Not to keep anyone in suspense, my answer is ‘no’. But perhaps suspense was never a live possibility—for you might have thought it pretty obvious that a militant deflationist such as myself couldn’t possibly regard “true” as a normative term. After all, we deflationists claim that the meaning of this word is fully given by the Equivalence Schema, “It’s true that \(p\equiv \hbox {p}\)”—a principle that appears to make no use of ought or good or of any other normative concept.Footnote 1

Conversely: you might have thought that anyone who believes truth is normative must have in mind Michael Dummett’s famous 1959 objection to deflationism—an objection that has been elaborated more recently by Wright (1992), Price (1988), and Lynch (2004). Dummett (1959) argued that the Equivalence Schema fails to capture the concept of truth, since our endorsement of it won’t explain the enormously important fact that we value true belief and assertion—so truth will have to be defined (at least in part) by some normative commitment: perhaps, that the proper aim of belief is truth, or that we ought to try to ensure that our beliefs will be true. Again, the implication is that no deflationist—no-one who rejects Dummett’s reasoning and who therefore continues to hold that Equivalence Schema is all that’s needed to define truth—could maintain that the concept is normative.

But matters are not so straightforward. I’ll explain in a little while why I disagree with these two correlated points. (This postponement is my feeble attempt to re-inject some suspense into the paper!).

To start with, it’s necessary to settle what’s going to be meant here by saying that a given concept is, or isn’t, “normative”. For philosophers have construed such talk in a variety of ways. And how can our main question be sensibly addressed without a definite idea of what that question is?

To that end, a vital first step is to distinguish between the following two senses of “normative concept”.—On the one hand, a given concept may be a concept used to evaluate (thereby making the term that expresses it a normative term); ought and good are paradigm examples of such functionally normative concepts. And, on the other hand, the application of a concept that isn’t evaluative, so isn’t functionally normative, may well have normative significance. Its application to a thing may imply that the thing is good, or ought to be done—or, more generally, that the thing falls under a concept that is functionally normative.

For instance killing isn’t a functionally normative concept; we can fully explicate it in purely naturalistic terms (roughly, as  causing a death).  But its application evidently has normative significance—killing a person is prima facie wrong. Similarly, umbrella isn’t a functionally normative concept, but its referent has normative import since an umbrella is good to have in the rain.

Clearly, just about everything is subject to evaluation of one kind or another; so it would be surprising if truth were an exception; and of course it isn’t. In particular, and as Dummett rightly observed, there’s something normatively positive about true beliefs; they are the ones that are correct; they are the beliefs we should to prefer to have. So, like kill and umbrella, the concept truth surely does stand for something normatively significant.

But this doesn’t show that it’s a normative concept in the way that ought is, and that our word “true” is a normative term. For no matter how obvious and general and basic the normative significance of a concept’s referent might be, that won’t make the concept a functionally normative one. But it’s functional normativity, not mere normative import, that’s our present concern. Whether a concept is functional normativity is clearly the more controversial and interesting question about it—the question I think most philosophers have had in mind when they have asked whether a given concept is normative—e.g. meaning, or pain, or belief, or truth.Footnote 2

So what would make truth a functionally normative concept? (Henceforth, I won’t usually bother to insert the “functionally”). What’s the general criterion for being a concept of that sort? I’ve mentioned a couple of paradigm examples—ought and good. But how can we tell which other notions are also normative?

Here we should distinguish two kinds of answer. One kind is relatively superficial and the other more profound. A superficial account would tell us that a concept is normative just in case it bears a specified intimate relation to one or another explicitly designated, paradigmatically normative concept, such as ought. Clearly such a story simply assumes that ought is normative and makes no attempt to explain why it is.

A deeper account would not explicitly presuppose that ought (or any other designated concept) is normative, but would provide a general criterion purporting to identify the feature of any normative concept (including ought) that’s responsible for it being normative.

Although an account of the more profound kind would obviously be more satisfying, it would also be harder to find and defend. And it would be nice to get to our primary topic with a minimum of further delay. So let’s be content, for the purpose at hand, with a superficial account.

But what should it be? Let’s suppose that ought is the sole fundamental normative concept. This assumption isn’t really needed for my line of thought, but it simplifies it. So our question becomes: what relation must a concept bear to the concept ought in order to qualify as a normative concept?

Before attempting to answer that question, it’s worth noting that there’s more than one concept expressed by the word “ought”—even after we’ve restricted its application to a particular domain of discourse: e.g. to moral contexts, or to epistemology, or to practical rationality.

