Davidson: Radical Interpretation

The Learning Zone Reading Groups
1

This is the next essay in an expanding series of expositions of the philosophy of Donald Davidson. The article is available at Jstor, or Common Ground. It follows closely on from the exposition on Truth and Meaning, in which Davidson sets out a methodical approach to interpreting the meaning of sentences in some natural language using an extensional, formal first-order language to make explicit the conditions under which a sentence of the natural language would be true. He proposes doing this by generating a T-sentence for a given utterance in the natural language. The foremost difficulty with such an approach is ensuring that the sentence of the formal language captures the meaning of the sentences in the natural language. That is the subject of this next essay:

…given a theory that would make interpretation possible, what evidence possibly available to a potential interpreter would support the theory a reasonable degree?

By way of example, given that Kurt utters “Es regnet”, and the theory that Kurt said that it is raining, what evidence is available to us that supports this theory?

Davidson notes that the issues here are not only between languages, but also within them. We might well ask how, if Kurt had instead uttered the sounds “It is raining”, what evidence might support the theory that Kurt said that it is raining. Roughly, the question is that, given those noises and some idea about what was being said, what evidence have we of what was meant?

Notice that we cannot simply use Kurt’s beliefs, intentions, or the meanings of Es regnet as evidence, because those are precisely what the radical interpreter is trying to discover. Whatever evidence we use to justify our interpretation must be publicly observable.

Which takes us to the top of the second page of the essay.

Davidson: Truth and Meaning
2

On the one hand it won’t do to explain interpreting “es regnet” by saying it means the same as “it’s raining” - that’s just repeating the problem. On the other hand it won’t do to say “es regnet” is just noise. It is indeed noise, but not just noise. There’s something in the middle. Davidson considers and rejects a few options: “But such theories fail to reach the evidence, for it seems clear that the semantic features of words cannot be explained directly on the basis of non-linguistic phenomena”. We need to consider utterances in their context, and find our evidence there, in both the “interests and activities” in which they occur.

Similarly a fine and discriminating analysis of the many intentions with which a sentence might be used might only be helpful if it includes the agent’s intentions and beliefs.

All this by way of pointing out that what we do with words is deeply embedded alongside the other things we do, and so in what we take to be the case: in our beliefs and intentions.

And this gets us towards the end of p. 315.

1 Like
3

Seeing what is happening in the paragraph that crosses pages 315-16 needs some background. I think the argument might be filled out somewhat as follows.
The interoperation here is for the whole of some language, nothing left out. That is, the interpreter must be able to interpret any sentence constructed in that language. In order to be knowable, what we might know in order to do this must be finite. That’s fine, since for any given language there is at most a finite vocabulary and other resources. So far so good. But if we were instead to look for a way to interpret any possible language, our resources would become innumerable, and hence unknowable. The base on which our interpretation might be built would become unfixed, and unusable. Hence, it would make no sense to ask for any sort of interpretation here.

This is much the same point as that made early in Truth and Meaning, that we must keep to a finite, if quite large, set of linguistic resources.

4

The next point is one of triviality; interpreting utterances in one’s own language doesn’t require a theory as such, since such would consist of tautological sentences such as “‘It is raining’ is true IFF it is raining.” But in radically interpreting “‘Es regnet’ is true IFF it is raining,” one cannot make use of that sort of recognition. The evidence cannot consist in terms of meaning, translation, synonymy, and so on, since these already presume some level of interpretation. A different sort of evidence is required.

Next, Davidson is moving away from Quine’s notion of Radical Translation. What will not do is a theory that gives something like a one-to-one equation between the terms used in two languages. The difference here is that between the truth sentence “‘Es regnet’ is true IFF it is raining” and the equivalence “‘Es regnet’ = ‘It is raining’.” The T-sentence is reliant on the semantics of both languages in a way that a mere equivalence is not.

5

Thanks for this. I found it much easier to understand than the actual article.

This is really just a comment. You won’t be surprised that this chimes well with my approach. But I don’t find much trace of it in his more detailed discussion. One could be forgiven for thinking that he is treating language as a distinct domain.

1 Like
6

FWIW, I think Sellars nailed this, adding some meat to “meaning is use.” “It’s raining” plays the same-enough role in one community that “es regnet” plays in another.

I like that you emphasize that is “noise.” It is “arbitrary noise.” We need some noise or other, but many noises will do.

