Comments on Bradley Armour-Garb: ‘Goodness Deflated’

Bradley Armour Garb objects to a parallel I drew, between the deflationary explanation of the behavior of the truth predicate (or equally the truth operator: it is true that) and a similar explanation, as I saw it, that could be offered for the use of the ‘good’ predicate. These explanations were not supposed to be speculative pieces of history, or accounts of the evolution of language. They pointed to similar useful functions, rather than making the additional claim that it is those utilities that drove the evolution. Other psychological or social needs may have done that. But as I saw it there is an account of the utility of ‘true’, and an interestingly similar account of the utility of ‘good’. A language without a term for truth would be impoverished in certain important ways; a language without a term doing something the term for good enables us to do, would similarly be impoverished in important ways.

            I never thought or said that the parallel would be exact in all respects. For a start there is nothing in our use of ‘good’ that is directly corresponding to the biconditionals that play such a large role in discussions of the utility of truth: it is true that p iff p; the sentence S is trueLiff p, and so on. Then ‘good’ has attributive uses (good tennis player, good husband…) which some have thought fundamental, whereas ‘true’ has relatively marginal claims to attributive uses: ‘what he said was true, but it did not add up to a true account of the events’, for instance, suggests the same kind of difference between true propositions and true accounts as might hold between ‘good husband’ versus ‘good cyclist’. But the usage is marginal, at best. Furthermore the impoverished language I imagined as a starting point was rather different: it had a device for expressing endorsement (the H! operator), whereas a language without the truth predicate or operator need have nothing, since simply putting a sentence forward in the right way in the right context amounts to asserting it. It does not take a different operator or other lexical element to manage assertion. This is a fact about language that puzzled Frege, for one, and it certainly deserves reflection.

            However these points do not feature in the complaints that Armour-Garb levels against my suggestion. Rather he directs his fire against three things I said in the course of developing my analogy. The first claim he dislikes is my saying that like ‘true’, ‘good’ is a device of indirect reference and generalization. He includes in his target here the rider that the good predicate will indeed resist analysis, but since he directs no specific objections at that target, I shall not discuss it here. Second he dislikes the contention that like ‘true’ ‘good’ is a device of indirect (or as Armour-Garb prefers to call it, opaque) endorsement. The third major blunder is my saying that there is an exact parallel between the way that ‘true’ enables us to generalize and the way that ‘good’ does (p. 375). Like Armour-Garb, I shall take these in turn.

(1)  As Armour-Garb admits, I am following many deflationists, and indeed philosophers who are not deflationists, in noticing that the truth predicate enables us to do things that we cannot do without it. Like them (he cites David, Field and Horwich among others) I put this by saying that the truth predicate is a device of indirect reference, or generalization, or indirect endorsement. A related claim, as he says, is that it is a device of infinite conjunctions and disjunctions, although I was not particularly concerned with infinite sets in my paper, and shall put it to one side. In any event, Armour-Garb dislikes these ways of putting anything. The truth predicate, he tells us is not a device of  indirect endorsement, indirect reference, generalization or infinite conjunction and disjunction. Rather, although the truth predicate can be a device for doing these things (given some further resources, such as quantifiers and variables, or presumably ordinary language devices having similar powers), this is only because it is actually a device of doing something different, namely enabling semantic ascent and descent (Armour-Garb’s italics).

            The distinction between being a device of (say) indirect endorsement, and a device for indirect endorsement, yet not a device of it, but a device of something else, which nevertheless can be used for  indirect endorsement or the other functions, does not wear its meaning on its face, even when the prepositions are italicised.[1] Presumably what Armour-Garb has in mind is that it is only because it is a device of (for?) semantic ascent and descent that the truth predicate has these other useful functions. The image would be that its semantic ascent and descent-enabling properties are primary, and any other functions for it are secondary and arise out of those.

            I have two responses to this. First, it would not matter in the least to the analogy I was drawing. My aim was to find a parallel between some functions the ‘good’ predicate serves, and ones that the truth predicate serves. The parallel would not be impugned by finding that the truth predicate has other properties, and even ones that explain the utility in question. I can quite happily say: look, the truth predicate serves as a device of indirect reference, generalization, and opaque or indirect endorsement—and so does the good predicate, even if there is another, perhaps primary function that the truth predicate also serves, and which explains its utility in these respects. But second, I am surprised at Armour-Garb’s confidence in his direction of explanation. For he does not tell us why a device of semantic ascent and descent would be particularly useful except in facilitating the possibility of indirect reference, generalization, and opaque endorsement, (and neither did Quine). Why should we want to skip from referring to the properties of sentences or other bearers of content to facts about the world, or vice-versa except when these other functions are being served? [2] For what it is worth—and it is not important to my defence—I should have thought that we more readily need to say things like ‘he’ll tell you about banking, and you can take what he says as gospel’, or ‘Einstein’s theory is right, but I can’t remember the details’ more readily than we need to skip through quotation marks or other devices of reference to linguistic vehicles, in which case it would be plausible to suppose that semantic ascent and descent is just the handmaiden of these functions. I say more about semantic ascent and descent shortly.

