Peter Hylton reckons he’s got to the bottom of the debate on analyticity between Carnap and Quine. He hasn’t, but he comes surprisingly close to getting Carnap very right at certain points — which makes it all the more disappointing when he then backpedals and decides not to follow through on those episodes of insight.

I’ll start with what goes right. The idea of using Putnam’s 1962 paper “The Analytic and the Synthetic” as a common point of comparison with Carnap and Quine is inspired. And Hylton’s section 3, describing a hypothetical Carnapian response to Putnam, is pretty much right on the money. At the beginning of the paper, things don’t look so good; Hylton seems to want to claim, like Ebbs, that an analytic-synthetic distinction requires the principle of tolerance (or vice versa) — but actually, it turns out he is making a much lower-key and subtler claim, to the effect that a distinction between analytic and synthetic is important to Carnap because he assumes the principle of tolerance in the background. Hylton is clearly not in the business of correct and incorrect, right and wrong, here, but of what “matters” and what “doesn’t matter” or is “significant” or “of no importance.” Hylton makes clear that we are in the realm of the normative rather than the descriptive.

This is made fully explicit in section 3 of his paper. Where Putnam claims that no sentence susceptible of “justification” can be analytic (i.e. no sentence with “systematic import,” since those are susceptible of justification), the Carnapian response is that there is “an ambiguity in the notion of justification.” (I should mention as an aside here that Carnap himself of course never thinks in terms of “justification.”) Internal statements are justified by the rules of the language in which they’re expressed — this much agrees with Putnam. But — Hylton continues — Carnap also sees a different form of justification at work when it comes to choosing among languages or language frameworks. “Justification of the choice of language must therefore appeal to kinds of consideration different from those involved in justifying the choice of one theory rather than another in a language that is not being called into question. . .”

. . . on Carnap’s account, truth is a language-relative notion. When it is the choice of language itself that is in question, we cannot aim at truth but rather at something vaguer, perhaps finding a language which, all things considered, facilitates our theorizing better than others. But this is not an all-or-nothing matter, like truth and falsehood. It is a matter of finding the most efficient and convenient language for certain purposes; it is thus relative to our current aims and is a matter of degree. The relevant considerations here are practical or pragmatic rather than theoretical. . . (pp. 7-8)

Precisely! True, Carnap wouldn’t have regarded any of this as “justification”; for him it was simply a matter of practical choice, but never mind, the relevant distinction between normative and descriptive is clearly stated and accurately described. And Hylton is right that this is what the distinction between internal and external questions in ESO amounts to — though he muddies the waters a bit when he brings the analytic-synthetic distinction into this picture:

The analytic sentences of the language are also, indirectly, susceptible of pragmatic justification, because they are fixed by the choice of language. So the sense in which analytic sentences are justified, or not, is quite different from that which is in play when we take the language for granted and speak of this or that theory as being justified. (p. 8)

As Hylton himself acknowledges elsewhere in the paper, the rules governing the relevance of empirical observations to theoretical statements are also “fixed by the choice of language,” so there is no difference here between analytic and synthetic sentences; the synthetic sentences of the language are also “indirectly susceptible of pragmatic justification.” So Hylton’s later summary (at the end of section 5), based on this mistake, is quite misleading:

. . . our question was: why should analyticity be epistemologically significant? We can now sum up Carnap’s answer: because the [analytic/synthetic] distinction is also the distinction between theoretical questions and practical questions. The former have right and wrong answers, which require theoretical justifications; the latter are a matter of language choice, which is a practical matter of better and worse for a given purpose, not a matter of right and wrong. The two kinds of questions are epistemologically quite distinct — which is why the principle of tolerance applies to the latter but not to the former. (p. 14)

The last part is fine; what is misleading here is the claim that the analytic/synthetic distinction “is also the distinction between theoretical questions and practical questions.” Practical considerations underlie not only the choice of analytic sentences in a language but also the choice of rules governing empirical evidence, i.e. the choice of synthetic sentences in a language; practical (normative) considerations govern the choice of language period (they govern any choice), and as I argued in a recent post, this distinction between considerations governing choice of language and considerations governing reasoning within a language can be made without an analytic-synthetic distinction.

On his way to this later, misleading, summary in section 5, Hylton retreats somewhat from his earlier Carnapian response to Putnam in his section 3. Unlike section 3, section 5 portrays the distinction between practical and theoretical as rather tenuous; “Carnap is (at best) walking a very fine line here. . .” (p. 13) But what follows after that summarizes what we saw above in section 3; so what is the problem? Perhaps Hylton worries that normative considerations are “vaguer” than theoretical ones, as he puts it. Well, they are purpose-relative and value-relative; there are no purpose-neutral or value-neutral answers to practical questions; the solutions will not be the same for every situation and every utility function, as they usually are in cognitive frameworks. But this needn’t make them vague or ill-defined. There is nothing vague or ill-defined about which Allen wrench is “correct” for a particular hex screw, but it depends entirely on the size of the hex socket (which in turn depends on what the screw is supposed to hold together).

