It has been nearly forty years since the publication of "Two Dogmas of Empiricism" (Quine [1951/1961]).  Despite some vigorous rebuttals during that interval,<1> Quine’s rejection of analyticity still prevails--in that philosophers en masse have either joined Quine in repudiating the "analytic"/"synthetic" distinction  or remained (however mutinously) silent and made no claims of analyticity.
     This comprehensive capitulation is somewhat surprising, in light of the radical nature of Quine’s views on linguistic meaning generally.  In particular, I doubt that many philosophers accept his doctrine of the indeterminacy of translation, which directly underwrites the rejection of analyticity even though it did not figure prominently in "Two Dogmas" but was made explicit only in Word and Object (Quine [1960]) and subsequent papers.<2>  (Indeterminacy of translation spells the death of individual word meaning as well as that of propositional or locutionary meaning, and so leaves no room for truth by virtue of word meaning as is required by analyticity.)
     In this chapter and the next I shall make a Quinean case against analyticity, without relying on the indeterminacy doctrine.  For I would like to join the majority in denying both analyticity and indeterminacy:  Contra Quine, I think there is determinate propositional meaning, as represented by a structured set of possible worlds, but with Quine I doubt that any sentence of a natural language is true solely in virtue of its meaning.<3>  However, we shall see that the attack on analyticity is partly blunted by the actual practice of stipulative definition, and we shall also have to concede a sense in which some sentences are "true in virtue of meaning," though it is not a philosophically important sense; those issues will be the focus of Chapter 12.

 1.  QUINE AGAINST ANALYTICITY

     As is well known, Quine begins by complaining that the "analytic"/"synthetic" distinction is imprecisely drawn.  It should by now be equally well known that that is the least of Quine’s complaints.  He further objects to the circularities that infest philosophers’ informal explanations of the "family circle" of intentional notions he is attacking: "analyticity," "synonymy," "contradiction in terms," and so on.  Grice and Strawson (1956), among many other commentators, objected that all systems of definitions have this "circular" character.  But Quine’s point is that ordinary systems of definitions normally contain at least one term which is antecedently and nontechnically understood, in which case the circularity (of course) does not matter, while among the family-circle words there is no such term--they are all technical.  (One might think that "synonymy" is nontechnically clear enough, but as Harman (1967, pp. 135-137) has emphasized, ordinary people use that word far more broadly and loosely than philosophers do.<4>  One might also think that "definition" is all right, but as we shall see, "definition" is a morass.)
     Yet the real objection to analyticity is neither the imprecision nor the unspecifiability of a distinction between analytic and synthetic; the first half of "Two Dogmas" was historically very misleading in this regard.<5>  Defining analyticity is not the problem.  In fact, analyticity is rather easily defined, as a strongly modalized variety of truth: not just nomologically or even metaphysically necessary truth, but "conceptual" truth, truth by virtue of lexical meaning alone.  A sentence is analytic iff its own meaning suffices to make it true, regardless of any (other) contribution from the world.
     This way of understanding analyticity has obvious epistemological consequences:  An analytic sentence would be unrevisable, in the sense that to deny or reject it would be eo ipso to abandon its standard meaning; one who called it false would be, as Quine says, not denying the doctrine but changing the subject.  Thus nothing could count as evidence against the truth expressed by an analytic sentence, and more generally we could have no rational grounds for doubting that truth (we could be mistaken only about the meanings of the relevant words).
     Quine’s real complaint is that analyticity thus understood is just unexemplified.  There are no analytic sentences, because no truth is necessary in any stronger sense than the nomological, and no belief is unrevisable.  (There is, if you like, "analyticity" relative to a theory on an axiomatization of that theory:  A sentence may be treated as analytic within a theory, by being used as an axiom and introduced in textbooks by an axiomatic definition.  But mere analyticity-within-a-theory-on-an-axiomatization is not what partisans of analyticity have had in mind.)
     So let us review some attempts to explain how a sentence can be both necessary in some stronger sense than that of physical or causal necessity ("logically," "conceptually," etc.), and epistemically unrevisable in the way described.  There are three main kinds of account expressed or presupposed in the literature.  The first two I shall mention but dismiss briskly.  It is the third that will detain us.

 2.  THE FIRST UNPROMISING ACCOUNT

     The first account is a straight shot: the idea that truth in virtue of meaning is just that.  Sentences have meanings; some meanings are such as to guarantee truth, and that is (virtually) all there is to it.  Call this the "Simple Account."
     At the time of "Two Dogmas" meanings were thought of as propositions, language-neutral abstract entities "expressed" by sentences.  The positing of propositions was supposed to explain the properties of language-bound sentences individually and in pairs--meaningfulness, ambiguity, synonymy, entailment, and the like--in terms of the relation of expressing.  Psychological facts about the propositional attitudes were complementarily to be explained with the aid of the further relation of "grasping" between a subject and a proposition.
     Understood in these terms, the Simple Account of analyticity has it that a sentence is analytic just in virtue of expressing a proposition of a special sort--the null proposition that rules out no contingent possibility.  Alternatively, one might say that although an analytic sentence is well-formed and trivially or degenerately true, it expresses no genuine proposition.  Either way, such a sentence--meaning what it does mean--cannot be false.
     Quine’s objection to this propositional theory of analyticity is that it invokes propositions.  His main mature objection to positing propositions is that if there really were propositions then translation would be determinate; yet translation is indeterminate.  However, I myself am sworn not to rely on the indeterminacy doctrine, and so must find other objections.  Quine himself makes two, and a third can be pressed on his behalf.
     The first objection is akin to the indeterminacy claim, but does not rely on it:  The proposition theory does not square with the translation habits of ordinary people and with their ordinary talk of synonymy and the rest.  In the real world, people count as "synonymous" any two sentences that serve roughly the same contextual purpose on an occasion, and such pairs will seldom be candidates for doing what philosophers call expressing the same proposition.  Ordinary people rather engage in the loose free-for-all practice that Quine calls "paraphrase" ([1960, sec. 33, p. 208, and secs. 53-54], and see Harman, p. 142).  Secondly, propositions themselves are entia non grata, "disorderly elements," "creatures of darkness," etc.; they are nonspatiotemporal and acausal, their identity conditions are inscrutable, and so forth (Quine [1960, Ch. Six; 1969c]).
     The third objection is emphasized on Quine’s behalf by Harman.  It is that propositions are flatus vocis or dormitive virtues, and do no explanatory work whatever.  ("Why are these two sentences synonymous?" -- "Because they express the same proposition." -- "Oh, I see.")  The proposition "theory" of meaning is a pseudo-theory, not a substantive theory at all.
     I disagree with Harman on this general point, for I think the theory of propositions is a genuine and even substantive theory, even though a very bad one (Lycan [1974]).  But there is a serious question of the explanatoriness of the proposition theory with respect to analyticity in particular.  A sentence is supposed to be true, and true in virtue of meaning, because it "expresses the Null Proposition."  Like Harman, I fail to discern the explanatory force of that "because."  Even if we know what "expressing" is, what do we know of the "Null Proposition" except that sentences that express it are supposed to be analytically or conceptually-necessarily true?
     One may of course attempt to answer that rhetorical question nonrhetorically, by offering a story about what "propositions" are that says in particular what makes the "Null Proposition" special.  The possible-worlds story comes instantly to mind, for propositions can be taken to be sets of worlds.  Indeed, David Lewis’ theory of analyticity is an instance of the Simple Account by way of propositions by way of possible worlds:  For Lewis, a sentence is analytic iff it is true at every world--and so expresses, though Lewis does not use the term, the Null Proposition, i.e., the proposition that excludes nothing.<6>  Thus Lewis avoids my foregoing criticism of the Simple Account, and advances the issue by explicating analyticity as truth in all possible worlds.  An analytic sentence is true in virtue of expressing the proposition that it does, the Null Proposition, for that proposition is the set of all worlds and is nowhere false.
     Lewis maintains that the truth of a proposition throughout all worlds is a mind-independent fact of the pluriverse.  Thus for him an analytic statement is true in virtue of the modal structure of the world, in virtue of what is objectively possible.   In a sense, then, although a Lewis-analytic sentence is true in virtue of its meaning, it is not true solely in virtue of that but partly in virtue of an objective feature of the pluriverse.<7>  That is perhaps only a quibble, but in any case, there ensues the more trenchant Quinean question:  Why should anyone think there are any sentences that are true in all possible worlds, if "possible" means something seriously weaker than nomologically possible?  Lewis’ theory makes no headway at all against this most basic misgiving.<8>
     Might one perhaps pursue the Simple Account without reifying meanings, i.e., without appeal to propositions at all?  That would be to hold that the truth of certain "necessary" sentences is still explained solely by reference to considerations of meaning, though now in some ontologically non-toxic sense.  But if the Simple Account is to be left at that, we are given no hint of how the explanation of truth by meaning alone could proceed.  We are told "how" sentences can be true by virtue of their meanings alone, only by being told that those special sentences’ meanings suffice to make them true.  Unless some entities comprising a mechanism are posited--at least propositions and/or possible worlds--no explanation emerges at all.