            Armour-Garb then chides me (along with Field) for  claiming that the truth predicate is a device for opaque endorsement on the grounds that this implies that it is exclusively a device for opaque endorsement (p. 377). But this is not an implication of what I said (how could it be, since as Armour-Garb describes, I explicitly mentioned other functions, such as indirect reference and generalization?); furthermore Armour-Garb is in no position to insist that it is, since he himself has said (pp. 275–6) that ‘together with other machinery…the truth predicate can serve as a device for expressing opaque endorsement’.  Being or serving as something with one function does not preclude having other functions as well.


(2) We then turn to indirect or opaque endorsement: the analogy I sketched between being able to say that what Sally said was true (without identifying what she said) and that what Sally did was good (without identifying what she did). To enter on discussion of this we should know that Armour-Garb has identified a proposition

(T) <p> it T iff p

where in his words, the corner quotes ‘<’ and ‘>’ ‘nominalize any sentence that goes in for p’. He says that for deflationists relevant instances of this deflationary schema are both a priori and necessary. If they do hold this it must imply that ‘nominalizing a sentence’ is a rather specific process. For if the left hand side of this biconditional refers to a sentence in any ordinary sense, then instances of T will be neither a priori nor necessary. It is neither a priori nor necessary that the sentence ‘snow is white’ in the language used by English speakers is true iff snow is white. It is something foreigners need to learn by empirical acquaintance with the language (they cannot work it out in their armchairs), and it is not necessary, for a small tweak in the history would have easily led to it having been false.[3] The only natural way to read T as both a priori and necessary is to take it that the left hand side abbreviates ‘the proposition that p’, as it does for Horwich, for example.  But in that case there is no room for a notion of indirect or opaque commitment: if X believes <p>, then X believes p, and conversely. Nor is the biconditional any longer an example of semantic descent, since it no longer moves from a proposition about a vehicle of content,  to a distinct, worldly, proposition. It starts and finishes with the proposition that p.

            Leaving this aside, Armour-Garb then tells us that

Deflationists claim that if what Isabel said yesterday was that snow is white, and if you assert

       (1) What Isabel said yesterday is true

then, given the relevant instance of (T), together with what Isabel said, you thereby express a commitment to the whiteness of snow. Moreover it is through expressing this that you manage indirectly (indeed opaquely) to endorse what Isabel said. (p. 378)


But I doubt if deflationists do say this, and whether they should depends crucially on what is intended by ‘given’. The ambiguity is whether it is given to you, the person who asserts (1), and given in a transparent way that precludes failure of understanding, or whether it is just given in the sense that this is indeed what she said.[4] To put the matter clearly, do we have:

(A) X holds that what Isabel said yesterday is true, and is also fully aware that Isabel said yesterday that snow is white.

Or do we have

(B) Isabel said yesterday that snow is white, and X holds that what Isabel said yesterday is true.

Only (A) sees X committed to snow being white, either opaquely or indirectly or any other way. (B) is compatible with X having no beliefs about snow and no intention of committing himself to it being white, since there is no implication that X has any idea what Isabel said yesterday. He may have been sure that Isabel would have told the truth on prior inductive grounds.

            We have exactly the same possibilities transferred to ‘good’.

(A’) X holds that what Isabel did yesterday was good, and is also fully aware that Isabel stole Fred’s apple yesterday.

(B’) Isabel stole Fred’s apple yesterday, and X holds that what Isabel did yesterday was good.

Only (A’) commits X in any way to holding that stealing Fred’s apple was a good thing to do (i.e. commits A to admiring or endorsing Isabel’s theft). (B) has no such implication, since it is compatible with X having no inkling of what Isabel actually did, and having no inclination to admire or endorse acts of theft. He may have been sure of Isabel’s good behavior on prior inductive grounds.

            With this cleared up it is evident that the (A) cases are parallel, and so are the (B) cases.

            This analysis does suggest that for clarity it may be best to avoid both the locution of ‘indirect’ and ‘opaque’ commitment. In each B case the only commitment X currently has is to Isabel having yesterday been a truth-teller/good agent. This commitment makes X potentially vulnerable when or if it emerges in the one case what Isabel said (which may turn out to have been mind-blowingly false) or in the other case what she did (which may turn out to have been mind-blowingly wicked). In each case X would have the same kind of withdrawal to make, unless, of course he decides to brazen it out and declare his mind appropriately blown. X has made a commitment—he has stuck his neck out—and the revelation of what Isabel said, or did, can therefore embarrass him.

