Jared Warren fancies himself “the lone contemporary defender of conventionalism in logic and mathematics” (p. 343 of his book Shadows of Syntax, 2020). His notion of “conventionalism” is somewhat offbeat (see below), but even so, he exaggerates wildly.
Warren goes out of his way to distance himself from Carnap, though admitting in asides that Carnap comes closer to his own “conventionalist” position than any other philosopher of the past. (I’m not so sure.) Carnap seems mostly (I haven’t read the whole book yet) to be portrayed as the garrulous old uncle who insists on boring us at dinner parties, and who may have got certain things right (perhaps by accident), but was too easily led astray by Neurath and other colorful personalities to be emulated as an inspiring forerunner.
This affords Warren ample opportunities to rap Carnap over the knuckles and tell us what Carnap should have thought or written, to bring him into conformity with Warren’s own more consistent and better-informed version of “conventionalism.” Warren does not, however, despite this frequent use of such normative language (addressed to Carnap), regard the question whether his own conventionalism is “correct” as a normative question. No, there is a fact of the matter about that; Warren is right, in his view, and everyone else is wrong. And yes, he means factually right! He has arrived at the “uniquely true and correct theory.”
Warren spells out his case for this at some length in his Ch. 14 (“The Facts of the Matter”), and that’s what I want to focus on in this post. There are many other aspects of interest in the book, and I hope I can come back to them before long, but this issue of the nature of philosophical disagreement seems pretty fundamental, so I want to address it head-on before going into more detail on anything else. It seems fundamental to me, anyway; I realize many readers will regard it as a peripheral meta-problem, not worth getting into – after all, the book makes some actual substantive contributions to various debates, and shouldn’t one concentrate on those, rather than an issue on which Warren has nothing very original to say and is only making explicit something that many people take for granted without bothering to articulate it? – he’s merely “saying the quiet part out loud.” But that’s just why I want to address it; it’s this widespread assumption about the point and nature of philosophical argument that I think should be made explicit and addressed explicitly.
Warren shares the widespread conception that the point of philosophy, as of any discipline, is to get things right – to reach conclusions that are, as he puts it, “factually true.” If you’re not into getting things right, in this view, then you can’t be serious about philosophy. Or rather, he embraces this conception not for all of philosophy, only for the top level, the most general, meta-semantic level, and claims not only that such a “metasemantic factualism” can co-exist with “a healthy amount of indeterminacy about meaning, reference, translation, and the like” (p. 338) but that “Metasemantics is a matter of fact.” (p. 337) I will get to his arguments for this claim in a moment, but for now I want to point out that this view is inconsistent with much, perhaps most of what has called itself philosophy over the past couple of thousand years. Socrates set little store by factual correctness as a goal of dialogue. Skipping forward, Carnap was well aware that major philosophical differences often involved what he called “non-cognitive” matters, questions about which there isn’t a correct or incorrect answer or solution but rather different perspectives based on different attitudes and backgrounds and sets of values.
Warren should be sympathetic to this line of thought. He himself says, in this very Ch. 14:
The human activity of philosophy is not a natural kind. It is instead a wildly heterogenous collection of activities, in many cases held together only by contingent historical associations. Given this evident fact, there is no reason to expect a unified metaphilosophical analysis of the whole of philosophy (p. 338)
Fine. But not much of that heterogeneous collection fits comfortably under the rubric of Warren’s metaphilosophical factualism. The vast bulk of it, I would venture, is closer in spirit to the tradition from Socrates to Carnap. In this more traditional view, the typical philosophical problem is to give an explication of some vague or otherwise unsatisfactory concept, and this problem is best viewed on the model of an optimization problem in microeconomics – the optimization of some function (utility function, production function – in the philosophical case more like an insight function or enlightenment function) subject to constraints. In economics, the constraints are the prices and quantities of the goods or factors between or among which trade-offs are envisaged; in philosophy, the constraints are the facts more generally.
