Quine’s Tarskian Angle on Truth: Immanence, Semantic Ascent and the Importance of Generality

Ia. Tarksi/Quine Basics

It is, I said, the capacity for generality that calls for the substance of the Tarskian approach to truth, beyond underwriting the T-schema. The dummy letter ‘S’ in the T-schema is assuredly not a variable or pronoun; it is why we speak of a T-schema — not a T-formula, a T-axiom, or a T-principle. To advance it is only to advance its infinitely many instances (to replace the dummy letter with a variable or pronoun, bound by a quantifier, is on its face a confused attempt to have variables doing two duties at once).⁸ This leaves us without instructions for dealing with ‘x is true’ where ‘x’ is a variable. For Quine, Tarski’s positive results show that we can nevertheless get the effect of generalizing using a truth-predicate, as when we say that all theorems of a theory are true. As Quine puts it, what a suitably explained ‘true’ does is to make available another “dimension” of generalization — “oblique generalization” he once called it (PL: 97). We can generalize on ‘Socrates is mortal’ by writing ‘Every man is mortal’. Similarly, we can generalize on ‘If time flies then time flies’ by writing ‘Every sentence of the form “if p then p” is true’ (more explicitly: ‘For every x, if x is a sentence of the form “if p then p”, then x is true’); on ‘“If 2+2 = 4” is provable in Q then 2+2 = 4’ by writing ‘Every sentence provable in Q is true’ (more explicitly: ‘For every x, if x is a sentence provable in [the intended interpretation of] Q, then x is true’). Such are particular examples of what Quine calls “semantic ascent” (PT: 80–2; PL: 10–13; 1960: §56 270–6). We ascend to a meta-level, quantifying over sentences, mentioning rather than using them. Oblique generality, one might go so far as to say, is the core theoretical reason for the truth predicate, just as the more familiar kind of generalization is the reason for the quantifiers.

A truth predicate must however be only “incompletely” disquotational, notes Quine (PT: 83).⁹ The lurking spectre of the liar paradox shows that such a predicate “must not disquote all the sentences that contain it” (PT: 83). Tarski’s way to bring about the capacity for oblique generalization whilst keeping consistent involves an inductive definition of satisfaction. Simplifying, and following Quine in Pursuit of Truth (PT: 84–6; the presentation in PL: 35–43 sticks closer to Tarski), and in particular referring with Quine to arbitrary assignments of values to variables rather than to Tarski’s original infinite sequences of objects, the base will contain clauses appertaining to the true-free object language such as that an assignment satisfies ‘x is a dog’ iff the object assigned to ‘x’ is a dog — and so on for each of the finitely many simple predicates in the language, including relational predicates (we assume that the language lacks names, since for scientific purposes we can always replace them with singular descriptions, contextually eliminable in Russell’s way; similarly we can ignore function-symbols). Then, for the recursion (following the syntax of the language), we say that an assignment satisfies the negation of a formula iff it does not satisfy the formula, and that an assignment satisfies a disjunction iff it satisfies either disjunct; and finally that an assignment A satisfies an existential quantification ‘∃x(…x…)’ iff some assignment A* satisfies ‘…x…’ that matches A save, at most, what it assigns to ‘x’. Other truth-functional connectives and the universal quantifier are definable in terms already given. Truth then is just satisfaction by all assignments (or what comes to the same, by some assignment). It’s not for nothing that Tarski’s is sometimes called a “compositional” theory of truth.

