Peter Hylton has a paper in the Monist, now a few years old, that I’ve been meaning to comment on because it exemplifies a bad habit much of the Carnap-Quine literature suffers from: comparing the early Carnap with the later Quine. This is tempting, of course, because Quine himself did it (as e.g. Gregory Lavers has pointed out), and even Burton Dreben (Peter’s doctoral supervisor), though far more scrupulous than Quine, tended to fall into it; I guess it became a sort of Harvard thing, and Peter can’t be blamed too much for slipping into the ruts of his elders. However, he happens to have chosen a subject where this mismatch gets him into serious trouble, since if he’d actually compared the mature Quine with the mature Carnap, his main points wouldn’t be just questionable, they’d have collapsed entirely.

Nothing in the paper is quite as blunt as the synopsis at the beginning, where Peter says “Carnap (post-1932) takes philosophy as a wholly abstract subject. This makes it impossible for him to give an account of evidence.” What he means by “wholly abstract” is “not concerned with anything empirical” — whereas for Quine, of course, philosophy was continuous with empirical science, on which it had sometimes had to draw.

So “Quine rejects the principle of tolerance: what we take as the basis of our knowledge is not settled by a free choice of language; it is, rather, a factual question about how we come to know the world. . . it is an empirical question and, since science embodies our best knowledge, a scientific question.” (p. 217) Well, what if the same old philosophical controversies show up in the relevant sciences? Sensory neurophysiology is a controversial field (or grouping of fields), every bit as riven today (or in Quine’s lifetime) as it was in Helmholtz’s time between rationalists (or Kantians, or innatists, or whatever you want to call them) and empiricists (or Gibsonians, believers in “direct perception”). Sure, they all agree that we need the sensory input, but they don’t agree (any more than Poincaré agreed with Mach) about how our socio-cognitive apparatus processes that input, and how much it contributes to the result. So how does Quine’s obiter dictum help? Its rhetoric is scientific but it ignores the actual state of the sciences it’s supposedly drawing on. We all know we need observation (and we didn’t need “science” to tell us that), but by itself this gives us no assurance whatever that our language for talking about the perceptual basis of (which?) knowledge needs to be one of observation sentences (of what form?), or one whose basic items are physical objects. Or something else completely.

For Carnap, Peter says, there is “no place for philosophical argument to defend” his choice of a physicalistic (or sense-datum) protocol language. Well, there is no place for simply calling it a factual question and leaving it at that; clearly there are ideologies or values, normative questions of some sort involved here. And so for Carnap this appropriately became a practical (normative) question rather than a factual one. I’m not sure why Peter thinks that it’s a “philosophical” argument if it’s factual but not “philosophical” if it’s normative. (Not that Carnap would have cared; his colleagues all thought that what he was engaged in wasn’t philosophy but some sort of engineering — and according to Dick Jeffrey, that was just fine with him!)

Carnap shared Quine’s scientistic prejudices, but was able to step back from those (perhaps because he was at Chicago rather than complacent Harvard) and leave such questions about language choice open. He had his own preferences about languages but thought that since there were ultimately normative questions involved, and “in logic there are no morals,” his own preferences shouldn’t decide the issue; he distinguished framework issues from content issues, and like the putative Voltaire he might despise what you said but was willing to die for your right to say it.

That attitude was there all along, even in the early Carnap, certainly during the Vienna years (if you buy e.g. Thomas Uebel’s notion about the “bipartite meta-theory” as a kind of ideal joint project arising from the dialogue between Carnap and Neurath). But it became more articulate and explicit in Carnap’s later years. In the late 1930s he introduced a third branch of meta-discourse, alongside syntax and semantics, which he called “pragmatics.” Admittedly he made it sound pretty trivial, in his rather wooden English; it was the study of language in use, he said, “incorporating the context of the user.” But when Howard Stein pointed out in an essay he wrote for one of Carnap’s graduate classes that much of the philosophical weight had shifted to pragmatics (which now included all the traditional sectors of philosophy — epistemology, ethics, even the philosophy of philosophy — within its scope), Carnap agreed enthusiastically.