 3.  THE SECOND UNPROMISING ACCOUNT

     The second unpromising account of analyticity--which historically replaced the Simple Account so far as the latter modestly appeared in Russell and Moore--is the positivist/verificationist theory of  truth in virtue of unfalsifiability.  A sentence that is verified by every possible experience, or at least falsified by no possible experience, would be a candidate for analyticity.  And this Verificationist Account avoids Quine’s and Harman’s objections to the Simple Account, for it is genuinely substantive and explanatory.  Verificationism in general is an undeniably explanatory theory of meaning, even if a false one, and it entails in particular that if a sentence is radically unfalsifiable, there is something very special about that sentence’s meaning: so far as the sentence is true, it is true solely in virtue of meaning (for its meaning, i.e., its verification-condition, affords no room for falsity).
     There are basically three objections to the Verificationist Account of analyticity.  First, how could we ever know that a sentence had this character of in-principle unfalsifiability?  At least, how could we possibly know that a sentence was immune in this way to refutation, unless we had some deeper antecedent reason to think the sentence was impregnable, e.g., that it was already true by convention.  A look at the more accessible Positivist writings suggests that irrefutability is not itself the explanation of a truth’s "conceptual necessity"; rather it is explained by necessity, which is in turn explained by something more familiar.
     It is sometimes urged against such points that we know a sentence to be unfalsifiable simply because we cannot imagine or conceive the sentence’s denial.  Quine’s answer to someone who pleads just plain inability to imagine a thesis’ being false is simply that the person has a poor imagination.  What one can imagine at any given time is almost entirely dependent on one’s current theory, as the history of science and the history of ideas indicate; even the great intellectuals of the past were "unable to imagine" many of the things which today are commonplaces.
     In any case (second objection), the operative term in the slogans "verified by all possible experiences," "falsified by no possible experiences" and the like is, of course, "possible."  The Verificationist Account cannot do without this qualification, and "possible" is a member of the family circle.  Explaining necessary truth in terms of "possible experiences" is much like explaining it in terms of "possible worlds."  Once again we must ask why certain imagined experiences--e.g., as of seeing an object that is simultaneously red all over and green all over, or as of a square circle--are supposed to be impossible in any stronger sense than that of nomological impossibility.  To insist that such experiences just are inconceivable in that very strong sense  adds nothing to one’s initial insistence that the corresponding negative sentences are analytic.
     Quine’s most famous objection to the Verificationist Account (and my third) is a corollary of his Duhemian point against atomistic verificationism generally: that "statements about the external world face the tribunal of sense experience not individually but as a corporate body" ("Two Dogmas," p. 41).  Any single sentence of our total theory may be retained in the face of apparently recalcitrant experience, if enough more or less drastic revisions are made elsewhere in the theory.  Thus no individual sentence has its own specific verification-condition.  Someone might reply that while Duhem’s point is true of contingent and empirical sentences, it obviously fails for sentences which are true or false no matter what; but that would, as before, beg the question of whether there are any such sentences in the first place.
     Happily (for them), the Logical Positivists had a positive account of analyticity, in addition to experiential unfalsifiability, their negative criterion.  It was that analytic sentences are true by convention.  The Convention Account of analyticity is the third and last view that I shall consider.  Notice that it is entirely independent of the Verificationist Account, even though the Positivists held both; historically, it was preserved among the Ordinary Language philosophers who rejected verificationism, and continues to be accepted by at least one (otherwise) hard-core scientific realist, Armstrong (1981, especially Chs. 5 and 6).
     A self-respecting contemporary metaphysical realist and referential semanticist finds the idea of "truth by convention" grotesque.  How could a sentence be true, i.e., correspond to reality in a referential manner, entirely by convention?  But of course the Positivists and the Ordinary Language philosophers were not (thoroughgoingly) either metaphysical realists or referential semanticists.  More to the point, the idea of truth by convention is defensible and the Convention Account of analyticity is far more compelling than was either of its two predecessors.

