to be true. Use theories are built on the idea that meaning is determined and expressed only in the use of language.

Davidson's "Truth and Meaning"

OK, friends, don't tell anyone I said this -- don't post it on the web or anything -- but I think this paper is terrible. I apologize for requiring you to read it. Even Davidson seems to have known it was terrible; he keeps referring to this paper inside the paper, as if trying to explain it to himself and us. I especially appreciated his observation 1/3rd of the way through that "the path to this point has been tortuous." Like we hadn't noticed. But, the paper was very influential, so we're kind of stuck with it.

I'll follow the article in roughly its order, to be of some use perhaps in reading it. Hopefully with my notes you can beat some sense out of this thing (I can't promise my notes would have been acceptable to Davidson, by the way).

Davidson's paper takes roughly the following form:

The Tortuous Background and Sentence Meaning

Davidson begins by asking whether a theory of meaning should explain the meaning of a sentence in terms of the reference of its parts.

Frege proposed once a view in which the meaning of "Two is even" would be constituted by both the reference of "two" and the reference of "is even" being put together. You had to assume that the reference for "two" and "is even" are fixed. But at least in theory you could find a finite number of elements to your language which you explain with your theory of reference or just assume they have a reference, and then you could show how you could build infinitely many sentences out of these things.

Some thought there was a bad regress here. The idea of the regress concern was that to say
the meaning of 'two is even' is given by putting together 'two' and 'is even'
depends upon "putting together" having a meaning, which you cannot explain alone. You'll need to use your theory of meaning to explain what it means to put them together, and then again to explain your explanation, and so on. I confess I don't find this very threatening, but Frege addressed this by calling predicates like "is even" unsaturated, his techincal term for them being in need of a term to be properly used. This is, I think, to take "putting together" as a primitive; Davidson says this was to just restate the problem (which is, I guess, equivalent to complaining that it means you take the solution as primitive).

More importantly, Davidson argues that functions and presumably predicates are not objects. We don't need a referent for things like "the father of".

What about focussing on the sentences? Frege in fact claimed that the meanings of the parts of sentences arise solely from their role in sentences.

Here there is a problem: Frege proposed that the referent of true sentences was the true. Then all true sentences have the same referent! If meaning is somehow made out of reference, and if we adopted this theory, this fails (2+2=4 and 4+4=8 do not have the same meaning, obviously).

So, the obvious move, if we're to keep something like the idea that meaning starts with sentences, is to drop the notion that the meaning is constituted by reference. We'll use Frege's distinction between sense and reference (meaning and reference); and use some basic meanings instead of basic references to explain meaning.

The problem, however, according to Davidson, is that we don't have a good account, or even a start of an account, of how the meaning of sentences is composed of the meaning of words (there is nothing like the story above explaining the meaning/reference of "Two is even" in terms of the reference of "two" and "is even").

We want something like "two" means x and "is even" means y, so we can analyze and explain how "two is even" has meaning and what that meaning is.

Davidson also claims we can't assume meanings of the parts. He is concerned that we will use the meanings of words to fix their reference, and so mix the meaning and reference theories up again. (He also claims meaning can't "do work" in the meaning theory. This is really obscure; I think his point here is that it won't explain anything about meaning if we start by assuming meanings.)

Nor, Davidson claims, will a theory of syntax help. If we had a syntax for a language, and also the meaning of its smallest parts as given, you would think the two would give us the meaning of sentences. He says this is not so because there are sentences like:
Tom believes that Malcolm X was tall.
(my example, not his) that cannot be analyzed with some syntactic analysis. That is, this sentence has a meaning not at all like, "Tom runs to his house," thought they would seem to be similar in syntax (Subject verb object). But otherwise Davidson does not defend his claim that syntax is insufficient.

Convention T to the Rescue

What we need, Davidson says, is a recursive theory that will tell us some property that sentences have which relates them one-to-one to some other sentences that we can identify as their "meaning." He says Tarski's convention T can do this. Convention T is a criterion for any theory of truth; Tarski argued any such theory must for every sentence P allow
"P" is true if and only if P
Davidson wants to run with this, and say a theory of meaning can be built on the idea that it will provide a series of sentences of the form
"P" is true if and only if s.
Where s is not the sentence (as it is in Tarski's Convention T), but some kind of analysis of the sentence.

One advantage that Davidson sees is that this approach does not refer to meaning. Meaning is revealed by this theory, but not taken as a primitive.

