The Seas of Language.
The Dummett of popular imagination is a confident anti-realist: the dogmatic purveyor of a sophisticated and comprehensive brand of idealism. As these essays make plain, however, the real Dummett is very different. He's not an anti-realist at all; he's just a guy with an argument - the celebrated Dummettian Case for Anti-Realism - according to which the various "realisms" we tend to treat as Moorean data are in fact philosophically untenable. Unlike his imaginary counterpart, however, the real Dummett does not assert this conclusion. He concedes that The Case is unlikely to convince; it is too long and complicated; its premises too contentious; its conclusion so firmly at odds with "common sense". It is put forward instead as a challenge: either accept the conclusion and embark on the project of delineating a coherent alternative to realism (a project Dummett regards as hardly begun outside mathematics), or else say where the argument derails. Dummett is clearly sympathetic to The Case; but he is concerned to stress that at present we are not entitled to a final verdict. It may yet turn out that while there is no fault to be found with the critique of realism that The Case embodies, it is nonetheless impossible to describe an intelligible alternative; in which case we would have not a proof of anti-realism but a paradox, and hence the occasion for a re-examination of our most basic assumptions.
We may think of The Case as proceeding in stages:
(1) The problem of realism is initially given to us in palpably inadequate metaphysical terms. Sometimes the question is framed in the jargon of post-Kantian idealism, as when we are instructed to ask whether reality as a whole or some interesting part of it exists "independently of the mind" or perhaps "independently of our linguistic and social practices". In other cases the jargon recalls a much older tradition according to which reality is somehow a matter of degree. (Of course, numbers exist in a sense; that's just mathematics. But are they really, robustly real? Are they really out there, part of the furniture of the world?) And the trouble is that any issue framed in these terms is simply too obscure to discuss.
If the issue is to be tractable we require a "criterion" of realism: some clear, debatable thesis upon which the acceptability of this "rhetoric of reality" may be seen to turn. The Case thus begins with the famous suggestion that many (though not all) disputes over realism can be recast as disputes about whether the language we use to describe the disputed region of fact respects the principle of bivalence. With some qualifications to be mentioned below, Dummett holds that to be a realist is to accept bivalence, to reject it to be some sort of anti-realist (p. 230).
(2) If bivalence is the issue then the issue is clear enough. Yet for all its clarity, it is far from obvious how to pursue it. Here Dummett's proposal is to see a thesis in philosophical logic as turning upon a more basic thesis in the philosophy of language. Bivalence holds for a region of discourse only if the meanings of its sentences are sufficient to determine a truth-value for each, even when it is impossible for us to discover what that truth-value is. However, a sentence can possess such a meaning only if it is sometimes appropriate to regard a competent speaker's understanding as consisting in his grasp of such "evidence-transcendent" truth-conditions. For Dummett, a theory of meaning must be a theory of understanding (p. 4): a representation in propositional form of what a competent speaker knows when he knows his language. The question of bivalence (and hence the question of realism) is thus reduced at this stage to whether linguistic understanding can be said to consist in a grasp of truth-conditions of this epistemically unconstrained sort.
(3) In the third part of the Case - the destructive part - general considerations are brought to bear against the possibility of any such conception of meaning. Meaning is use; and use is the deployment of linguistic objects in public circumstances. So whatever it is to understand a sentence as possessing certain truth-conditions, this state must be "exhaustively manifestable": it must be the sort of thing that can (and frequently does) show up in one's overt behavior, otherwise languages would be unlearnable and communication impossible. When S's truth-conditions can be recognized as obtaining when they do obtain, there is no problem: I can manifest my grasp of S's meaning by asserting it when it is true and rejecting it when it isn't. But when S's truth-conditions are undetectable, I cannot manifest my knowledge of them in this way. And since there is no other way for the fact that I assign one set of truth-conditions to S rather than another to show up, it follows that there is no way for a speaker to manifest his grasp of a sentence as having any determinate verification-transcendent truth-condition. This stage thus concludes: since there is no way for a speaker to manifest his assignment of this or that verification-transcendent truth-condition to a sentence, no sentence can possess such a truth-condition; hence we are never entitled in the interesting cases to accept the principle of bivalence, and realism must therefore be abandoned.