 4.  TRUTH BY CONVENTION

     There are three main ideas of truth by convention.  One is that of lexical meaning as codified in authoritative dictionaries of natural languages.  The second is that of lexical rules or "meaning postulates" codified in formal languages.  The third is that of truth by stipulative definition.  In the rest of this chapter I shall discuss the first two; the third will be the topic of Chapter 12.
     The dictionary idea is that word meanings are conventional and that the governing conventions are codified by the entries in a good dictionary for the language in question.  Any "information" contained in or easily read off from a proper dictionary is thus purely conventional, and cannot be doubted or questioned by anyone who has mastered the language.
     An obvious objection to this idea is that, in the real world, dictionaries are not stone tablets bearing our conventions in axiomatic form.  Most entries in a typical dictionary do not even purport to codify analytic truths, and none does anything whatever to distinguish mere factual information from what philosophers would consider strictly parts of the meanings of the definienda; usually the entries contain just informal explanations of how to apply the concept or recognize the item in question, recognition criteria mixed with a motley of commonsensical facts.  Hence Quine’s apt remark that dictionaries differ in no principled way from encyclopedias.
     One might respond that in real life, lexicographers get sloppy (many of them are free-lancing, and all are underpaid); that is just real life for you.  But Quine has a more powerful objection to the dictionary idea:  Considering the suggestion that ""bachelor’...is defined as "unmarried man’," Quine asks, "Who defined it thus, and when?" ("Two Dogmas," p. 24)  This question is sometimes passed by as a rhetorical flourish or as merely a superficial piece of sarcasm, but it has two deep points.  One is that on pain of regress, conventions of language could not all have been established explicitly by syndics sitting in council.  (There are infinitely many logical truths.   But there could not be a complete infinitely totality of single conventions which fix all the logical truths, one by one, immediately.  Therefore if logical truths are true by convention, they must be given by general conventions, we must rely on an antecedent knowledge of the very same sort of logical truths, in mediating our inferences.   Thus the regress.)  Quine originally took this to embarrass the idea of a linguistic convention quite badly, but he agrees it has been circumvented by Lewis’ (1969) account of tacit convention.<10>
     The second and more telling point is that the lexicographer is in exactly the position of the field linguist in a case of radical translation; the lexicographer merely observes language already in use and theorizes about it.  Thus he or she cannot experimentally distinguish between sentences that are fixed by convention as "analytic" in the target language and those that are merely considered obvious by the native speakers, since all that meets the eye and ear is one class of "stimulus-analytic" strings, the sentences that command universal assent and whose hypothetical denials elicit funny looks.
     Granted, a lexicographer’s evidence base is not exhausted by the linguistic behavior of others.  The lexicographer who is a native speaker may appeal to his or her own linguistic competence and internal sense of word meanings.  But the same criticism applies to that appeal, as it does to the Ordinary Language philosopher’s introspective test of meaning ("Would we still call a thing an X if it weren’t a Y any more?"):  We cannot say even of ourselves, simply by introspecting, whether some belief of ours is true by definition or whether it is merely a very well-entrenched truism.  For we have nothing to go on but the feeling of obviousness and thorough-entrenchedness.  One might say that if the lexicographer knows what genuine synonymy is, he or she is entitled to use that notion in constructing an ideal, nonsloppy dictionary of a language, and can make the appropriate distinctions on sight.  But this notion of "genuine synonymy" is just what is in question, and claimed by Quine to be an unexemplified philosophers’ artifact.  The appeal to existing or even idealized real-world dictionaries cannot discredit Quine’s claim.
     Perhaps ordinary natural language is not the proper locus of analyticity.  Formal languages are better suited to foment analyticities, since they are under the control of logicians and philosophers in a way that real natural languages are not.  And Carnapian "semantical rules" or "meaning postulates" are splendid candidates for analyticity.  They are assimilated to the definitional postulates of an axiomatized theory.<11>  A logical consequence of a definitional postulate, or at least the postulate itself, ought to be true in virtue of meaning alone.
     Quine has a good deal to say about this;<12>  two criticisms emerge.  First, postulates are postulates only relative to a given context of inquiry; they can be changed around at will without this effecting any substantive change in the theory they inhabit.  Notoriously (Hanson’s [1965] example), the second law of classical mechanics, F = m(d²s/dt²), can be taken or employed variously in the organization of theory--as an axiom, as a stipulative definition of "F," as a definition of "m," or as a purely empirical generalization, depending merely on how the containing theory is axiomatized.
     A Positivist might well reply here that postulates can indeed be added and dropped at will for reasons of convenience, and theories re-axiomatized, but that such tampering necessitates corresponding changes in the meanings of the terms which occur crucially in (and are "implicitly defined by") the theories; different axiomatizations give different meanings to the theoretical terms, even though the theories do not vary in overall empirical content.  But Quine rejoins that there is no independent evidence at all of any such meaning changes’ having occurred, when the postulates of actual theories have been readjusted in real life.  Put the matter up to any scientist or any intelligent but philosophically untainted specialist in some other cognitive discipline, and neither informant will know what we are talking about; though they might grant loosely that "what we understand by" a theoretical term has changed when we reaxiomatize or introduce a new definitional postulate, they would never maintain that the term had "just plain changed its meaning," or that it now simply meant something different from what it meant before.  (Hilary Putnam gives numerous examples of relevant scientific changes that plainly have involved no meaning changes in any but a question-begging sense.<13>)  The Positivist insists that the changes occur, because they must occur if the Positivist theory is correct.  But on pain of severe ad hocness, the meaning changes would have to be exhibited in a non-question-begging way.<14>
     Quine’s second criticism of the appeal to definitional postulates is that theories can turn out to be false; so "truth by convention" does not guarantee truth.  (Harman puts the criticism in this way on pp. 130-131.)  As always, someone may reply that a theory’s definitional postulates are special, in that even when their containing theory goes down in flames, their analyticity exempts them from falsification.  But as always, the alleged analytic exemption is just what is at issue.
     One might think that at least the theorems of classical logic would be exempt, before any special vocabulary had been introduced.  And the notion that they hold purely in virtue of the meanings of the logical constants is a very appealing philosophy of logic, whether or not one approves of "truth by definition" in science or in other empirical discourse about the material world.  But the present point holds for logical truths also:  As the Law of Excluded Middle is questioned both (for epistemological reasons) by intuitionists and (for scientific reasons) by quantum logicians, so might we find epistemological or scientific reasons for qualifying even the Law of Noncontradiction.  Not even logic is epistemically unrevisable in the strong sense required for analyticity.  Moreover and contributorily, as we saw in Chapter 10, the alleged distinction of kind within linguistic semantics  between the "logical constants" and other words of English is dissolved by even a little examination.
     The third idea of truth by convention, based on stipulative definition, will require a new chapter.
 
 

1 Bergmann (1953); Grice and Strawson (1956); Bennett (1959); Katz (1967, 1974), and elsewhere; and of course many others.

2 Quine (1968, 1969, 1969, 1970, 1987).

3 Callaway (1980, 1984) has also championed meaning without analyticity.

4 Harman's splendid essay (hereafter just "Harman") is the classic exposition of Quine's views on meaning; my own understanding of analyticity derives very largely from it.