For example, suppose that someone opts for a certain medical procedure, which turns out to leave him much worse off than he was before. If he says, “I ought not to have done it”, he might be admitting that he was irrational and subject to criticism for not having properly taken into account the solid evidence for regarding the procedure as extremely risky. Or he might instead mean merely that the operation happened, unpredictably, to have unwanted consequences—conceding, not that he was irrational, but merely that if he had known all the relevant facts (including how the operation would turn out) then it would have been irrational for him to have agreed to it. This second sense of “ought” (the so-called objective ought) might in that way be explained in terms of the first (the subjective ought), which is the one immediately tied to motivation and criticism. Anyway, the assumption I’m making is that it’s a concept’s special relationship to the subjective ought that will qualify it as functionally normative. (In what follows, when I talk about the concept ought I’ll mean the subjective concept).Footnote 3

But what should the account be? I can think of three tempting alternatives:

  1. (a)

    A concept is normative iff its analysis invokes ought. (Thus, if the analysis of x  is good  is we  ought to want x,  then good qualifies as normative).

  2. (b)

    A concept is normative iff simple applications of it entail claims of the form ‘x ought ought to be done’ or ‘x ought not to be done’. (For example, perhaps  x is  delicious entails  one ought to like the taste of x).

  3. (c)

    A concept is normative iff a person can fully possess it only in virtue of deploying the concept ought. (For example, perhaps full possession of the concept generous requires having an inclination to explicitly believe that one ought to be generous).

But I don’t like (a) because it requires that every normative concept, except ought, must be explicitly definable. And if we philosophers have learned anything at all since Plato, it’s that explicitly definable concepts are few and far between.

And I don’t like (b) because, although it’s perhaps less implausible that (a), still I don’t see how we can be confident that every simple application of a normative concept entails that sort of ought-proposition.

So—acknowledging that this issue calls for more discussion—let me settle on (c). Notice that in making this choice I’m bending over backwards to be concessive to those who suspect that truth is normative—because, quite plausibly, anything that’s normative according to either (a) or (b) is also normative according to (c)—but not the other way around.

Now at last we’re in a position to address our central question:—Is truth a normative concept? And let me start by defending the pair of unjustified claims I made at the very beginning of this paper: namely, that we should not go along with the impression that deflationary views of truth obviously imply its non-normativity. And nor should we think that truth could be functionally normative only if, as Dummett maintained, grasping it requires explicit recognition of the value of true belief.

Both of these negative claims of mine derive from the observation that although the Equivalence Schema, “\({<}\hbox {p}{>}\) is true \(\leftrightarrow \hbox {p}\)”, doesn’t deploy any obviously normative concepts, still it clearly does deploy the concept of proposition.Footnote 4 But our recognition that “p” expresses the proposition that p—i.e. that this proposition is the meaning of “p”—is essential to our grasp of that concept. So, if meaning is normative (as quite a lot of philosophers have insisted it is), then so it proposition. And truth, will inherit that normativity. For someone will be able to accept the Equivalence Schema (and thereby fully possess the concept truth) only if she already possesses the concept meaning, which will in turn presuppose possession of ought. This shows not only that a deflationist might well come to accept that truth is normative, but also that a normativist about truth needn’t base his position on the Dummettian anti-deflationist idea that the value of truth is built into the concept.

So, at this point, we have two potential routes to the normativity of truth. Let me first consider the more familiar route—namely Dummett’s. As we’ve seen, the contention here is not merely that true belief is valuable. Nor is it that this value is universally recognized. The crucial thesis is that it must be recognized by a given person in order for her to qualify as fully possessing the concept truth.

Now I can’t think of any good reason to think that this thesis is correct. Yes, there’s an extreme and implausible position known as “meaning-holism”, whereby everything we say with the help of a given predicate, “f”, contributes to its meaning. But leaving that aside, we’ll always need definite grounds, given a specific commitment of ours concerning the property of f-ness, to suppose that this commitment is required for possession of the concept, F. In particular, the mere fact that we recognize the value of true belief isn’t enough to show that respect for some such norm plays a role in grasping the concept of truth.

Such grounds would be present if it struck us as trivially analytic that truth ought to be pursued and untruth avoided. But we don’t have that impression. Consider the following competing interpretations of the statement, “Murder is wrong”. On one conceivable account, the word “murder” means “wrongful killing”—in which case the concept murder is functionally normative, the statement is true by definition, and so it has no normative force. On another conceivable account, “murder” means something roughly along the lines of, “premeditated killing of someone against his or her will”. And taken in that sense, “Murder is wrong” clearly would not be analytic and clearly would have normative force. Like many simple substantive norms it would begin by picking out some phenomenon in entirely descriptive, naturalistic terms, and then proceed to evaluate this phenomenon (as good, or bad, or obligatory, or permissible, or correct, etc.).