(4) Or at least, that is what we should say provided we can give an alternative theory of meaning for the language in question. We have one promising model. An intuitionistic semantics for the language of mathematics represents a speaker's understanding of a mathematical sentence as consisting in his knowledge of what counts as a proof of it: a condition he can manifest by accepting and rejecting purported proofs. The statements of such a language do have truth-conditions; but the notions of truth and falsity play no central role in an explanation of their meanings, this function having been shifted to the notions of proof and refutation. Intuitionism so understood is to be the paradigm for anti-realism in other areas. The problem is that the relatively sharp notion of intuitively acceptable proof in the mathematical case corresponds to the much messier notion of empirical warrant in the general case. The central challenge for the anti-realist who is not content simply with the destructive phase of The Case is therefore to provide a corresponding "justificationist" theory of meaning for each of the interesting empirical cases; and here the problems are considerable. For example, it cannot be said that to understand a sentence about, say, the past is to know what observations in the present would conclusively verify it, since conclusive justification is impossible in such cases. Nor can it be said that to understand a sentence is to know everything that would count as prima facie grounds for asserting it. For whether P counts as evidence for Q is nearly always an empirical matter - a matter of scientific expertise, broadly construed - and not something one knows simply by being a competent speaker of the language. The question, then, is whether there exists a coherent generalization of the intuitionistic idea to the empirical case. This is clearly the part of the argument that Dummett regards as most provisional.
Much has been said about The Case: so much, in fact, that it comes as something of a disappointment to discover how little Dummett has had to say about the enormous critical literature that has grown up around his work on realism. In other areas Dummett has been keen to take on his critics; and much to our benefit. Dummett is a particularly engaging polemicist, and his responses to even the most misguided objections have always been among the best ways to find out what he thinks. The present volume contains references to only a handful of contemporary philosophers, and there is no sustained discussion of efforts by sympathetic dissenters like Crispin Wright and John McDowell or by more impatient critics like Michael Devitt or Simon Blackburn to come to grips with The Case. The main lines of criticism all concern points stressed above. The central premises in the philosophy of language - that a speaker's understanding of a sentence must be fully "manifestable" in his behaviour; that what matters is manifestability "in principle" and not "in practice"; that a sentence cannot possess a certain truth-condition unless a speaker's use of it can be said to be guided by a grasp of that condition, or more generally, the view that the psychological state that underlies a speaker's competent use of a sentence determines its semantic value - these have all been the subject of extensive critical discussion, especially in America where they are widely rejected. A collection of this sort would have been a natural place to join the debate, whether in the preface or in postscripts to selected papers. One regrets that the opportunity has been missed.
Just to illustrate the well-known difficulties that Dummett's view seems to face and the manner in which they are addressed in the present volume, let me briefly take up the crucial first step, the suggestion that obscure but exciting metaphysical disputes about realism can be rendered tractable by treating them as disputes about classical two-valued semantics. We have become so accustomed under Dummett's influence to speaking of a "realist semantics" or "realist truth-conditions" that we have a tendency to forget that realism does not begin life as a semantical thesis, but rather as a metaphysical picture, and that it is therefore a substantive assumption that a semantical thesis like bivalence can go proxy for it. It needs to be shown that verdicts about the semantical thesis correlate sufficiently well with our verdicts about the realist's rhetoric of "mind-independence" and "robust reality" to allow a dispute about one to go proxy for a discussion of the other.
Dummett's considered view is that bivalence is necessary though not sufficient for realism. So we must ask: is it really the case that anyone who holds that statements in a certain class sometimes lack a truth-value is thereby precluded from thinking of their subject matter as altogether "mind-independent" or "robustly real"?