5 See Quine (1960, p. 65, n 3).  I conjecture that all the charges of circularity and emptiness of characterization were really meant to defame the idea that behind philosophers' "intuitions" of analyticity lies a substantive and good theory of meaning that supports the intuitions and has explanatory merit of some sort.  I agree with Quine that that idea is false, even though unlike him I believe there is a substantive and good theory of meaning.

6 Lewis (1969, pp. 174-177)--a book on linguistic convention addressed in part to the analyticity question, in which, ironically, Lewis repudiates the understanding of analyticity as truth by convention.  Though what proposition an English sentence expresses is (obviously) a conventional matter, Lewis insists that the truth at every world of the proposition expressed is not conventional at all  but is a mind- and cognition-independent modal fact about the universe (p. 207).

7 Against this, a referee has argued that since convention establishes what proposition a given sentence expresses, then if, for a particular sentence, convention establishes that that sentence expresses the necessary proposition, then convention a fortiori establishes its truth and so it is after all true by convention.  But, I reply, it is not true (solely) by convention that that proposition expressed by the sentence, the necessary proposition, is the necessary proposition.  So convention alone still does not guarantee the original sentence's truth.

8 N.b., Lewis does not himself claim that there are any analytic sentences.  Also, a response can be hazarded on his behalf.  Suppose (waiving cardinality problems) that there is a set of all possible worlds.  Any set of worlds is a proposition, and so the set of all worlds is a proposition, the Null or necessary one.  But a proposition corresponding to a neatly delineated set of worlds is surely expressible in English, even if there are some sets of worlds that are just too heterogeneous to match single English sentences (which I doubt).  Thus, some English sentences express the Null Proposition.  Which ones?  The standard philosophical examples of trivial verbal truths are the obvious candidates; therefore probably they are analytic.  My objections to this argument are (i) that it does nothing to show what it is about the trivial verbal truths that makes them express the Null Proposition, and (ii) that in the face of Quinean skepticism we have no reason to think that there are multiple possible worlds distinguished from each other only by "logical" or "conceptual" possibilities that are not nomological possibilities.

9 Nerlich (1991) points out that Leibnizian verificationist-cum-imaginability arguments against absolute space simply though tacitly assume geometrical theses that hold only for Euclidean space--e.g., that a given figure has similar figures of any size!

10 The original objection was stated most fully on pp. 85-98 of Quine (1966tbc); the recantation occurred in Quine's Foreword to Lewis (1969).

11 So far as natural languages can themselves be formalized, a Carnapian structure can be imposed; a formal "semantics for" a natural language such as English might contain a box of "meaning postulates" or "nonlogical axioms" intended to exhibit facts of lexical meaning.  We have seen in the previous chapter that Carnap's practice is a vexed one.

12 "Two Dogmas," pp. 35-36; Quine (1963, p. 392); and elsewhere.

13 See Putnam (1975a, pp. 310-315; 1975b; 1975d; 1975e, pp. 254-257), and other papers collected in Putnam (1975c).

14 I think, incidentally, that this rejoinder is what Quine has in mind when he makes a somewhat misleading remark on p. 37 of "Two Dogmas," calling the analytic/synthetic distinction a "metaphysical article of faith."  He means, not that the distinction is a prediction that may or may not come true, but that it is something that is dogmatically believed come what may in the absence of any convincing evidence at all.  Pp. 122-127 of Quine (1963) bear a similar interpretation.
 
 

     Let us turn to our third idea of truth by convention, inspired by the occasional practice of novel stipulative definition within more or less natural languages.

 1.  STIPULATIVE DEFINITION

     Suppose I offer a novel coinage:

                                                Dudent  =df  Dumb student.

And suppose I add, now employing my newly augmented object language, "All dudents are students, you see."  How could I possibly be mistaken?  My assertion seems a clear example of the truly unrevisable.  To question it would be to question either the success of the wisdom of my stipulation, and to reject it would be to reject the stipulation and refuse to join in speaking the augmented language; my audience could not "deny the doctrine" and genuinely disagree with what I had said, for the ostensible denial, "There are dudents who are not students," simply makes no sense in the context.  It is either a straightforward self-contradiction or it means nothing at all.
     Quine himself is startlingly respectful of this idea:

There does...remain still an extreme sort of definition which does not hark back to prior synonymies at all: namely, the explicitly conventional introduction of novel notations for purposes of sheer abbreviation.  Here the definiendum becomes synonymous with the definiens simply because it has been created expressly for the purpose of being synonymous with the definiens.  Here we have a really transparent case of synonymy created by definition; would that all species of synonymy were as intelligible.  For the rest, definition rests on synonymy rather than explaining it.  ("Two Dogmas," pp. 25-26)

 2.  THE OBJECTIONS

     The obvious Quinean objection is based on the indeterminacy doctrine:  Suppose a friend comes by and offers a stipulative definition, say: "‘Veline’ shall mean vegetarian cat."  We could accept a translation manual according to which our friend’s phrase "shall mean" is taken onto the term "obviously denotes" and suitable readjustments of syntax are made.  The idea is that our friend, in her role as unspoiled native, behaves toward the indicative in question, "All velines are vegetarian cats," exactly the same as she behaves toward an admittedly synthetic but trivial and obvious truth--except for the attachment of a particular label: "analytic," "=_df," "shall mean," or what have you.  And our translation of the label is indeterminate; there is no fact of the matter.  And so there is no fact of the matter as to whether our friend’s utterance is really a stipulative definition or merely the reiteration of an obvious truth for the common good.
     On Quine’s view, her utterance is indeterminate in a second way as well:  There is no fact of the matter as to what her putative definiens means.  Even if we were able to bypass the indeterminacy of "=_df" & Co., and were able to say determinately that "veline" was analytically connected to another expression, we would not thereby know a necessary truth unless we knew what the definiens meant, to the exclusion of what it coextended with but did not mean--and this, according to the indeterminacy doctrine, we cannot know, because there is no fact of the matter to be right or wrong about.
     As before, I shall not rely on the indeterminacy doctrine.  But the first of these two appeals to translation has some force independently of the indeterminacy claim.