Quine also presents Tarski’s idea not in terms of a sequence of metalanguages — with L₁ defining a truth predicate for L₀, L₂ for L₁, and so on — but in terms of a single language, with an ascending sequence of truth predicates (and of satisfaction relations): “The hierarchy begins with a predicate ‘true_o’, which disquotes all sentences that contain no truth predicate or equivalent devices”; then a “predicate ‘true₁’ disquotes all sentences that contain no truth predicate or equivalent devices beyond ‘true_o’”…and so on (PT: 84; for more detail 1991b: 219–22, 224–30).¹⁰ Then, with variables separated in such a way that each sort of variable ranges over a given Russell-style type and lower, every membership condition determines, if not a set, always a class — that is, a class which may be a proper class, in von Neumann’s (1967 [1923]) sense (PT: 88–9).¹¹ We would thus have a single language adequate for representing the set-theoretic hierarchy simultaneously with the closely related hierarchy of truth-predicates. Similar to a hierarchy of sets, the hierarchy of truth predicates is unbounded above: “[w]e get a self-contained language with a hierarchy of better and better truth predicates but no best” (PT: 89). “This is how I like it”, says Quine (PT: 90; also 1991b: 221); not only for the structural insight it affords, but because it comes as close as is feasible to a framework adequate to a single statement and analysis of what is known, to a “limning of the most general traits of reality” (1960: §33 161).

Ib. Semantic Ascent

This last move fits with Quine’s famous if difficult pronouncement that “[t]ruth is immanent, and there is no higher. We must speak from within a theory, albeit any of various” (1981a: 21–2). Elsewhere he states:

Whatever we affirm … we affirm as a statement within our aggregate theory of nature as we now see it; and to call a statement true is just to reaffirm it. Perhaps it is not true, and perhaps we shall find that out; but in any event there is no extratheoretic truth, no higher truth than the truth we are claiming or aspiring to as we continue to tinker with our system of the world from within. (1975: 327)

Semantic ascent is not metaphysical ascent, whatever that would mean. All the same, the vital importance of fashioning a truth-predicate that allows for semantic ascent cannot be overstressed. For one thing:

Semantic ascent serves also outside of logic. When Einstein propounded relativity, disrupting our basic conceptions of distance and time, it was hard to assess it without leaning on our basic conceptions and thus begging the question. But by semantic ascent one could compare the new and old theories as symbolic structures, and so appreciate that the new theory organized the pertinent data more simply than the old. (PT: 81; see also 1960: 271–2)

The reference to Einstein shows that big fish are to be fried. For another thing, perhaps the most central use of semantic ascent, therefore the most central generalizing use of ‘true’ — especially for one with the interests of Quine, author of so many books on the subject — concerns logic itself, as intimated in the first sentence. Quine’s answer to the question “What is Logic?”, as expressed in Philosophy of Logic, is as well-known given not in terms of logical implication or a set of axioms but in terms of logical truth. And “[a] logical truth is, on this approach, a sentence whose grammatical structure is such that all sentences with that structure are true” (PL: 58). This gets refined as “a logical truth is a sentence that cannot be turned false by substituting for lexicon, even under supplementation of lexical resources” (PL: 59; Quine’s emphasis). By “lexicon”, he means an open-ended list of expressions with (possibly null) extensions; this then excludes the “syncategorematic” particles and in particular the logical particles (PL: 27). We have truths of chemistry, truths of physics, truths of mathematics, and finally the truths of logic. We can pare away lexical content until we reach the hard skeleton of lexicon-indifferent grammar, the instances of which can nevertheless be generalized about by means of semantic ascent. We can say, and indeed prove, for example, that every sentence of the form ‘If p then p’ is true.¹²

The possibility of such a definition shows in spades the value of semantic ascent, of generalizing by means of truth. But the existence of this along with the possibilities lately canvassed is separate from the theoretic account of truth (from “what truth is,” one almost can’t help but say). Semantic ascent is vital to all these activities, but these activities do not enter into the description of semantic ascent itself — in particular not into the very definition of truth (or rather the inductive definition of satisfaction). Quine writes of oblique generalization as the “crucial purpose” of the truth-predicate (PL: 97), not the multitudinous inferences licenced by the T-schema (for illuminating detail on such generalization, see Heck (2021)).