What matters is the way that “noise” functions like a “machine part” in the community understood as an organism. This community of primates needs food, water, shelter. The signs only have meaning in the context of helping us deal with non-sign objects that we primarily care about.

To me it looks hopeless to discuss “meaning” as if something that is in the sign system ripped out of its visceral context.

7

This last point, that more is needed than a mere translation, is teased our over the next page or so. Davidson is very much placing the utterance in a context, as what others might call a part of the “form of life”, not as a “distinct domain” and within it’s “role” in a language game. Radical interpretation is anthropological, not linguistic.

8

Huh, I have the reverse opinion. Sellars’s “role”-semantics is too wooden, I think, to take account of language use. Sellars’s seems to think that we can identify the role of a given word as a stable pattern of use, which is then projected onto a community of language users as the meaning of a term. Davidson, by contrast, by putting interpretation at center stage, can accommodate the fact that there is no such stable meaning, no such stable pattern of use, and that we negotiate the meaning of our words at each new encounter.

2 Likes
9

As you describe this “negotiation,” I can only agree. I’m with Bakhtin on this issue. We are primates holding up flashcards. A particular primate can do new things with an old sign.

But we have to be vaguely trained into an “average use” in order to play against this average use. So to me it looks like a tension between the average and the long term with the now of the dialogue.

For me the fundamental breakthrough is the “role” metaphor. I don’t assume that Sellars invented this, but I happened to pick it up from him. It gels with the work of Derrida. Basically we just have signs ( the signifiers ) and don’t need “signifieds” as mysterious essence crystals. We enact the “meaning” of signs. But signs as perceptual objects are also “experienced” in their “significance.” I can only express ( in this context ) this experienced significance by holding up more flashcards.

10

I’ve re-written the next bit several times, hence the delay. I was attempting to bring the next few pages together in one example, but now think that the approach Davidson uses will work best.

So to page 318, and the paragraph beginning “What characterises a theory of truth in Tarski’s style is that it entails…”. This is a synopsis of Tarski’s work, on which Davidson builds. Best to break it down, but that’s no simple task.

What Tarski is after is a theory of truth that entails, for every sentence s of the object language, a sentence of the form:

s is true (in the object language) if and only if p

These are what Davidson, and now most other folk, call T-sentences.

They are generated by replacing “s” with a canonical description of s, and “p” with a translation of s.

A canonical description of s is not simply a restatement, not “the cat is on the mat”, although this is often the way the idea is introduced. It’s closer to an individual utterance, more: “The cat is on the mat” as spoken by Banno last Saturday at half past ten.

Satisfaction is the key undefined semantical notion in the theory — it relates sentences (open or closed) to infinite sequences of objects belonging to the range of the object language’s variables. It is used to explain truth.

The axioms (finite in number) are of two kinds:

  • those giving the conditions under which a sequence satisfies a complex sentence, on the basis of the satisfaction-conditions of simpler sentences
  • those giving the conditions under which the simplest (open) sentences are satisfied

Truth is then defined for closed sentences in terms of satisfaction.

For “The cat is on the mat” as spoken by Banno last Saturday at half past ten, the axioms used might be along the lines of:

  • “the cat” (as spoken by Banno last Saturday at half past ten) refers to y iff y is the cat demonstratively indicated by Banno last Saturday at half past ten.
  • “the mat” (as spoken by Banno last Saturday at half past ten) refers to z iff z is the mat demonstratively indicated by Banno last Saturday at half past ten
  • a sequence ⟨y, z⟩ satisfies “α is on β” iff y is on z.
  • present-tense “is on”, as spoken at t, is satisfied at the time of utterance t, not the time of interpretation; we are evaluating the sentence after the fact, on the report of it.
  • “the cat is on the mat” (as spoken by Banno last Saturday at half past ten) is true iff the referent of “the cat” (as spoken by Banno last Saturday at half past ten) is on the referent of “the mat” (as spoken by Banno last Saturday at half past ten), last Saturday at half past ten.

Notice how the axioms each take a piece of the object sentence and give it a value in the metalanguage. Together these set out the conditions under which the object sentence is true, in the terms of the metalanguage.

Notice also that “truth” here makes use of satisfaction, and so is extensional. And again, while Tarski explained truth in terms of satisfaction, and so took meaning as granted, Davidson is going to take truth in terms of satisfaction as granted, and use it to set out the truth conditions for sentences in the object language, and hence, in his view, the meaning of those sentences: after all, if you know when a sentence is true, what more is required of a theory for the meaning of that sentence?