     Objection 1:  On p. 27 of "Two Dogmas," Quine lists three possibilities as to what is going on when new notation is stipulatively defined in a formal system.  Of the first two (both technical equatings of long technical locutions with shorter terms, but for different purposes) he complains, as usual, that they already presuppose a prior notion of strict, analyticity-generating synonymy and so cannot be appealed to in an argument for the existence of analytic sentences.<1>  In keeping with his concession quoted above, he expressly refrains from bringing the same complaint against the third possibility that "...the definiendum may be a newly created notation, newly endowed with meaning here and now."
     But we may bring it.  The idea of adventitious stipulative definition as it would have to be construed for the purpose of saving analyticity must presuppose the notion of strict, analyticity-generating synonymy.  Otherwise it would collapse back into mere definitional postulation.  Stipulators must have the power to create synonymies and truths by fiat.  The "dudent" example seems to show that people do sometimes have that power.  But notice that the "dudent" argument still presupposes the analyticity of logical truths:  "All dudents are students" is true because "dudent" has "student" as part of its meaning and all students are students (cf. Quine [1966c], pp. 71-72).  As we have seen, Quine has argued against the analyticity (in the strong sense) of the logical truths; so he would hardly give in on that of sentences obtained from logical truths by substitution.  Recall Harman’s point that a sentence’s being "endowed with meaning" on the spot does not prevent it from being false and thus does not solely explain its being true.
     Notice too that if "All dudents are students" is truly analytic, its syntactic denial is meaningless; if a sentence really is true just in virtue of its meaning, then that sentence’s syntactic denial cannot mean what it appears to say, and so either means something else or means nothing.  But "There are dudents who are not students," though (if you like) logically false, is not meaningless.  We all know what it means; if we did not, we would not know that it was logically false and that to token it is to contradict oneself.

     Objection 2: Let us revisit and generalize the first of Quine’s two arguments against analyticity through definitional postulation (cf. Harman, p. 141).  The main problem was that the same theory or belief system can be axiomatized in different ways, treating different elements as "analytic."  More formally:  Suppose

             (1)  A sentence S in theory T1 is "true by definition."

And

             (2)  The very same sentence S appears in theory T2 as a purely empirical generalization.

(Let S, for example, be "F = m(d²s/dt²).")

             (3)  T1 and T2 are merely two different axiomatizations of more or less the same body of truths.

             (4)  A sentence’s analyticity/syntheticity is a function of the meanings of its component terms.   [Tenet of the "analyticity" view].

             (5)   Analytic sentences are those which are "true by definition"; others are synthetic.  [Assumption for reductio]

But

             (6)  The meanings of S’s component terms do not change (in any ordinary sense) from T1 to T2.  [Seemingly supported by (3), but also
                plausible on its own]

             (7)  Either S is analytic in both T1 and T2 or S is synthetic in both T1 and T2.  [(4),(6)]

             (8)  S is analytic in T1 but synthetic in T2.  [(1),(2),(5)]

             (9)  CONTRADICTION!   [(7),(8)]

             (10)  (5) is false.  [Since our premises (1), (2), (3) and (6) are true and we grant the analyticity theorist (4)]

So "definition does not hold the key to analyticity."
     This formulation of the Quine-Harman argument opens the way for the standard reply simply put: that premise (6) begs the question, because on the "analyticity" view it follows trivially from (1) and (2) that the meanings of the component terms of S do change.
     Here as in the previous, more specific case of definitional postulation, I am in sympathy with Quine’s and Harman’s replies on this point:  Who says the meanings change from T1 to T2?  It seems that all we have here is two organizations of the same theory differing only in elegance or in convenience for particular purposes.  It is up to the defender of analyticity to show why that natural reaction is wrong.  But the analyticity theorist cannot do this without relying on the essential tenets of just the philosophical view that is in question.  Thus, it is the analyicity theorist, not Quine and Harman, who is begging the question here if anyone is.
     Just as there is no difference between a "meaning postulate" and an ordinary postulate (in the sense of a very general truth chosen to generate other truths), that has other than ad hoc considerations to recommend it, there is no non-ad-hoc difference between a sentence "true by stipulative definition" and one simply true ex hypothesi.  Even stipulative definitions, like postulates, can rationally be given up in physics without apparent change of meaning.  Although the clearest cases of postulational change without meaning change involve natural-kind terms when new empirical information comes in, there are other, more ordinary occasions for semantically harmless readjustment of taxonomies as well.<2>

     Objection 3:  In stipulative definition, there is an unexpected problem of getting from the stipulator’s token to an indicative sentence held to be analytically true.  The problem takes a bit of explaining.
     One strong temptation to see stipulative definition as yielding (however harmless) analytic truths arises from the fact that, when a person utters (tokens) an explicit stipulative definition, one cannot at all appropriately respond, "That’s false."  (Lucy: "Here’s my first definition: A ‘freebish’ is a dog eating pizza."  Linus: "That’s false, Lucy, because..."  Sound effect: POW!)  Naturally enough, this is felt to show that explicit stipulative definitions and their consequences cannot be false, and thus are necessarily true.
     I shall show here (expanding on pp. 71-72 of Quine [1966c]) why this approach to necessary truth is blocked.  Let us first look at some typical instances of explicit stipulative definition:

             (a)  "Veline"  =df  vegetarian cat.

             (b)  All velines are by definition vegetarian cats.

             (c)  We shall use "veline" to mean "vegetarian cat."

             (d)  Let "veline" abbreviate "vegetarian cat."

             (e)  "Veline" shall denote vegetarian cats.

             (f)  Let "veline" mean vegetarian cat.