And the idea that truth is desirable is surely in the second category. It’s something we try to instill in our children by means of encouragement and criticism—just as we do in the case of other substantive moral norms, pragmatic norms, and epistemic norms. We’re surely not teaching them the meaning of the word “true”. The point of the education is character building not linguistic expertise.

Further evidence against the Dummettian brand of normativism about truth derives from an appealing way—an almost irresistible way—of explaining the role of truth in ‘value of truth’ principles. This concept is doing what it’s nearly always doing for us—serving as a device of generalization.

Consider, for example, the logical facts expressed by “Av-A”, “Bv-B”, “Cv-C”, ... and so on (where “A”, “B”, etc. abbreviate particular English sentences). Evidently there’s some general logical fact in the offing—something that captures all those particular ones. But how are we to state it precisely? Surely not with two or three conjunctions of them, followed by the indefinite and imprecise, “. . ., and so on”. The answer is that our concept of truth is tailor-made to solve this sort of problem. We can simply say, “Every instance of <pv-p> is true”. And this captures all the particular facts we wanted to capture. For it yields, “<Av-A> is true”, which—in virtue of the Equivalence Schema - entails “Av-A”. And by parallel reasoning it also gives us “Bv-B”, “Cv-C”, and all of the other instances.

Similarly, we can very naturally regard a value-of-truth principle—for example

  • If something is true, then it’s better to believe it than to

  • disbelieve it

as exploiting the concept of truth merely to make explicit our disposition to accept indefinitely many proposition-specific norms such as

  • If \(\hbox {E}=\hbox {mc}^{2}\), then believing that \(E={ mc}^{2}\) is better than

  • disbelieving it

  • If pigs can fly, then believing that pigs can fly is better than

  • disbelieving it

and the like. No employment of truth in those. It’s only brought into the picture, as in the case of the Law of Excluded Middle, so we can package up all the specific commitments into a single precise and stateable principle.

And exactly the same account can be given of all the other plausible value-of-truth principles—for example, “A belief is correct just in case it’s true” and “One ought to want that one believes only what is true”.

It seems to me that this confirms what was earlier suggested in our discussion of the non-triviality of belief-truth norms: namely that Dummett was mistaken—it’s not a condition on full possession of concept truth that we appreciate the value of truth.

Now I was careful to say that my deflationist story about the role of truth in value-of-truth principles is only almost irresistible—because it has in fact been resisted, and in various ways. (No doubt some valiant members of the resistance are included in the present collection!) But, for now I’ll have to confine myself to just one of their arguments—one that’s been emphasized by Michael Lynch.

He rightly points out that, on my account, the epistemological relationship between a general belief-truth norm, for example:

  • We should want that our beliefs be true

and the specific norms that it encompasses—e.g.

  • We should want that (we believe that pigs can fly only if pigs can fly)

is that we first come to believe some of the latter (including, let’s suppose, the very example I just gave); we then bring in the notion of truth via the Equivalence Schema to get, for example,

  • We should want that (we believe <pigs can fly> only if <pigs can fly> is true)

and we finally arrive at the universal value-of-truth principle by generalizing such proposition-specific normative statements.

But Lynch’s objection is that this proposed inferential route makes a mystery of our alleged premises. Why would anyone accept any of them? Why, for example, accept the premise that we should want that we believe that pigs can fly only if pigs can fly? Isn’t it much more plausible, he suggests, that the correct epistemological order is exactly the opposite of mine: that we believe these particular things by inferring them from the value-of-truth generalization?

Well I continue to think not—and for the following reasons.

First, it’s typically the case that generalizations are justified on the basis of some of their independently credible instances—and it would be strange if our belief-truth norms were exceptional in this regard.

Second, I grant that in the case of such norms it’s harder than it is in many other cases (e.g. “All ravens are black”), to recognize that it can be independently rational for us to accept their instances. But I think that this sense of relative difficulty stems entirely from the fact that we tend to be especially confused and puzzled by the rationality of epistemologically basic judgments when they are a priori.

However, the fact remains that there are rational judgments of that sort; and if we are determined to be puzzled about this phenomenon, it won’t help to suppose, with Lynch, that they concern generalizations. We might as well allow (i) that we have dispositions to believe the instances of certain schemata; (ii) that, in the case of some of these schemata, the underived beliefs that instantiate them are perfectly rational, despite being justified neither by argument nor experience; and (iii) that these schemata include those (such as “Only if p, is it correct to believe that p” and “One ought to desire that one believes that p, only if p”) that are easily rearticulated, thanks to the Equivalence Schema, into our belief-truth norms.