Socrates and Carnap (and most of those in between who thought of themselves as philosophers) of course thought that in philosophy one had to get everything right about the constraint structure. You had to get your “descriptive pragmatics” right, in Carnap’s lingo, you couldn’t refuse to face facts or carelessly leave relevant facts out of your picture. You couldn’t violate the factual picture established by the theoretical and empirical standards of the disciplines in which that theoretical and empirical work is done (what Neurath called the “encyclopedic” picture). But the optimand you come up with within these factual constraints is a proposal, not a claim or an assertion. You come up with a language, say, for articulating the bearing of evidence on substantive scientific theories, but you don’t claim it’s the right language – you propose it in the same spirit in which a software engineer proposes a new programming language. It may be found useful, it may open up new perspectives, it may make it easier to reveal connections between apparently heterogeneous fields of inquiry.
These are all good things, but you’d never claim it was therefore the correct language or the true language. Engineering is like art, not like physics; in engineering you don’t go for true (as opposed to false) – you go for elegant or productive or fruitful or for optimal according to certain requirements or values. Carnap’s principle of tolerance basically stated just this: philosophy is a matter of engineering, not of science – it’s a matter of optimization within constraints, not a matter of truth or falsehood. So the principle of tolerance is another way of putting Carnap’s basic non-cognitivism (or “functionalism” as Huw Price calls it).
There is a long tradition of misunderstanding this basic idea. Quine didn’t get it (as Rick Creath has shown over the years, in painstaking detail), and kept insisting – despite his own professed non-cognitivism — that the distinction between semantics and pragmatics, for instance, was a “pernicious error” (Quiddities, p. 211). Like Warren, in other words (but on the other side, more or less), he didn’t just think this distinction was unfruitful or clumsy or otherwise inelegant, but that it was mistaken; he was right, in other words, and Carnap was wrong. (In Carnap’s own presence, it seems, Quine could be reminded that the difference between them “was not a cognitive one,” but then invariably slipped back into his default mode.) Gödel’s famous critique of Carnap, based on the second incompleteness theorem, again tries to show that Carnap is wrong. Gödel and Quine (and other less illustrious critics such as Popper, Ryle, Strawson, and Putnam) still have defenders; the appeal of this idea that the principle of tolerance is mistaken will never fade. But the very attempt to argue that the principle of tolerance is mistaken reveals the attempter’s failure to understand it. The principle of tolerance can no more be “mistaken” than freedom of speech or the rule of law – or the staircase of the giants in the Doge’s Palace, or Wozzeck, or The Ambassadors. All these things can be criticized, of course, but not on the grounds that they are cognitively false, or factually mistaken, or incorrect. And cognitive considerations can be relevant to their appraisal – again, the constraint structure faced by proposals or ideals or engineering and/or artistic creations can be partly or largely cognitive (see above). But the decision whether to admire or accept or advocate or propagate such things is not itself a cognitive one.
Warren’s argument in his Ch. 14 for meta-semantic factualism is disappointingly feeble. It has three main components. First, he tries to revive Alberto Coffa’s claim that Carnap (among others) was himself a meta-semantic factualist. He doesn’t actually go into the (very cogent) arguments against this claim advanced by Ricketts in his 1994 paper (in the Reading Putnam volume) or Goldfarb in “Semantics in Carnap” (Philosophical Topics 1995), he just dismisses them, as Putnam dismissed Ricketts on the grounds that the Carnap he portrays “wasn’t the Carnap I knew and loved” – indeed Warren even cites Putnam’s dismissal and defends it, rather than engaging with Ricketts’s or Goldfarb’s actual arguments against Coffa! In any case, he insouciantly concludes, it doesn’t really matter whether Carnap or Reichenbach or anyone else actually ever shared Warren’s own correct version of “conventionalism,” and he moves on to the second component of his argument.
This second component is a misleadingly named section called “How Conventionalism Could Be Factually True” that begins with the sentence, “Given the evident factuality of metasemantics, there is a fact of the matter about the correct philosophical theory of logic and mathematics.” – In other words, he begins by assuming what his title indicates is to be shown. He evidently regards himself as entitled to this assumption on the basis of his claim in the preceding section: “Metasemantics is a matter of fact. This isn’t to say that what our utterances mean is always fully determinate, only that any semantic indeterminacy in a given utterance is itself a matter of fact.” (p. 337) No argument is given for this. Nor are examples given, but if we take Coffa’s example of a semantic fact, “all mathematical axioms are conventional,” as a case in point, Warren’s claim amounts simply to restatement of his assertion that “conventionalism” (in his sense) is correct.