How would this work? Unstated, but in the background, Davidson is assuming something like Frege and Tarski's logical work. In a standard logic with a semantics (a "model"), you have the smallest basic elements of the language, and then you specify both their meaning or referents, and how they combine with other elements to form meaningful larger formulas (typically, sentences). These rules for combining elements are "recursive" -- which means you can apply them again and again, and form more and more large sentences.

Here's a simple example. Let P and Q be any sentences. Then, we have a recursive syntax that says (P and Q) is a sentence. And a recursive semantics that says that if P is true and if Q is true then (P and Q) is true. Note now that you can make infinitely many sentences, and you can tell what is required for them to be true. For example: ((P and Q) and (P and Q)) is to apply the syntactic rule twice. And so on.

Davidson is imagining that we would undertake something like this for English. We would have a theory of how some basic, smallest parts combined (the syntax) and when those combinations were true or false (the semantics), and this would be a recursive theory so we would then be able to determine for any sentence of this language when it was true (given the truth values of its smallest parts).

This is a huge task, one on which little progress has been made. But Davidson is claiming only that it is possible and it would be a way to understand a theory of meaning.

Confronting the Problems with a Logical Theory of Language

But there are problems. Tarski, the great logician that inspires all this, thought that it was very unlikely we'd ever get a decent analysis of a natural language with this approach. Two things in natural langauges made him worry: (1) contradictions and semantic paradoxes; (2) confusion and vagueness and ambiguities. Davidson also adds (3) indexicals.

(1) Contradictions and paradoxes are easily said in a natural language. Consider:
This sentence is false.
It is meaningful, surely, but if its meaning is based on its truth conditions, what does it mean? If it's true it's false, if it's false it's true!

Davidson says we can just rule out those features of a language that allow for this kind of thing. You can do that in formal languages and in fact we have done it in many formal languages.

(For what it's worth, I think Davidson is way too quick here. The way that you rule out in a logic some paradoxes like the one above is largely by weakening the language very strongly. He says, for example, we'll not allow "True in this language" to be part of our formal specification of the meaning for the language. Note then that you would not be able to extend his theory of meaning to his theory of meaning!)

(2) Davidson argues that the vagueness and ambiguity we can hopefully handle by letting it go all the way through. So, if I say:
"P" is true if and only if s.
And part of P is ambiguous, I'll have it so that the corresponding part of s is ambiguous.

(3) For indexicals, such as sentences where the truth conditions depend upon a time and place or the speaker ("It is raining" or "I am tall"), he suggests we might put time and place and speaker into the truth specification.
"P" is true as spoken by A at place xyz at time t if and only if s is true of A at place xyz at time t.

Other concerns Other reasons that we might have concerns about a truth-based theory of meaning are that (4) it does not in any obvious way handle nondeclarative sentences; (5) it is not clear how one handles possibility and other modal notions; (6) it is in trouble for co-extensive terms.

Defenders of a truth-based theory of meaning have offered several answers to these problems.

The boldest answer to (4) is to claim that all language can be understood in terms of truth conditions. Thus, if Jones says, "I do" at her wedding, we must analyze this as somehow true or false. For example, it might be analyzed to mean, "Jones sincerely intends to act as this oath requires" -- or somesuch analysis. You'll have to ask yourself whether you think it is plausible that sentences like these can be properly understood as sentences with truth conditions: Problem (5) concerns the fact that it is not clear how you analyze something possible in terms of truth value alone. The sentence "Nixon could have lost the election" might be true, but how could that truth be given in the relations of its parts? It seems that possibility is a property of whole sentences, not a part of a sentence that can be analyzed by itself. Most philosophers think you need to take notions of possibility and necessity as primitive. The solution to (5) that is most commonly adopted by the truth-based theorists is to actually add these in. This is a big benefit because it handles problem (6).

Problem (6) concerns a fact we discussed when talking about reference. We noted that many predicated can be coextensive. Perhaps all the things that have a liver also have a kidney, and vice-versa. Suppose this is true. Then "___ has a liver" is coextensive with "___ has a kidney" -- which is to say, these predicates are true of all and only the same things. Here then, the truth conditions for both properties would be the same, but their meaning is different.

One Davidsonian answer is to say that the analysis of "Jones has a liver" and of "Jones has a kidney" will be different. That means the stuff on the right side of our biconditional is different in each case, maybe even in a way that is more than just replacing "liver" with "kidney." I find this an unconvincing move, but ask that you note we would then have meaning being more than just truth-conditions; meaning would also include syntactic analysis.

Problem (6), the problem of coextensive predicates and other elements of language, is the main inspiration for an alternative version of the truth based theory: the possible-worlds theory of meaning.