Ic. General vs. Particular

It is nonetheless best sometimes to plant our feet on the ground and focus on the particular, not to be carried away by generalization. In a passage in From Stimulus to Science, Quine writes:

Along with this seriocomic blend of triviality and paradox, truth is felt to harbor something of the sublime. Its pursuit is a noble pursuit, and unending. In viewing truth thus we are viewing it as a single elusive goal or grail. (FSS: 67)

The language of “is felt” in the first sentence shows that Quine does not quite commit as a scientific philosopher to this sentiment, even if he sympathizes with it informally and certainly does not think it misguided, however vague it is. He goes on:

In sober fact the pursuit resolves into concern with particular sentences, ones important to us in one or another way. Pursuit of truth is implicit, still, in our use of ‘true’. We should and do currently accept the firmest scientific conclusions as true, but when one of these is dislodged by further research we do not say that it had been true but became false. We say that to our surprise it was not true after all. (FSS: 67)

The three words at the beginning — in sober fact — more emphatically indicate a professional distance from this feeling. The lesson — what the concern “resolves into” — is more vivid if we adhere to a single example, availing ourselves of disquotation. When one of these — Quine’s example is ‘Light rays are straight’ — was dislodged by further research, we did not say that they had been straight but are no longer. We say that to our surprise they were not generally straight after all (adapted from Quine 1994a: 498).

All the same, what is said to be “implicit. in our use of ‘true’” — the generalizing function — opens the door to philosophers. Quine continues:

Science is seen as pursuing and discovering truth rather than as decreeing it. Such is the idiom of realism, and it is integral to the semantics of the predicate ‘true’. It fittingly vivifies scientific method, the method of interrogating nature by conjecture and experiment and abiding by the consequences, (FSS: 67; see also Parsons 2020: 223–4)

In a perhaps more alarming passage, written not long after the above, he writes:

To call a sentence true, as I said, is to include it in our science, but this not to say that science fixes truth. It can prove wrong. We go on testing our scientific theory by prediction and experiment, and modifying it as needed, in quest of the truth. Truth thus looms as a haven that we keep steering for and correcting to. It is an ideal of pure reason, in Kant’s phrase. Very well: immanent in those other respects, transcendent in this. (1995b: 353)

It might seem as if Quine is not using Kant’s phrase merely for its connotative appeal, as if he were having second thoughts about his pronouncement of the immanence of truth. But again, defusing the rhetoric slightly, the serious non-figurative point is only to illustrate the possibilities afforded by semantic ascent, of generalization by means of truth. That is the respect in which he can go along with Kantian phraseology, that truth is “transcendent”. The “idiom of realism” is displayed by the transfer of a certain attitude towards ‘Light rays are straight’ just considered to assertions in general. “Where the truth predicate has its utility”, he writes, “is in just those places where, though still concerned with reality, we are impelled by certain technical complications to mention sentences”. For “the truth predicate serves, as it were, to point through the sentence to the reality; it serves as a reminder that though sentences are mentioned, reality is still the whole point” (PL: 97).

This and other philosophical predilections, I am saying, are implicitly only made possible and fully intelligible by our understanding of the sentential generalizing function of the truth-predicate (I admit that this is more my assertion than Quine’s, as no statement of his quite amounts to it). Indeed, the generalizing function makes it intelligible to speak of truth in first place, i.e., by means of an abstract noun rather than an adjective, or rather by means of the predicate ‘__ is true’ (one can thus venture such grand pronouncements as "[t]ruth thus looms as a haven that we keep steering for and correcting to"). To put the point in a way reminiscent of Wittgenstein, the generalizing function — a suitably augmented disquotationalism, a compositional theory of truth — adequately explains the use of ‘true’ (or rather explicates it; that is, replaces the term for regimented purposes with a closely related but cleaned-up alternative, as in Quine 1960: §53). These further philosophical questions and tasks put the term to work. It is just that we are apt, if we are not careful, to mistake these further philosophical questions as bearing on the definition of truth — on the “nature” of truth.¹³

IIa. Main Criticism

An obvious worry would be over extensional equivalence: I have spoken of different “dimensions” of truth, but surely the two dimensions do not determine the same class. That is so, as indeed Bergström recognizes. But even if it were not so, there is a criticism to be made that operates, so to speak, at the intensional level. In this section, I leave aside the extensional worry.