In one sort of case there can be no objection to this. A philosopher who maintains that ethical statements are expressions of feeling, or that conditionals express personal probabilities or that mathematical statements are mere marks on paper must hold the bivalence fails in spades for statements of these sorts, since on his view no sentence in the area possesses a truth-value; and of course no one would call such a theorist a realist. Non-factualism is precisely the denial that the statements in question have a subject matter. They are not representations or descriptions of a world. A fortiori, the world they describe is neither independent of the mind nor robustly real.
But of course for this very reason it would be just as wrong to say that they describe a world that is somehow mind-dependent or less than fully real. This sort of anti-realism provides no elucidation of these obscure but intriguing ideas, and so does not serve to clarify any deep and difficult metaphysical issue.
Dummett has never been interested in the non-factualist paradigm (see 466-7). His attention has always been focused on regions of discourse whose statements are clearly "in the market" for truth and falsity but which nonetheless feature local truth-value gaps. When the sentences in an area are truth-apt in this way, it makes sense to speak of them as possessing a subject matter. The question is then whether the presence of local gaps in a region of discourse provides grounds for withholding the realist's rhetoric of reality from the region of fact it purports to describe.
The obvious worry here is that truth-value gaps have been said to arise for a wide variety of reasons, most of which seem quite remote from any deep metaphysical view. A sentence has been said to lack a truth-value if it contains an empty name; if it has a false presupposition; if it contains a vague predicate applied to a borderline case; or if it contains the truth- predicate applied to a paradoxical sentence. Does a philosopher who accepts one of these doctrines thereby commit himself to rejecting realism?
Dummett's discussion of the first case is illuminating. The problem is pressing, since Frege was the first to hold that a sentence containing an empty name lacks a truth-value and Dummett's criterion therefore seems to imply that Frege of all people was something less than a realist. Can this possibly be Dummett's view? The short answer is "yes" (p. 468). To be sure, it would be seriously misleading to call Frege an anti-realist tout court. Still, in one respect he was less of a realist than he might have been. In holding that a sentence containing a fictive name lacks a truth value, Frege rejected the Meinongian view that such sentences have a domain of non-existent objects as their subject-matter. Dummett can therefore say that to the extent that Frege admitted failures of bivalence for fictive sentences he was something less than a full-fledged realist. He was an anti-realist about non-existent objects.
It is to be observed, however, that this sort of anti-realism once again offers no clarification of the obscure idioms of mind-dependence or attenuated reality. The Fregean holds that some meaningful sentences lack a subject matter; in this respect he is like the non-factualist. But for this very reason his rejection of realism does not involve the view that some genuine subject matter is less than thoroughly there or somehow dependent on the mind.
Dummett also discusses the second case. It is a familiar thought that the philosopher's focus on bivalent formal languages involves an idealization. Natural language is shot through with vagueness, and where this is present bivalence will fail. There is of course no fully adequate semantic model for all of the linguistic phenomena associated with vagueness. Dummett himself was among the first to point out how potent an obstacle the sorites paradox presents to the project of formal semantics (in "Wang's paradox" in Truth and Other Enigmas). Whatever the right account, however, it is simply preposterous to suppose that a statement like "Jones is bald," said of a man with a dozen scattered hairs, must be determinately true or false. So any account of vagueness will have to recognize it as a source of truth-value gaps. Does the theorist who recognizes these gaps thereby become something less than a realist in any intuitive sense?
It is hard to see why he should. On the face of it one does not flirt with idealism simply by admitting gaps due to vagueness. One may accept vagueness and still hold that the facts (such as they are) about baldness and the rest obtain independently of us, our minds and our linguistic practices. So why isn't the fact of vagueness a decisive counterexample to Dummett's proposed criterion?