Pragmatics had two parts: “pure” pragmatics, where proposals were developed for languages (or, more locally, for explicata) in which to articulate aspects of language in use: e.g. logics of belief, of normative statements and imperatives, of decision and interaction (decision and game theory would have counted as pure pragmatics), and so on. “Descriptive” pragmatics, on the other hand, is all the empirical knowledge we have available that bears (or potentially bears) on practical decisions, such as which language or explication to choose for a particular purpose in a particular case.

Naturalized epistemology is, of course, part of descriptive pragmatics, as are, for instance, the fascinating reflections on the tangled histories of the concepts and languages employed in various sectors of applied physics and chemistry by Mark Wilson. The study of evidence, which belongs in that general neighborhood, has two aspects; the empirical part, which draws not only on the neurophysiology of perception, but on the engineering physics of the instruments we employ, the information theory of the transmission of relevant data from the phenomena to be measured to the instrument output, and much else. The other aspect is the pure pragmatics part: developing concepts and languages capable of coordinating all this empirical knowledge into a coherent theory of evidence. And part of that would be languages for determining the relevance of particular pieces of evidence (data points) or collections of them on the conclusions we draw, e.g. to confirm or refute empirical hypotheses or theories. This was, of course, precisely the thing Carnap focused on from about the age of 50 to the end of his life; and unlike Quine, he actually got his hands dirty with the messy details. It makes little sense to claim that, unlike Quine, the later Carnap had no account of evidence.

Carnap’s lack of attention to the empirical part — the naturalized epistemology of evidence, if you like — was a matter of specialization. He claimed no expertise in neurophysiology, information theory, engineering physics, or other possible components of the relevant naturalized epistemology, and didn’t see the need to weigh in on them. What he did insist on was that all these sciences needed to be capable of articulation in a single language; you couldn’t have a coherent theory of evidence if each science is framed in a language system that is incommensurable with the others. This insistence on the unity of science as a goal was motivated ultimately by practical concerns. We can’t solve the problems our species is faced with, he said, unless we can weigh the knowledge available from various sciences in contexts of action. To deal with a forest fire, for instance (an example he used in a talk about the unity of science in the 40s), we have to be able to coordinate chemical and botanical knowledge about flammability of various plants under different conditions of moisture, meteorological knowledge about likely winds and rainfall, economic and sociological knowledge about how people will respond, political knowledge about the management of the firefighters and the resources available to them, and much else. More obvious examples for us today would be the environmental challenges we face globally, including climate change, or the challenges posed by the outbreak of a pandemic. Our abject failures in dealing with these challenges (even fires, as any resident of California knows very well!) are largely failures of practical coordination, but the absence of an accepted lingua franca among the sciences — even of the rough and ready sort advocated by Neurath — makes it way too easy for those who oppose science to get away with sowing confusion by claiming that science gives no clear and unambiguous answers.

Quine agreed about the unity of science, of course, but why? Presumably Peter would argue that science tells us that all the sciences must ultimately be articulable in the same linguistic framework, but that is obviously going out on a limb — extrapolating rather wildly from the unifications we’ve observed over the past few centuries. Carnap left open whether complete unification could ultimately succeed; he saw it as a goal, not a fact. And meanwhile, he thought we have the option, where decisive evidence about some question is hard to imagine, of kicking the question upstairs, from the empirical realm to the realm of language choice. While he sympathized with Feigl’s attempts to pin down the mind-body problem empirically, for instance, he suspected that at this stage, it might make more sense to just go ahead and use a monistic language for this issue (and reject a dualistic one), simply on grounds of potential fruitfulness and the fit with relevant sciences (biochemistry, neurophysiology, etc.) — a normative choice informed by descriptive pragmatics. When and if enough evidence accumulates to give us hope that it might actually decide the larger questions of mind and body, we can always return them to the empirical realm.

In practice, Carnap and Quine probably wouldn’t have differed much here. But Carnap at least had an overall account of what we are doing as we shift from an empirical to a normative framework or vice versa. For Quine, as Peter reaffirms, there is no difference. And yet Quine explicitly endorsed non-cognitivism; why did he not think it extended to choices among linguistic frameworks? Or, to look at it from the other side, why did his otherwise thoroughgoing naturalism not extend to normative sentences and frameworks (i.e. why wasn’t he also a “naturalist” in G.E. Moore’s sense)? Either way, there is a glaring inconsistency here (or at least the appearance of one) that I wish Peter had addressed in his Quine book.