My main criticism is rather simple and may be anticipated from subsections Ib and Ic. There is no doubt that the concept of truth is used out of necessity in conveying Quine’s characterization of the route from the stimulation of one’s nerves to an articulate, warranted, and realistic theory of the world. It’s why his penultimate book is called Pursuit of Truth. However, that a theory requires the use of a concept need not affect what the concept is, as one might put it. More carefully: that a certain linguistic expression (or some equivalent) is needed for the expression of a theory does not mean that the theory is thereby imported retroactively into the very definition or logic of that linguistic expression (or its equivalent). The term ‘fruit’ may be essential to a fruit-picking manual, but is fully understood independently, apart from the material on fruit-picking. The term ‘acid’ may be vital for characterizing the engineering of industrial cleaning supplies, but the chemical theorist may define the term quite independently of that business. Likewise for claims about inference, testimony, assertion, induction, evidence, or belief-revision: often these manifestly require the generalizing function of truth, but they don’t revise the theory of truth by making essential use of the relevant words. From Quine’s point of view, that these theories — and in particular Bergström’s empirical theory of truth — are made possible by the generalizing function of truth does not indicate that Tarski’s work was in any way inadequate or incomplete. Assuming we accept Tarski’s compositional theory of truth, the only way in which such backwards influence could happen would be, for example, if developments in some special discipline called for a material alteration of the Tarskian theory itself, which admittedly is conceivable but presumably most unlikely. Bergström’s empirical theory is not felicitously described as a theory of truth, not if by that appellation is meant an analysis or explication of truth itself. (Quine himself spoke of a certain monkey wrench being thrown into empiricism if somehow extra-sensory perception or soothsaying had to be admitted as real, but he gives no hint that our conception of truth would be disturbed; PT: 20–1).

Enthusiasts of ‘Two Dogmas of Empiricism’ (1961 [1951]) might complain that this sounds too essentialist. If everything is connected with everything — if the famous “holism” holds — then we cannot rightly talk as if conceptual borders were impermeable, as if concepts retained their identities irrespective of the theories in which they figure. And it is true that from a highly abstract philosophical point of view — as at Quine (1976: 76) — no such boundary is sacrosanct; they are ultimately matters of more and less, even if in practice at least some may be regarded as categorical. But still, this exaggerates or over-stresses Quine’s holism. Not only would it follow from this maximal holism that every discipline which uses a given form of words cannot develop without disturbing the semantics of that form of words (that indeed one cannot alter one’s beliefs without changing the subject and without changing the significance of every belief and thus making it virtually impossible properly to disagree, as Fodor and Lepore put it in their 1992 criticism of Quine; 1992: 37–58). More to the point, this reply on behalf of Bergström fails to take account of the considerable back-pedalling in which Quine engaged after the broad if not reckless statement of holism in ‘Two Dogmas of Empiricism’. In ‘Empirically Equivalent Systems of the World’ of 1975 and ‘Two Dogmas in Retrospect’ of 1991, Quine speaks of holism as only a “legalism” — that if in principle any theoretical change can have repercussions for any region of theory, that is hardly so in practice (1975: 316; 1991a: 268). “In later writings”, he emphasizes, “I have invoked not the whole of science but chunks of it, clusters of sentences just inclusive enough to have critical semantic mass” (1991a: 268; inclusive enough, that is, to entail the relevant observation categoricals). There is no reason on the score of this now “moderate” holism to discount the existence of boundaries, often propped up by sound theoretical principles, between domains — not least of all the boundary that sets logic off from the rest, which Quine devoted more or less an entire book (Philosophy of Logic) to articulating (I mean this as consistent with the fine-grained account of Quine in Verhaegh (2017)).