Dummett's remarks on this issue have always been elusive. I quote the latest discussion in full:
A formulation of bivalence must allow for vagueness. The thesis that every statement is determinately either true or false, even if it is vague, can be sustained only on the implausible supposition that our use of vague expressions confers upon them meanings which determine precise applications for them that we ourselves do not know. A realist must therefore hold that, for every vague statement, there is a range of statements giving more precise information of which a determinate one is true and the rest false. An anti-realist may deny this, holding that reality itself may be vague, whereas for the realist, vagueness inheres only in our forms of description. (p. 468, my emphasis)
This seems at first to be a sensible concession: a realist may recognize failures of bivalence due to vagueness, provided that he thinks of vagueness as a matter of "semantic indeterminacy" and not of there being "vague objects or properties in the world". However, the penultimate sentence seems to saddle the realist with a bizarre commitment to there being, for each range of vague predicates, an underlying range of precise terms which can be used to convey non-vague information about the same subject matter. In the case of baldness this may seem unproblematic. If "hair" were not vague itself (which it is), then the underlying range could consist in claims about the precise number and distribution of a person's hairs. But consider colour. Suppose Bloggs maintains, as is immensely plausible, that our colour vocabulary is vague "all the way down". General colour terms like "red" and "blue" admit of any number of sharpenings: "scarlet", "cerulean", etc.; but at no point in our language does the vagueness disappear. Does this make Bloggs something less than a realist about colours? It is hard to see why it should. It certainly does not imply that he views the distribution of colours as somehow constituted by us or our linguistic practice. Would extending his colour language to the point where he was like the ideal interior decorator in possessing a non-vague name for each determinate shade somehow change his metaphysical commitments in this regard? Again, no. And what are we to make of the closing suggestion that unlike the anti-realist, the realist cannot tolerate "vagueness in reality"? It is, of course, totally unclear what it means to say that reality itself is vague. But to the extent that it is intelligible it would seem to involve no clear connection with the rejection of realism. If anything, the opposite holds: the theorist who rejects vague objects is an anti-realist about vague objects in the same sense in which Frege was an anti-realist about fictional objects.
Dummett has had less to say about the other alleged sources of truth-value gaps. But on the face of it, a philosopher who holds that a sentence like "Fred has stopped chewing tobacco" is neither true nor false if it turns out that Fred never did chew tobacco has taken no stand whatsoever on any metaphysical issue. Or consider the more philosophically resonant case: Kripke has proposed a family of formal theories of truth for languages that contain their own truth predicates. Most of these have the consequence that the truth-predicate is partial: its extension and its anti-extension do not exhaust the universe. (Paradoxical sentences fall between the cracks.) Does this approach to the semantic paradoxes commit Kripke to a metaphysical position that deserves to be called a rejection of realism about truth? Again, it is hard to see why it should.
The connection between the semantical thesis of bivalence and the metaphysical thesis of realism appears to be much looser than Dummett would have us believe. Perhaps the various reasons for rejecting bivalence canvassed above are not good reasons. But the question at this stage is "What is it to be a realist?", and if philosophers who reject bivalence are not for that reason anti-realists, then Dummett's proposal for a necessary condition misfires. One wonders why he does not take Crispin Wright's long-standing suggestion to set the issue of bivalence to one side and to identify anti-realism directly with the thesis that truth is epistemically constrained. An anti-realist of this sort might well reject bivalence; but that would not be what made him an anti-realist. The rejection of bivalence would rather be a symptom of anti-realism, one which may not figure in every case and whose presence would not be not an infallible sign of the underlying disease.
I noted earlier that Dummett regards a commitment to bivalence as necessary but not sufficient for realism, and I would like to conclude with some questions about the examples that motivate the claim of insufficiency. One paradigm for anti-realism is reductionism, the view that the meaning of a statement S in the disputed class is to be explained or analysed in some less-than-obvious fashion by a statement [S.sup.*] in another vocabulary. Phenomenalism and logical behaviourism are the paradigms here; and one often has the sense that views of this sort amount to something less than a full-blooded realism about their subject matter. However, as Dummett is well aware, there is no obvious reason why such a reductionist scheme should not respect the principle of bivalence. If statements in the reducing class are themselves bivalent and if the reduction assigns each S to a single [S.sup.*], then the language as a whole will be bivalent. How then are we to capture the sense in which these views are nonetheless species of anti-realism?