Needless to say, it would be absurd to look for ways of justifying the underived beliefs, or justifying our disposition to have them. The most we can hope for is to explain why we have them. And that sort of understanding does seem to be achievable. The explanation could well be roughly along the following lines.—So many cases of a person’s ‘believing that p only when p’ tend to promote his or her goals, and so many beliefs are acquired via testimony, that communities, simplifying for the sake of effectiveness, inculcate a disposition to accept all instances of ‘I should try to ensure that I believe that p only if p’—where the “should” is no longer to be construed instrumentally.Footnote 5 So it seems to me that we are able to get a rough sense of why we are inclined to accept specific instances of a belief-truth normative generalization without deducing them from it.

So much for why I think we must reject the Dummettian rationale for supposing that truth is functionally normative.

Let me now turn to the second above-sketched route to this conclusion.—It’s that the truth-defining Equivalence Schema is, appearances to the contrary, a normative principle in virtue of its deployment of the concept of proposition, which depends on the concept of meaning, which is normative.

The view that meaning is normative has many illustrious advocates—for example, Kripke (1982), McDowell (1984), Gibbard (1994, (2012), Brandon (1994), Hawthorne and Lance (1997), Boghossian (2005), and Ginsborg (2011). But are they right? I think not. But before I explain why, I’d like to digress for a minute to try to persuade you that, even if they are right—even if truth is normative in virtue of presupposing the concept of meaning—it would only be a little bit normative.

That’s because the above-emphasized function of truth—its being a device of generalization—can be performed (although admittedly not quite as well) by a somewhat weaker concept—namely, a concept of idiolectal sentential  truth, \(\textsc {truth}^{*}_{\mathrm{ME}}\), that’s captured by the disquotation schema

  • My “p” is \(\hbox {true}^{*}_{\mathrm{me}} \leftrightarrow \hbox {p}\)

This schema can be applied by each individual exclusively to the sentences of her own idiolect. It involves no meaning-related notions, hence no suspicion of normativity. However—since its attributions of \(\textsc {truth}^{*}_{\mathrm{ME}}\) to utterances are equivalent to the utterances themselves—it will nonetheless provide a much needed way of expanding the range of phenomena about which generalizations can be formulated (including logical phenomena, as we have seen).

Still, this device is somewhat limited. It doesn’t specify the circumstances in which sentences beyond one’s own idiolect are true*. Therefore, it can’t be used to express agreement (“That’s true”) with what someone else says or thinks. And, a fortiori, it can’t be used to capture generalizations about such matters (“None of Sid’s claims is true”).

It’s this defect that’s rectified by recourse to ‘that’-clauses (such as “that dogs bark”), which designate the propositions expressed by (i.e. the meanings of) their component sentences. Our possession of the concept, proposition, is grounded in our recognition of the circumstances in which we can reasonably accept instances of “Utterance, u, means that p”. And with that concept in play we can supplement the above disquotational device with the principle that a sentence is true* if and only if the proposition it expresses is true—arriving at the equivalence schema, “The proposition that p is true if and only if p”. We’re then in a position to generalize instances of those schemata that concern the assertions and thoughts of others (—such as, “If Sid claims that p, then not p”).Footnote 6

Thus meaning isn’t involved in the heart of the concept of truth—only in a secondary peripheral part of it. That’s what I mean by saying that even if meaning is normative that wouldn’t make truth normative through-and-through, but only a little bit.

But is meaning normative in the first place? Let’s compare the potential reasons to say “yes” with the potential reasons to say “no”.

Applying the conclusions reached at the outset of this paper about what it takes in general for a word (or for the concept it expresses) to be normative:—one can establish that “meaning” is normative only by showing that a full understanding of this term rests in part on an understanding of “ought”. But how might that be done?

One possible strategy (adopted by Gibbard 2012) is to try to make it plausible that any attribution of meaning to a given word analytically entails a simple proposition about how that word ought to be used. Perhaps, for example: <S’s word w means and> entails <S ought to use w as we ought to use “and”>. Or perhaps, more specifically, it entails <S ought not to reject a given sentence, u, whilst accepting the result of operating with w on a pair of sentences, one of which is u>.

But convincing evidence for such analytic entailment claims is hard to come by. For it wouldn’t be enough that the corresponding material conditionals be true—or even that they be both necessarily true and a priori knowable. All the mere truth of those conditionals would show is that meanings, like umbrellas, have normative import. And all that their necessary and/or a priori truth would show is the necessity and/or apriority of their normative import.