That is all I’m claiming for Quine’s account of truth. Its purpose — officially, in promoting an explicated concept engineered for regimented use — is to underwrite a certain kind of generalization, and there is no reason to deny that it is suitably insulated from Quine’s other philosophical projects. That it so underwrites is the essence of truth, if we want to find a place for that form of words.

IIb. Morality, Aesthetics, and Superfluous Sentences

The foregoing might be said not really to disturb the kernel of what Bergström claims–that it implies only that insofar as any such a claim is true, the claim may be part of epistemology or whatever but it is not part of the theory or analysis of truth itself, not from Quine’s perspective. But there are some more specific claims of Bergström’s that I think are false. They are indeed entailed by Bergström’s thesis, and to see this will in turn illuminate further dimensions of Quine on truth. In his reply to Bergström, Quine (1994a: 497) says:

I see truth in general as far exceeding, in turn, anything that can be checked in observation categoricals…[m]uch that is accepted as true or plausible even in the hard sciences, I expect, is accepted without thought of its joining forces with other plausible hypotheses…to form a testable set.

Such statements are matters of filling in, of the rounding out of theories — normally for the sake of explanatory or aesthetic ideals if not for mere convenience — but they are not strictly speaking essential to any testable set of statements.¹⁸ The Continuum Hypothesis is a likely example; we hear that string theory might be another, and it stands to reason that there are many less sensational examples. Bergström, as in the above quotation, holds that some statements — the “superfluous” ones — pass the disquotational test but not the empirical test, and thus are not strictly true (1994: 429). Further and striking support seems to emerge from what Bergström says in connection with statements of “moral and aesthetic value”, that famous inflection point for logical empiricism or positivism (1994: 433):

Many empiricists have felt that moral and aesthetic value statements are neither true nor false. Quine is also attracted to this view [in private conversation]. But it seems clearly wrong on the disquotational account. On the other hand, it seems very plausible on the empiricist account of truth. This is a further reason why Quine should accept the empiricist account.

Set aside any remarks in private conversation (and note that Quine is silent on the issue in his response). A perhaps wooden reply, but nevertheless an interesting and substantive reply, can readily be inferred from Quine’s public remarks.

In brief, the reply runs, value-statements can indeed be true or false, even if incapable of empirical justification, but this does not generate any pressure to adjust the account of truth. Such statements are not going to fit as they stand into Quine’s epistemology, as Bergström notes. To cite empirical evidence for them, let alone to verify them, is not the thing, as so many have insisted (and let us grant that they do not play the filling-in-and-rounding-out role just mentioned). But Quine recognizes this and takes steps to integrate them into his world-picture. In ‘On the Nature of Moral Values’, he describes an empirical third-person test for detecting values in an individual subject, which, at the level of brass tacks, involves a chicken pecking at levers as in a Skinner experiment (1981c: 55–9; presumably Quine is speaking of even the most selfish non-transitory desires as ‘values’). Moving up the phylogenetic scale, he describes a process whereby what might begin as an egocentric value is instrumental to the acceptance of a moral value (Quine 1981c: 57):

We learn by induction that one sort of event tends to lead to another that we prize; and then by a process of transfer we may come to prize the former not only as a means but for itself… The transmutation of means into ends…is what underlies moral training. Many sorts of good behavior have a low initial rating on the valuation scale and are indulged in at first only for their inductive links to higher ends… But by association of means with ends we come gradually to accord this behavior a higher intrinsic rating. We find satisfaction in engaging in it and we come to encourage it in others. Our moral training has succeeded.

It is a mark of moral values that they are apt to “blend in social harmony” (1981c: 60), and indeed there are biological and sociological reasons not to rule out that some should turn out to be universal (1981c: 61–2). But any corresponding agreed statements — those expressing value as illustrated by the form “x is good” — are not themselves going to slot into Quine’s framework of empirical science, having observation sentences as its base. Quine writes: “we have to deplore the irreparable lack of the empirical checkpoints that are the solace of the scientist” (1981c: 66; see also 1974: 50–2, and for a consonant view of an ethicist, see Stevenson 1963: 214–20).