What I find deeply mysterious, though, is that Hylton then, in his final section, backs away entirely from his portrayal of the Carnapian standpoint he describes in section 3, where the distinction between the two forms of Carnapian “justification” amounts to the distinction between theoretical and practical, descriptive and normative. In fact he now, suddenly, out of nowhere, wants to claim that Quine rejects this distinction:

Carnap holds that analyticity is significant because he holds that the [analytic/synthetic] distinction corresponds to a distinction between two kinds of justification. But Quine does not accept the latter distinction. On his account, all of our theorizing has the same very general aim, and so every proposed change is subject to the same very general criteria. . . The crucial point is that for Quine the notion of justification is not ambiguous; it does not bifurcate into two separate kinds. Carnap’s view, as we saw, requires that we have a notion of theoretical justification, applicable to choice of theory within a language, and a distinct notion of practical justification, applicable to choice of language. The distinctness is precisely what Quine denies. (pp. 14-15)

Precisely wrong! Quine not only does not deny the distinctness of descriptive and normative, he explicitly embraces it, e.g. in §13 of The Roots of Reference (and then more elaborately in the 1979 paper reprinted as Ch. 6 of Theories and Things). When the dog, in his metaphor, ignores the turnip on the other side of the tree, “two hypotheses, or a combination of them, may be advanced to explain the dog’s inaction with respect the turnip: perhaps he is not aware it is there, and perhaps he does not want a turnip. . . Observe then the duality of belief and valuation. . .” (p. 55 of Theories and Things). He advances a genetic hypothesis for this duality that fits in with the overall genetic account of language and reference in The Roots of Reference, and says that belief and valuation are often “intertwined,” but he certainly doesn’t deny their distinctness, as Hylton claims. He is what Carnap called a “non-cognitivist.”

Now it’s true that where the choice among languages or explications is concerned, he forgets his non-cognitivism, or for some reason thinks it doesn’t apply to this kind of choice. As far as I can tell, though, he nowhere gives a reason for limiting the scope of valuation this way, to exclude choice among languages or explications; in fact he never explicitly acknowledges this limitation of scope. Nor do any of Quine’s many brilliant current expositors (such as Hylton) give any attention to this question (yes, I know, I’ve complained about this before). But they need to! Either they need to explain this inconsistency away, or, if it’s really as glaring as it seems, they need to figure out how to repair it in a way that is consistent with whatever they think is most important about Quine in other respects. Should Quine — who prided himself on his “naturalism” — have also been a “naturalist” in G.E. Moore’s sense, to be consistent, i.e. should he have embraced what Moore called the “naturalistic fallacy” of supposing that ought reduces to is? Or should we rather see the “intertwining” of belief and valuation, descriptive and normative, as analogous to the (not quite) black and (not quite) white threads of empirical content and convention, which may perhaps be distinguished in principle but are indistinguishable in practice, unavoidably always resulting in a uniform “pale grey”? This approach is hard if not impossible to reconcile with what Quine actually says in Roots and the 1979 paper, though it might perhaps seem to fit better with Carnap’s own diagnosis of the difference between himself and Quine, as recounted by Howard Stein (pp. 278-9).

In any case, there is a serious tension here in Quine’s thought that those who know it better than I do need to address. Though he was a non-cognitivist, that seems not to have been so important to him; it didn’t cut very deep, whereas for Carnap it was fundamental. I’ve been looking recently at the roots of Carnap’s non-cognitivism in his earliest writings (as far back as 1911), and it does seem to have been something that (although often invisible on the surface) was central to his thought from the very beginning. So naturally, it was essential for Carnap to keep these radically different components of our collective cognitive enterprise distinct — to keep track of when we are relying on facts, or entailments in accordance with the existing rules of whatever framework we are working within, and when we are exercising choice, applying values, to choose among sets of rules. Like Poincaré, Carnap thought our languages and frameworks are our most important tools, and as Dick Jeffrey liked to point out, this meant that for Carnap it was critical that we maximize our conscious control over what we are doing in our inquiries, and shape our languages to meet our needs rather than drifting along at their mercy. It’s somewhere in that vicinity, I think, that the fundamental difference between Carnap and Quine is to be sought — which would certainly qualify under Carnap’s depiction (on that occasion reported by Stein): “Quine and I really differ, not concerning any matter of fact, nor any question with cognitive content, but rather in our respective estimates of the most fruitful course for science to follow. . .”