Analytic entailment is another matter, and there’s reason to doubt that meaning-attributions bear such an extremely tight relation to simple ought-propositions. For one thing, attributions of the form, <Our word “f” means F>, are trivial—surely not informative enough to analytically entail debatable contentions as to how “f” should be used. And for another thing, specifications of the distinctive normative imports of distinct meaning-attributions appear to have substantial critical force—they are used to correct a person’s usage. But, as we saw in the context of our initial discussion of ‘value of truth’ norms, analytic entailments have no such force. (Our paradigm example was that if “x murdered y” analytically entails “x acted wrongly”, then “murder is wrong” doesn’t articulate a genuine norm).

A second way of trying to show that “meaning” is a normative term would be to argue that the meaning-constituting fact about that word is some regularity in how it’s used in relation to the word “ought”. (Perhaps it’s our tendency to accept the material conditional, “If w means the same as v, then w and v ought to be used in the same way”.) But again, supporting evidence for the general idea is elusive. And as for the specific proposal used to illustrate the idea, a point against it is that the alleged meaning-constituting fact is our acceptance of a principle that appears to have substantive normative force. But it’s hard to think of any substantive norm that must be respected in order for the expression designating the evaluated phenomenon to be fully understood. For example, we don’t need to agree that killing a person is prima facie wrong in order to fully understand the expression, “killing a person”. Similarly, it would seem, a person’s failing to accept, “Words with the same meaning should be used in the same way”, still permits him to have a full understanding of “meaning”.

A third normativist strategy (mentioned in footnote 2 above) would be to argue that the properties of words that ground their meanings invariably concern how the words ought to be used. For it’s indeed hard to see how that could be so unless the superficial meaning facts to be constituted—i.e. facts articulated with the word “meaning”—were themselves normative.

But the initial claim here is quite implausible compared with its polar opposite:—that meaning-properties are always grounded non-normatively. For the most plausible view of their constitution will be the one that best squares with our central pre-theoretical convictions about what such properties are supposed to explain and how their presence is recognized. More specifically, it’s obvious common sense: (i) that, whenever we accept and utter a sentence, that is explained in part by what we mean by it (and therefore by the meanings of the words in it); and (ii) that we discover (via inference to the best explanation) that what others means by a certain term of theirs is just what we mean by a given word of our own, by observing that they tend to use theirs in the same basic way that we tend to use ours.Footnote 7

Thus we might well expect what grounds the meaning-property of a word to be the common factor in all of the thousands of explanations of the thousands of particular events that consist in one or another sentence containing the word being accepted. And if “meaning” is intended to designate a basic tendency of use then one can begin to see how this might be so, and one can thereby understand why the epistemological bases for meaning-attributions are what they are. However, on the face of it, a normative property of the word won’t do, since—on the face of it—how a word ought to be used cannot help to explain how it actually is used.Footnote 8

This emphasis on the naturalistic features of meaning is not to deny that some of our pre-theoretical convictions about it are normative—e.g. that if a word of ours means dog then we should want that we apply it only to dogs; and that if a symbol, “#”, of ours means and then inferences from “p # q” to “p” are legitimate. But as we have seen, there’s no reason to deny that these (like nearly all other normative propositions) are substantive—that they articulate the normative imports of phenomena (in this case, meanings) that are characterized in purely naturalistic terms. Nor is there any reason to doubt that our acceptance of such norms can be explained via the conative attitudes we have towards the meaning-properties so characterized.

We can conclude, I believe, that the meaning-properties of words are grounded naturalistically at every level, and therefore that the conception of w’s meaning that’s implicit in our pre-theoretical convictions is not well accommodated by supposing that it’s functionally normative. So I think we can safely say that meaning-attributions are not normative.Footnote 9—From which it follows that the conceptual dependence of truth on meaning won’t make truth intrinsically normative (not even a little bit).

Thus if deflationism is correct—if truth is captured by the schema, “The proposition that p is true iff p”—then it’s not a functionally normative concept. And there’s substantial reason to think that this schema is indeed fully adequate.—Certainly it’s better than any of the traditional definitions of truth, based for example on correspondence, provability, utility, or consensus.Footnote 10 Granted, we mustn’t neglect the Dummett-Wright-Price-Lynch pro-normativist and anti-deflationist considerations. But these fail on various grounds, as we have seen. So the bottom line is that our critical analyses of the two potential routes to the normativist view of truth (one via belief-truth norms, and the other via the alleged normativity of meaning) leave us, not only with no reason to adopt that view, but with good reason to reject it.Footnote 11