But of course we have no idea what Warren or Coffa means by “all mathematical axioms are conventional.” As Goldfarb points out, “When Coffa puts forward the claim ‘all mathematical axioms are conventional’ as a candidate for factual status, he does not seem to be using ‘conventional’ to refer to the distinction that is made, relative to each framework, between what the framework contains as factual and what it contains as analytic. (After all, a mathematical axiom is not conventional given the framework – that phrasing makes no sense.) To my ear, Coffa is relying on a global conventional/ nonconventional distinction. But this would bring in just the framework-independent conception of fact that Carnap is at pains to avoid.” (p. 58 of Goldfarb’s 1995 paper)
Is Warren, like Coffa (apparently), “relying on a global conventional/nonconventional distinction”? It’s unclear, but seems likely. In any case, the second component of Warren’s argument does not actually address the question of how “conventionalism” could be factually true. Instead, it summarizes some of points made previously in the book, to the effect that “conventionalism” (in the global sense that Warren, like Coffa, seems to have in mind) solves all sorts of problems better than the alternatives – it is “unmatched in its theoretical simplicity”; it is “not at all ad hoc,” it is “extremely general and thus has unmatched explanatory scope,” and so on. But these are usually referred to as “theoretical virtues” – not as facts! (Appropriately, since “better” and “best” are paradigmatically normative words.) So the section devoted to showing “how conventionalism could be factually true” does no such thing; its argument is not factual at all but normative. Perhaps Warren does not recognize this distinction (perhaps his global conventionalism embraces a form of cognitivism regarding normative language), but then his own claim that any semantic indeterminacy is a matter of fact falls to pieces, since it is not even clear whether his apparently normative claims are to be understood normatively or cognitively. There is no fact of the matter, here, about whether there is a semantic indeterminacy; no facts bear on this question. What is missing is a rule framework governing these expressions, rather, that would disambiguate them, telling us whether they are to be taken as descriptive (factual) or normative or something else. And rules are generally classified as conventions rather than as facts.
The third and final component of Warren’s argument addresses a couple of ad hominem meta-objections that needn’t detain us here, as they are more relevant to the social acceptability of accepting what he calls “conventionalism” than to its semantic or cognitive status.
I have no illusions that there is any prospect of overcoming this disagreement about philosophical disagreement, since although Warren is willing to admit he may be wrong, he doesn’t have room for the idea that philosophy might be more like engineering than like chemistry, and its problems less like factual problems than like optimization problems within factual constraints, where it makes no sense to say that a chosen optimand is either factually right or factually wrong. Each of these two conceptions of philosophy is itself a proposed optimand, and the arguments on either side must therefore be normative — though again, within factual constraints. One such constraint might be the past history of the discourse that has called itself philosophy over the past millenia; and while I fully accept Warren’s point about the heterogeneity of that discourse (see above quotation), I think it’s clear that most of what has gone on under the heading of philosophy corresponds more closely to the Socrates-Carnap conception than to his. One can think of other factual constraints, and one can imagine normative arguments on each side.
But wait — in his Ch. 14, Warren has his opportunity to put forward his best possible case for his factualist conception, and as far as I can see, he presents no argument whatever. Perhaps this indicates that there is no such argument. Regarding the factual constraints about the interpretation of logical empiricism, he asserts that Coffa is right and all the newer stuff is wrong, but doesn’t even try to argue for this. And his argument for how “conventionalism” could be factually true is simply to assert that it’s factually true. Is there an argument in there somewhere? Have I missed something?
 Goldfarb suspects that Coffa shares Gödel’s misconception about Carnap’s language-relativity of meaning and truth: “Gödel’s criticism assumes that we first have a realm of empirical fact; given it, we then adopt the conventions that yield mathematics. In Carnap’s terms, this is just to presuppose a language-transcendent notion of empirical fact. Carnap rejects any such language-transcendent notion. Indeed, the Principle of Tolerance enshrines just this rejection. On Carnap’s view, it is only given the apparatus of a linguistic framework that we can formulate a notion of the realm of empirical facts. Carnap’s view here can be taken to be his version of the priority of the logical: only once logical relations are given is objective discourse possible, and logical relations are given by the linguistic framework. I suspect Coffa does not recognize how deep this goes.”(pp. 57-8) Warren appears to be in the same boat, but to establish just how far he goes along with this sort of “global” conventionalism I’ll have to finish reading the book. Perhaps this question will be addressed in a later post.