In his essay on Austin, Quine writes that Tarski’s paradigm “works for evaluations…as well as for statements of fact” (1981d: 90; he goes on: “it works equally for performatives. ‘Slander is evil’ is true if and only if slander is evil, and [likewise] ‘I bid you good morning’ is true of us on a given occasion if and only if, on that occasion, I bid you good morning”). Value statements fit the normal grammatical template of the declarative sentence, and indeed are susceptible to logic as indeed we typically so treat them — as when we demand coherence of them and seem to argue about them — yet they are not themselves science (Quine 1981c: 63). The relevant scientific facts by contrast will be facts of psychology or sociology, in the form, say, of the merely descriptive “x values y”, not the statements which themselves express value. Statements which express value will be excluded from science, from the overall theory that limns reality, but as matter of their lack of empirical checkpoints, not as a matter of their failure to be truth-apt. The Frege-Geach conundrum is simply not recognized. Theoretical statements which only serve to fill in their particular branch of science, meanwhile, are included in science but only for pragmatic reasons if not merely by courtesy. With either type of statement, it would be an overreaction to restrict the notion of truth as Bergström recommends. We may stick to Quine’s strict epistemological scheme as setting forth the language-game of science but acquiesce in Quine’s liberal Tarskianism when it comes to truth.

IIc. The Pursuit of Truth

Related to the foregoing in its emphasis on value is a further consideration to which Bergström appeals in making his case, touched on in the penultimate paragraph of section II in the exposition of the basics of Bergström’s view. It concerns the very idea of the Pursuit of Truth: why is it that truth is something valuable, something worthy of pursuit? He writes (1994: 433):

If disquotation is all there is to truth, it is very hard to see why truth is valuable or why truth is something that we want to pursue… Why should anyone attempt to find new sentences which are true in a disquotational sense? This is hard to explain. On the empiricist account, however, it seems quite reasonable that we should try to find theories which are true. For if the sentences we believe to be true are true in the empiricist sense, our expectations will tend to be satisfied. This can be taken as an indication that empiricist truth is more interesting than disquotational truth.

In his reply to Bergström, Quine answers (1994a: 498):

We choose to pursue truths conducive to our well-being and that of other deserving people, and truths that gratify our curiosity about the world. As for just how a true sentence can serve any such purpose, that varies radically from sentence to sentence.

But one could well wonder on behalf of Bergström why “gratifying our curiosity” would not suffice in answer, or some other general answer relating our cognitive states to reality; would this not have repercussions for the theory of truth? My explanation has two parts, the first trading on what I said in Ib and Ic, commenting further on what I take to be a central lesson Quine got from Tarski.

The lesson was that in order to explain the use of the scientific predicate ‘is true’, mere disquotation is not sufficient. Quine accepts the need for genuine semantic ascent, for generalization with “is true” and hence the need for a full-on Tarski-style apparatus. He thus concedes that there is nothing wrong with this sort of general question on the score of logical syntax. The first part of the answer then is to point out that the intelligibility of such questions does not mean that every such question is a good question, or that such breezy formulations as they “gratify our curiosity” really tell us anything definite or material. Indeed, as philosophers we do have a tendency to over-generalize, to submit to what Wittgenstein memorably called the “craving for generality” (1969: 17). In some cases, and this one included, we imagine there are general, pressing questions that have substantive answers, when in fact at anything but a superficial level there are only multitudinous answers to less sweeping questions. “In sober fact the pursuit resolves into concern with particular sentences, ones important to us in one or another way”, to re-quote Quine (FSS: 67).