Now none of these instances is itself a straightforward indicative sentence alleged to be analytic.  It is the prior occurrence of one of these that it supposed to legitimize the imputation of analyticity to some later, distinct sentence: "All velines are vegetarian cats." So we may well ask how this process takes place, and what sort of relation holds between the stipulative sentences (let us hereafter call them just "stipulatives") and the indicative, "All velines are vegetarian cats."
     Stipulatives (d) and (f) best capture the spirit of convention-fixing, being hortatory subjunctives.  Obviously they cannot be true or false; they state rules, more or less in the form of commands from the Great Convention-Giver to her people.  (c) and (e), I think, are best interpreted as announcements of the speaker’s intention to obey a certain rewriting rule--the idea being that if you want to understand the speaker, you must obey the same rule pro tem.  (c), then, is just a slightly more tactful version of (d).
     It is hard to assess (b)’s logical status; (b) seems to be an indicative, not a subjunctive or a command.  But one could, I suppose, state the stipulative definitions of one’s theory in this form, so long as one used a heading on the page which made it clear that the sentences codified stipulative and not reportive definitions.  (a) could be read aloud as almost any of the other stipulatives listed; it does not seem to have a separate logical status of its own.
     So we have two basic kinds of stipulative:  A hortatory subjunctive, or perhaps an imperative telling us to do such-and-such under such-and-such conditions; and an indicative sentence that either is flagged (by a key phrase such as "by definition" appearing within it) or is prefaced by a heading such as "Abbreviations" or "Defined Terms" or even "Semantical Rules."  In what way do stipulatives of either kind yield necessary truths?
     Let us take the latter kind first. (b) does seem to entail "All velines are vegetarian cats," but what entitles us to conclude that the latter sentence is necessary in the sense of analyticity?  Only the presence of the flag, the phrase "by definition," or perhaps the appearance of (b) under a heading of the kind just mentioned.  And it is the forces of just such phrases and headings that is in question here, the question of whether a speaker can make a sentence true by announcing that it shall be so.  So the stipulative occurrence of (b) does not suffice to distinguish (b) from the other stipulatives or to show that "All velines are vegetarian cats" could not possibly be false.  So let us go on to the more promising possibility of getting analytic truths directly from (rewriting) rules of the language itself.
     How is "All velines are vegetarian cats" obtained from, say, (d), or from a still more explicit rule, "Rewrite ‘vegetarian cat’ as ‘veline’?"  There is no entailment here (even if we could make sense of the notion of entailment independently of all the talk about meaning and necessary truth); since (d) is not an indicative, it has in the ordinary sense no truth-value and hence cannot entail anything.   The only obvious way to get "All velines are vegetarian cats" from our rewriting rule is to regard it as being the result of applying the rule to the logical truth, "All vegetarian cats are vegetarian cats."  But there is still no reason to regard  "All velines are vegetarian cats" as being true by convention, even though it is in part the product of a rewriting rule or conventional abbreviation which is (we may concede) in a clear sense purely a linguistic convention; because it has not been shown that the logical truth originally operated on by the abbreviative rule is true by convention.  We could regard "All velines are industrious" as being the result of applying our rewriting rule to "All vegetarian cats are industrious," but that does not make the former a necessary truth.
     A second way of putting Objection 3 is to point out that any way of getting a truth couched in the object language from an explicit or tacit stipulative must make a significant use/mention move and that that move is sure to be fallacious.  Stipulatives are metalinguistic, besides being hortatory or proclamatory in mood, while (in terms of truth conditions) the corresponding object-language trivialities are directly about the world, however little they illuminate it.  Therefore the object-language truisms do not follow deductively from the stipulatives; and there is no other reason to think that the stipulatives make them analytic.  (There is no objection to calling them "definitional truths," so long as it is understood that "definitional" does not mean "analytic."  Nor need we balk at calling expressly logical truths logical, any more than Quine would object to calling laws of physics nomological truths, or to calling laws of the state of Massachusetts legal truths.  The theorems of a particular system of logic form a well-defined and interesting class.  It is just that they are not true (solely) by definition or by virtue of meaning.)
     Incidentally, we now have a way of blocking the argument from the inappropriateness of "That’s false," mentioned above:  The reason "That’s false," uttered in response to a stipulative definition, is inappropriate or sounds funny or whatever  is not that stipulative definitions cannot-be-false in the sense of analyticity.  They are barred from falsity in the trivial sense of being truth-valueless by virtue of their syntax.  In the same sense they cannot-be-true.

     I close my Quinean case with an objection to "truth by convention" generally:  It is only recently that anyone--Lewis<3>--has made clear sense of the notion of a "convention" in the first place.  Lacking such an analysis, Quine’s positivist and Wittgensteinian opponents had no direct way of testing their intuitive idea that certain special sentences were true "by convention."  But using Lewis’ analysis, we can test it.  We can plug the idea of holding something true or treating something as true into Lewis’ analysis  and see if the result is plausible.  (Lewis’ analysis is surely accurate enough to use in such a test, though questions have been raised about its details.<4>)
     Lewis’ formula is as follows (p.78):

 A regularity R in the behavior of members of a population P when they are agents in a recurring situation S is a convention if and only if it is true that, and it is common knowledge in P that, in almost every instance of S among members of P,

 (1)  almost everyone conforms to R;

 (2)  almost everyone expects almost everyone else to conform to R;

 (3)  almost everyone has approximately the same preferences regarding all possible combinations of actions;

 (4)  almost everyone prefers that any one more conform to R, on condition that almost everyone conform to R ["any one more" is a weakening of the stricter "everyone," still in the spirit of "the more the better"];

 (5)  almost everyone would prefer that any one more conform to R’, on condition that almost everyone conform to R’, where R’ is some possible regularity in the behavior of members of P in S, such that almost no one in almost any instance of S among members of P could conform both to R’ and to R.