11

Onward, then, to the three questions. First, can we give such a theory of meaning for a natural language?

Davidson considers two cases, the first for our own language, with the interpretation already understood, the second for a “foreign” language, with at first no interpretation.

In the simple case, p, being its own interpretation of s, can simply replace s; “p” is true IFF p. And the point here is that there can’t be any ‘machinery’ in p that is not also in ‘“p” is true’:

Unless the original sentence mentions possible worlds, intensional entities, properties or propositions, the statement of its truth conditions does not.

Where there is no interpretation, one must be constructed. The path Davidson suggests is to interpret parts of the object language, and extend these to the remainder. The hope is to set out the logical form of all sentences. This project flourished for a number of years, however more general approaches developed that allowed better formalisations of parts of language without the restriction to an extensional first-order logic.

12

And so to the second question, and this:

What I propose is to reverse the direction of explanation: assuming translation, Tarski was able to define truth; the present idea is to take truth as basic and to extract an account of translation…

If we know what makes an utterance true, we have understood the utterance. So a theory that makes explicit the circumstances in which an utterance is true amounts to an account of the meaning of that utterance. That is, if for each sentence in the object language we have a sentence in the metalanguage setting out its truth, we have an account of meaning for the object language—an interpretation.

A speaker holds a sentence to be true because of what that sentence means and what the speaker believes. Hence belief and meaning are interdependent. To hold a sentence as true is to adopt an attitude towards that sentence, the same attitude towards all such sentences: that they are true.

So to the example. Kurt utters “Es regnet”. We hypothesize that “Es regnet,” as uttered by Kurt at a given time and place, is true if and only if it is raining at that time and place. The evidence we have is Kurt’s membership of a German-speaking community, and that it is raining at the time and place at which Kurt uttered “Es regnet”.

The approach then is to place the frame of a first-order predicate logic—those individuals, individual variables, predicates, truth operators, quantifiers, and identity—“over” our object language; by identifying the individual terms, predicates, and so on within the object language; the next step is to make sense of indexicals; the last step is to treat of what remains.

There are the formal constraints on the theory, given by the mechanism of the T-sentence; and there are empirical constraints, given by what we observe amongst the users of the object language. Our hypothesis is up for review on the basis of that evidence. In order to perform such a review, we are obliged to suppose that what the folk who speak our object language hold to be true is much the same as what we hold to be true; that they believe it is raining under much the same circumstances as we believe it is raining.

We interpret in such a way as to maximise agreement. Here we find the argument that would later be expanded in On the Very Idea of a Conceptual Scheme:

If we cannot find a way to interpret the utterances and other behaviour of a creature as revealing a set of beliefs largely consistent and true by our own standards, we have no reason to count this creature as rational, as having beliefs, or as saying anything.

This claim is methodological, a constraint on when we might count someone as rational. Hence, it might be that dolphins, for example, are rational, but hold beliefs so different from our own as to be untranslatable; and on that basis, while we do not have grounds to claim they are rational, we also do not have grounds to claim that they are not.

13

To the last section, then, and a few comments about what Davidson is not doing.

The first is that the T-sentence by itself does not give the meaning of the sentence of the object language. ‘“snow is white” is true IFF grass is green’ is true, since snow is indeed white and grass is indeed green. If all that were required in order that our theory give the meaning for “Snow is white” were the truth of the T-sentence, we could not rule out such oddities. Hence the theory requires more than just the T-sentence.

Nor will the canonical proof suffice, since again we might have a proof that ‘“snow is white” is true IFF grass is green’ is true, and have entirely missed the meaning of “Snow is white”.

Nor will it do to appeal to all the sentences so constructed, since knowing the truth value of other sentences does not help us with knowing the truth value of a particular sentence. Knowing when “Snow is white” is true does not tell us when “grass is green” is true.

We cannot assume that a T-sentence satisfies the translation criterion. It has to be shown. Tarski was able to make just that assumption, by stipulating the structural equivalence between the terms of his metalanguage and his object language. In a radical interpretation, we cannot make this assumption, but instead must treat it as an empirical issue, for ongoing examination and potential rebuttal.

So if we did theorise that “Es Regnet” is true IFF it is sunny, and so that “Es Regnet” means that it is sunny, our theory remains open to critique on the basis of additional information. We will find cases in which it is clear to us that it is not sunny, and yet the community under examination appear to hold “Es Regnet” to be true.