But Bergström might just hold his ground here, insisting that his answer — or something like it — is indeed of the requisite weight. The second and more substantial part of my answer begins with a fulsome acknowledgement that it is a major achievement for Quine to have schematized empirical science in a completely general manner. But again, as in subsection IIa, that we accept a certain theory — such as Bergström’s — does not imply that it plays a role in defining the terms it uses. In particular, Quine might be said to articulate what we do in doing science so that our “expectations will tend to be satisfied” (certainly there was no assurance to begin with that such a thing is possible, as Peter Hylton has in effect stressed in 2007: 365). However — and this is the core of the second part of my answer — it is wrong to think that an analysis of truth will tell us what those satisfactions will be in general. Beyond platitudes such as “Our curiosity is gratified!” or “We find out the way the world is!”, there is no doubt a subject for psychology or epistemology involving these attitudes. But truth itself, once again, is fully explained via its generalizing function.

IId. Realism

Finally, I will consider Bergström’s remarks on the perhaps more contentious matter of Quine’s attitude towards realism, especially in view of its connection with the earlier theme of the immanence of truth. Bergström writes, “Certain passages in Quine’s writings suggest that he has a realist account of truth” (1994: 425). He then quotes Quine as saying, “What is it that makes one complete physical theory true and another false? I can only answer, with unhelpful realism, that it is the nature of the world” (1981d: 179–80); and that the truth of a single sentence “consists in the world’s being as the sentence says” (PT: 81). “These are”, Bergström concludes, “formulations one might expect from someone who uses ‘true’ in a realist sense. They seem to point in the direction of a correspondence theory” (1994: 425).¹⁹

Quine responds to Bergström’s question with the generalizing function of truth: “I am a realist about truth in whatever sense I am a realist about light rays or straightness” (1994a: 498. But to those who take the question seriously, this might seem evasive, if not smug. What follows takes slightly more pains.

There are three points, which I present in ascending order of consequence. First, the passage from Bergström underplays Quine’s scepticism, his subtle irony about certain philosophical questions or turns of phrase. For example, Quine meets the unhelpful turn of phrase that something “makes” a theory true with another unhelpful turn of phrase that it is the “nature of the world” that does it (PT: 179–80). He is merely talking the talk without quite taking it seriously (see also PL: 96, where — in a book about logic — he speaks of questions that are “all sound, signifying nothing”). He is more serious when he adopts in response to Davidson’s “idiom of realism” (1994b; 500), avowing that we have indeed a “remarkable feature of our use of the truth-predicate”, but what licenses him is merely the logic of truth, not an inscrutable connection with the idea of the real (I refer the reader to the above remark about light-rays). Second, I play yet again the card of the generalizing function of truth. To one who persists in asking, “What is it that makes one complete physical theory true and another false?”, Quine can direct the questioner to a specific example — “What makes the earth round?” — and cite the relevant cosmological, geological, or astronomical facts; if the questioner protests that that is not what was being asked, Quine can reply with the Tarskian account of the generalizing function; if the questioner persists still, Quine can reply that there is no answer, that that question ultimately fizzles out.

Third, Quine does call himself a “realist”, but one must not be misled. According to his naturalism, there is no standpoint besides the scientific one from which to judge whether certain entities are real, whether they exist, whether the procedures that issue in our theory of such entities are fully objective, how such entities stand to one another, and so on. In ‘Posits and Reality’ (Quine 1976a: 246–54), in an important point that strongly parallels both Austin and Wittgenstein (and also, if less strongly, Moore, the early Ayer, and perhaps Carnap), judgements involving such concepts as existence and reality do not somehow reach categorically beyond the significance that is accorded to them in ordinary language training (Quine 1976b: 251–4; also 1963[1951]: 44; 1960: 22; 2008b: 152; 2008a: 405). One learns to call narwhals real, unicorns unreal, and that is pretty much that. For Quine, the hyper-realist idea of things-in-themselves quite apart from our theories about them is an empty fantasy, playing no serious role in science or epistemology.²⁰