 3.  THE ANN ARBOR DEFENSE

     Paul Boghossian and William Taschek began building the case anew, in the discussion period following presentation of the previous sections of this chapter at the University of Michigan in 1990.<5>  Boghossian (1992) gives its present formulation:  Suppose, as I insist, that meaning is real and a matter of truth-condition (give and take a bit).  A paradigm case of truth-conditional specification of meaning is the stipulative definition of the truth-functional connectives.  Let us stipulate in particular that  A tribar B  is true iff A and B share a truth-value, and that this exhausts the meaning of the tribar.  [Netscape Composer does not afford me the tribar symbol; that's why I have here substituted the word.]  Then any univocal instance of "P tribar P" is true in virtue of meaning.
     If it be protested that no pure instance of "P tribar P" is known to occur in English or any other natural language,<6>  Boghossian points out (p. 21) that anyone might augment English with a biconditional connective (say, "biff") whose meaning is specified as follows:  Sentences of the form "P biff Q" are true just in case "P" and "Q" express the same meaning.  Then every instance of  "P biff P" in Augmented English will be true solely in virtue of its meaning.  He concludes that if (as I urge) we are meaning realists at all, then either there are analytic sentences or, because we could so easily manufacture some, there might as well be.
     To my disgust, I find Boghossian’s argument fairly convincing so far as it goes.  Thus I am prepared to give up the thesis that there could not in any language be sentences that are analytic in anything like the sense Quine attacks.  Of course, that thesis is a very strong one, certainly stronger than Quine’s own (given his startling concession on pp. 25-26 of "Two Dogmas").
     (Why have I flirted with the Very Strong Thesis at all, then, in the first place?  Because any very robust metaphysical realist ought to blanch at the idea that any sentence could be true, i.e., correspond to reality, by convention, i.e., in virtue of human agreement and nothing more.  But at least the metaphysical realist can still have the logical-atomist claim that atomic sentences are made true only by substantive correspondence with mind-independent reality; and, as we shall see, the admission of sentences that are "analytic" in the present sense is not particularly antiQuinean--nor does it after all mark such sentences in any clear way as true "by convention.")
     Actually, Boghossian’s point generalizes,<7> and damages more than just the Very Strong Thesis.  Supposing again that meaning is real and codified (give and take a bit) by systematic specifications of truth-condition, it is reasonable to expect that any natural language will contain a significant class of sentences that, in one clear sense, are "true in virtue of meaning."  This is because the semantics of any natural language L contains a class, however ill-defined,<8> of operators that act as logical connectives and whose truth-conditional role is specified as part of L’s Tarskian truth-definition.  (For the truth-functional connectives, at least, their truth-conditional role is all that the truth-definition has to say about them.)  Those operators may directly formalize morphemes of L, or they may have no surface reflections but figure only in the semantic underpinning.
     In either case they will allow a truth-definition to mark a special class of sentences as logical truths; a "logical truth" will be just a sentence that remains true under any uniform reinterpretation of its nonlogical terms.  And this embarrasses my Quinean case, by affording a clear sense in which some sentences of L are indeed "true in virtue of meaning":  (i) Given any logical truth T of L, T’s truth will be deducible from L’s truth-definition alone.  But (ii) a theory of meaning for L contains a truth-definition for L.<9>  It seems reasonable to suppose that (iii) a correct theory of meaning for L, strictly so called, codifies all and only the meanings of L’s sentences.  Therefore (iv) the truth of T follows from L’s meaning-facts alone.  And since (v) the derivation of T’s truth will require only those clauses of L’s truth-definition that focus on T’s component morphemes in particular, (vi) so there:  T’s truth is guaranteed by T’s own meaning, not just holistically by the semantics of L in its entirety.  Call this the "Truth-Definition Argument."
     Assuming L’s semantics also features some meaning postulates (however arbitrarily those may be distinguished from clauses in L’s truth-definition), an even larger class of L’s sentences will come out true in virtue of meaning--it will include not just L’s logical truths but further consequences of L’s truth-definition taken together with L’s meaning postulates, such as (in English) the classic "No bachelor is married."  This begins to look very bad for the meaning-realist foe of analyticity.
     The Truth-Definition Argument’s move from (v) to (vi) would require some assumptions, at least one of which would be highly controversial.  But I am more concerned to belittle the weaker (iv), and then to dispute the spirit of (iii).
     I concede that any truth-conditional semantics for a natural language will yield a subclass of that language’s sentences whose truth is guaranteed in the way we have seen by the semantics alone.  In that sense, some sentences are "true in virtue of meaning."  But finer distinctions are needed here.
     In the first place, what has strictly been shown to be entailed by L’s truth-definition is only that T is true and so is every other sentence having the same form--nothing about why T is true, about T’s modal status, or about T’s epistemic credentials.  Still, the Truth-Definition Argument itself shows (if it is sound) that a complete theory of T’s meaning alone entails T’s truth, and that ought to make T true in virtue of meaning, and truth-in-virtue-of-meaning is what I myself have insisted we should mean by "analytic."  Even though the truth-definition entails nothing about why T is true, its entailing T constitutes why T is true.
     That last point is disputable.  As Devitt (1992) observes, that a set of sentences entails a further sentence S does not itself entail that the facts expressed by the set of sentences are what make S true.  Only an unreconstructed fan of the Deductive-Nomological theory of explanation would automatically take deducibility as explanation.  For all that has been shown, though a truth-definition specifies and codifies meanings and even specifies some truths, it does not explain those truths.
     I am not completely convinced by that post-Positivist line of defense, though it is correct so far as it goes; whatever the failings of the Deductive-Nomological theory of scientific explanation, the Truth-Definition Argument still seems to have shown that there are sentences whose truth is guaranteed by their truth-conditional meanings.  But I do not need the post-Positivist line.  I can concede the Deductive-Nomological sense in which some sentences are true in virtue of meaning, those sentences whose truth follows from their containing languages’ truth-definitions (call such sentences "TD-analytic").  For the truth-in-virtue-of-meaning thus demonstrated is philosophically innocuous in each of two opposing ways.  First, it does not entail alethic necessity of any grade.  For all that has been shown, a sentence might be TD-analytic and still metaphysically contingent.  A cheap way of seeing this is to note that a theory of meaning for L (if it included a "semantic pragmatics" in Cresswell’s [1973] sense as well as a pure truth-definition) might entail that "I exist" is always true.  A more expensive way is to recall the possibility that the Law of Excluded Middle is contingent; even if our own world is not one in which the Law fails, there may be worlds featuring suitably different quantum mechanics.
     Second, TD-analyticity does not abet the traditional Positivist, antimetaphysical project of debunking necessity by trivializing it.  The Positivists wanted to show that analytic sentences are contentless, fail to correspond to facts, say nothing "about the world," "merely reflect our habits of usage," etc.  (This was the key to their elegant solution of what we now know as Benacerraf’s [1973] problem about arithmetical reference.  They held arithmetical sentences to be analytic, and therefore that questions of arithmetical reference, arithmetical facts and the like could not arise.)  But, so far as has been shown, the logical truths of L (such as "Boghossian and Devitt are philosophers just in case Boghossian and Devitt are philosophers") are still about real things and people, and say perfectly intelligible if boring things about them.
     The latter point can be strengthened.  When theorists have called a sentence "analytic," meaning "true in virtue of meaning," they have traditionally also meant that "the world" makes no contribution to the sentence’s truth, that the sentence holds no matter what the nonlinguistic world is like.  But the Truth-Definition Argument does not support analyticity in that sense.  Although the truth of sentence T is entailed by our theory of meaning for L, that theory of meaning itself presupposes worldly tautologous facts; and therefore its entailing T all by itself fails to show that the world plays no role in T’s truth.  If (the real people) Boghossian and Devitt were not: philosophers just in case Boghossian and Devitt are philosophers, the tautologous sentence expressing that fact would not be true after all.
     To see the dependence more clearly, consider the proof of T from L’s truth-definition.  Let us take as given the theorem

             (BT)  "Boghossian and Devitt are philosophers" is true iff Boghossian and Devitt are philosophers.

and the recursive clause

             (Biff)  /-A biff B-\ is true iff either A and B are both true or A and B are both false.

To derive the out-and-out truth of "Boghossian and Devitt are philosophers biff Boghossian and Devitt are philosophers," we need to detach somehow, both from the supposition that Boghossian and Devitt are (in fact) philosophers and from the supposition that they are not philosophers.  The proof would have to go by vel-Elimination or Dilemma, based on the tautology that Boghossian and Devitt either are or are not philosophers.  That tautology would have to be assumed as a premise licensed by our old friend the Law of Excluded Middle.  Thus, although we can indeed deduce our logical truth T from L’s theory of meaning "alone," the deduction relies on at least one tautology already held to be true.  Any use or application of a theory of meaning will in that sense presuppose some laws of logic, and so the world will after all contribute to the derivations of logical truths.
     (Someone may point out that in any decent natural-deduction system, Excluded Middle can itself be proved from the null set of premises; so we need not invoke it as an article of faith, in our subsequent derivations from truth-definitions.  But this only puts the problem off.  The rules of inference employed in the system are validated only by soundness proofs done by truth table, and the reasoning that comprises the soundness proofs itself presupposes Excluded Middle.  Perceptive students sometimes complain about this.<11>)
     Thus, the Truth-Definition Argument’s premise (iii) is true only if we do the Argument the courtesy of allowing that the class of "meaning-facts" is closed under entailment and so including the worldly tautologous facts.  But the Quinean maintains, to the contrary, that the tautologous facts predate linguistic meaning; so the Quinean has no reason to grant (iii) strictly construed, and will deny (iii)’s "and only" clause.  (iv) remains true, but only because of the worldly tautologous facts presupposed by the notion of "following from" that occurs in (iv).
     Finally, note that although the truth of a tautologous sentence is explained by the sentence’s meaning in the sense I have conceded, the fact recorded by the sentence is a different explanandum and is not explained by any truth-definition.  No truth-definition explains why the live people Boghossian and Devitt are philosophers just in case they are philosophers; au contraire, the truth-definitional explanations of the truth-values of tautologous sentences already presuppose tautologous facts of that sort.
     I have reluctantly conceded a sense in which logical truths are "true in virtue of meaning."  But that is not a sense in which they are true contentlessly and with no contribution from the nonlinguistic world.  Nor is it any help either to the metaphysician in search of immutable necessary truths or to the anti-metaphysician bent on debunking such things.