            It is appropriate to compare the difference between the (A) versions and the (B) versions with the de re/de dicto distinction, of which, when we reflect, it is indeed just an example. Consider:

(A’’) X holds that the man with the martini is an idiot, and is fully aware that the man with the Martini is her husband John

(B’’) The man with the martini is X’s husband John, and X holds that the man with the martini is an idiot

It would be infelicitous to hold that in (B’’) X is said to be ‘indirectly’ or ‘opaquely’ holding that her husband John is an idiot. Her commitment extends no further than the quantification, a de dicto proposition given by a Russellian definite description: she holds that there is a man, and not more than one, who has a martini and who is an idiot. Holding this she is very likely due for a surprise and potential need to recant when she learns who he is (de re). The situation with the descriptions ‘what Isabel said’ or ‘what Isabel did’ is the same. This implies, indeed, that it is better to get rid of the idea of a ‘device of indirect reference’ since in ordinary cases of definite descriptions it is not helpful, and it may be positively misleading to say that X indirectly referred to the actual satisfier of the description (although I do not think that in the case of truth the locution has led David, Field, Horwich or others into making specific mistakes). In any case, the content of her commitment is quite different. So all that needs saying is that each of ‘true’ and ‘good’ can be used together with definite descriptions to identify de dicto commitments, and this can be done when it is unknown which proposition, or which deed, is in fact the one satisfying the description, and that in each case this gives us an expansion of the functions of the language. Because the proposition is unknown, there is no implication that it is believed by the subject, and since the deed is unknown there is no implication that it is the object of admiration by the subject. The parallelism which was the point of my paper remains.


(3) Armour-Garb also dislikes what I say about generalization. When deflationists say that the truth predicate is, or serves as, or is a device for, or of, generalization, they have in mind that rather than give a long conjunction of propositons of the form ‘Isabel said yesterday that p and p’ together with a closure clause saying that we have included everything that Isabel said yesterday, we can sum up and say that everything Isabel said yesterday was true. I said that there is a parallel: if someone expresses admiration for Isabel’s doing D and…admiration for her doing DN and is committed to D… DN encompassing all that she did, then he can sum up and say that everything Isabel did yesterday was good.

            Armour-Garb protests that in the first case we have a deduction, but in the second case this is not so.

            Now it is true that as he presents it there is an asymmetry. The case with truth is given by a list of propositions that are the case: ‘Isabel said that p and p’. The case with good is given only by a list of propositions describing the doings of Isabel that a subject, X admires, or “hoorays”. This is inevitable for an expressivist, since one is not starting with descriptions: one is starting with attitudes, and it is only after the elaboration one is trying to explain (the arrival of a predicate) that descriptions arrive on the scene. And a moment’s thought shows us how to restore the parallel. Simply give the case with truth in the same form, in terms of the commitments of a person X. We then get:

Truth as a device of (for) generalization: If a person X holds a conjunction of propositions of the form ‘Isabel said that p and p’ and is committed to supposing that the conjunction encompasses everything Isabel said yesterday, he can sum up, inferring that everything Isabel said yesterday is true.

Good as a device of (for) generalization: If a person X  expresses admiration for Isabel’s doing D and…admiration for her doing DN and is committed to D… DN encompassing all that she did yesterday, then he can sum up, inferring that everything Isabel did yesterday was good.

            There is nothing ad hoc about this. The expressivist obeys Peirce’s excellent maxim: ‘We must not begin by talking of pure ideas—vagabond thoughts that tramp the public highways without any human habitation—but must begin with men and their conversation (CP 8, 112). Inferences are, in the first place, movements of mind that people make. In each case there is an inference, or what I called a summation, book-keeping the commitments of the subject, and the parallel is exact. The truth predicate, and the good predicate, alike give us this useful ability, which we would not have without them. They may have other functions, but it is undeniable that the each share this one.

Simon Blackburn




[1] Now that I am sensitized to the question of whether of and for are substitutable, I find it odd that the preposition of ever seemed natural. It is hard to find other cases. A lawnmower is not a device of mowing lawns, although perhaps money is a device of exchange or of  stored value, although is it not also a device for both these things? I cannot think of a natural case where x is a device of y, but not for y, or vice-versa, except in irrelevant uses, as when one uses the genitive: money is a device of the devil, for example.

[2]  Armour-Garb does not notice the pitfalls surrounding semantic ascent and descent, arising from the fact that it is contingent that sentences have the meanings they do. Pace Quine, skipping up and down from sentences to content and back used to be called use-mention confusion, and care needs to be taken to avoid it.

[3] Of course, it is possible to define ‘English’ in terms of sentence/content pairs, so that it becomes necessary and definitional that “snow is white” in English is true iff snow is white. It is then contingent that English is the language anybody speaks now or ever did speak, as opposed to some trivial variant EnglishĘ in which the sentence meant something else. And in any case the left-hand side of the biconditional is then not strictly referring to a sentence (as it might be, an inscription or phonetic structure) but a hybrid object identified partly by a meaning. So it would be misleading to say that we have semantic descent or ascent even in this event. As a small matter of history, the first philosopher to be utterly clear about this was Carnap, in his 1942 volume,

[4] I bypass issues about what might be meant by ‘given the relevant instance of (T)’, since if (T) is a priori and necessary for deflationists, there will be no need to give it as an additional premise in a deductively valid argument.