 4.  AND SO WHAT?

     Moreover, our Ann Arbor concession leaves the foe of analyticity with each of three further fallback positions, all of which I believe are true, interesting, and supported by the barrage of Quinean arguments I have given in the last chapter and this one:
     (i)  To grant semantic privilege on the basis of (assumed) meaning realism is not to grant any epistemological privilege to any sentence of any natural language.  For one thing, we may never know whether any existing sentence does reflect the tribar.  For another, even if certain laws of logic are in fact true and necessarily so, we may not be completely justified in believing them.  All the epistemological arguments against unrevisability hold.  What Boghossian’s point shows is rather that TD-analyticity does not after all require epistemic unrevisability.
     Qualification:  The point does still require that in some sense, someone who rejects a de facto TD-analytic sentence has failed to grasp that sentence’s meaning.  But we are or should be hard put to say how the difference between uncertainty about logical form and uncertainty about fact should be distributed over epistemology, given that there are Duhemian tradeoffs between the two.  I can deny a TD-analytic sentence, because I may have reasonable (though in fact misleading) grounds for doubting that the sentence expresses the logically-true proposition it does express.  I could even doubt the logically-true proposition, for either of two reasons:  First, because, having imperfect access to what proposition it is I am entertaining, I might entertain it without realizing that it is a logical truth.  (In this regard, logical truths are no better off than the Kripkean metaphysical necessities discovered by chemistry, or for that matter the necessary but a posteriori identities of persons such as Cicero and Tully.)  The second reason is that even if I do know very well what proposition it is I am entertaining, say the Law of Excluded Middle, I may nonetheless have rational epistemic grounds for doubting it; sometimes genuine metaphysical necessities do not carry epistemic credentials to match.<12>
     Thus, the new meaning-realist TD-analytic/synthetic distinction has none of the epistemological importance that has been pinned on it by (e.g., Positivist) philosophers.
     (ii)  Nor, obviously, has TD-analyticity much philosophical importance.  In particular, one cannot defend a philosophical premise or other claim by announcing that the claim is analytic and brooking no further dispute.  (The same goes for attacking a position one dislikes by calling it "unintelligible," i.e., analytically false, or by asserting that it "makes no sense" or the like.)  This is akin to Putnam’s (1975d) point that any analytic sentences there may be would be trivial and of no philosophical interest.  Even if stipulative definition does yield some analytic truths, they certainly are not the ones that are useful to philosophers, being absurdly trivial.  No sentence that has ever been put forward as a necessary truth by a philosopher seriously philosophizing is the product of a stipulative definition.  Even when philosophers construct elaborate systems of stipulative definitions, our interest is not in the definitions themselves, which can have no nontrivial consequences, but rather in the way in which the system connects up to the real world, and the latter cannot be stipulative.
     (iii)  Not even "It is raining (tribar) it is raining" is true by convention, if this is taken strictly as meaning conventionally held true (cf. my appeal to Lewis’ analysis of convention.)  At best its truth is a consequence of the meaning of the tribar conjoined with some linguistic assumptions.  Perhaps the assumptions are all individually conventional in some sense; but the notion of "consequence" that figures here again presupposes tautologous facts of the sort that are cited in soundness proofs.  Despite my concession of TD-analyticity, we robust metaphysical realists at least are spared the ignominy of truth by convention.
 

1 I think he would also want to add exactly what he later says about "explication" generally in sections 53 and 54 of Quine (1960): that theoretical "definition" is a replacement or substitution, not an analysis or uncovering of a pre-existing conceptual meaning.

2 See also Harman, pp. 140-141; Quine's remarks (1963, p. 113) on the rapidity with which stipulative definitions "fade away"; and again Putnam's (1975x) many examples.

3 An earlier stab was made by Schwayder (1965).

4 E.g., Jamieson (1975); Burge (1975); Gilbert (1981).

5 Actually, I had formulated an argument similar to theirs in a 1975 course handout, but then self-servingly forgotten it for fifteen years.
    It should be noted that neither party remains in Ann Arbor; Boghossian is now at New York University, Taschek at Ohio State.

6 One might say this because one believes natural-language conditionals are not truth-functional.  I do not myself believe any natural language contains either the horseshoe, any simple strict conditional, or any biconditional made of any of those.  (For the arguments, see Chapter III of Lycan [forthcoming].)  The semantics of ostensibly biconditional sentences of English is both tricky and infested with the pragmatics of restriction classes and parameter shift.  Thus it is not clear that every biconditional, or even every one with ostensibly identical LHS and RHS, is true, much less true by virtue of meaning.  (I know that sounds awful.  I also deny the validity of Modus Ponens; see Lycan [1992; forthcoming, Ch. III].)
    If one tries to make an analytic formula out of a nonconditional connective, the obvious candidate is "or."  But (surprise!) I do not think natural languages contain the vel, or any very well-behaved disjunction operator, either.

7 Boghossian has notified me in correspondence that he himself does not accept this generalization.  Nor is it his intention to defend "truth by convention or to rehabilitate the Positivist theory of necessity and logical truth.  So the anti-Positivist points I shall make below are not aimed at Boghossian's own view.

8 The argument of Chapter 10 does not intrude here, for the point I am about to make does not depend on there being any single best precisification of the notion of a logical constant.

9 I continue to make this Davidsonian assumption, defended in Lycan (1984).  L's theory of meaning would also contain whatever is needed to block Chapter 9's counterexamples to the Modified Intension Thesis, plus whatever anyone thinks should be counted as part of the "meaning" of a sentence over and above its truth-condition.

10 Yes, I know that sentence is a counterlogical.  Unlike David Lewis, I have no problem with counterlogicals.

11 If it be complained further--say by an inferential semanticist in the tradition of Wittgenstein, Carnap and Sellars--that rules of inference stand on their own as intuitively sound and need no extraneous license from model theory, the Quinean responds with the rhetorical question of what distinguishes the "intuitive" validity of a rule from a conditional belief about the world.

12 A similar point is made by Devitt (1992).