Dummett has two things to say on this matter. He first points out that while reductionists have traditionally had little to say about bivalence, it is nonetheless the case that their views tend to undermine any grounds we might have for maintaining that bivalence must hold in the disputed area. In the classical cases, statements in the disputed area are reduced to statements involving counterfactuals about observable matters of fact. But counterfactuals are notoriously gappy, so in any such scheme we have excellent reason to believe that the reducing language will manifest gaps that will transmit to the original idiom via the translation scheme. (Of course, if one maintains with David Lewis that the indeterminacy that affects counterfactuals is a species of linguistic vagueness, and if one has already said that truth-value gaps which derive from linguistic vagueness do not count against a claim to realism, then this will be no explanation at all for the sense in which phenomenalism and behaviourism constitute species of anti-realism.)
Still, it is conceivable that the reductionist might devise a translation scheme that secured bivalence for the original idiom, and our sense that he is something less than a realist would not then diminish. This brings us to Dummett's second ground for thinking that surprising reductionisms are versions of anti-realism. Consider the view, briefly entertained by Frege, that sentences involving abstract singular terms like "the direction of line a" are to be explained by a system of contextual definitions, e.g.,
"the direction of line a = the direction of line b" is to mean the same as "line a is parallel to line b".
Dummett has long maintained that to the extent that such contextual explanations for abstract terms are available, it is perfectly correct to say that abstract objects like directions and numbers (really) exist and that these singular terms refer to them. Still, for some time now Dummett has also maintained that the Fregean platonist is something less than a full-blooded realist about the abstract (p. 240). This can't be put down to a rejection of bivalence, since the Fregean translations can be arranged to respect this principle. So the criterion of realism must have another aspect that explains its failure in this case.
Dummett suggests that in such cases,
realism is abandoned, not because a truth-conditional account of the meanings of the statements is impossible, nor, necessarily, because there is any reason to repudiate the principle of bivalence as applied to them, but because the notion of reference no longer plays any role in the account of their meanings. Even if we continue to ascribe reference to terms for directions, we do not need to invoke the notion of reference, as applied to such terms, in order to explain how a sentence containing such terms is determined as true or false; the determination of the truth-value of the sentence does not proceed via the identification of an object as the referent of the term. (p. 240, cf. p. 468)
When a region of discourse supports a realist construal, the semantical theory that represents how the content of a sentence S containing a singular term t is determined by the contents of its parts will contain a special clause of the form "t refers to o", knowledge of which will consist in a recognitional capacity on the part of the competent speaker. In the case of abstract singular terms on the Fregean model, by contrast, the account of how the content of a sentence containing "the direction of a" is determined will contain no such clause, but will proceed rather through the translation scheme to a sentence involving no term for directions at all. Dummett's view seems to be that even when the terms in a certain class refer to certain objects, we are not entitled to view those objects "realistically" unless the fact that these terms refer to these things plays an essential role in explaining their semantic contribution to the sentences in which they occur.
One question that immediately arises is how this proposal classifies a view like Russell's in The Problems of Philosophy. Russell's view in that work was that terms for material objects are disguised definite descriptions in a language where the only genuine singular terms name sense data. The scheme here is not phenomenalist: rather, a material object is given to us as "the cause of such-and-such sense data", that cause being conceived of as something quite independent of the mind. Definite descriptions are of course incomplete symbols; so the explanation of the truth-conditions for sentences in which they occur involves no assignment of reference to them. Rather the sentence is first to be translated into a general statement in the more basic language in which no such description appears. Dummett's proposed account of why the Fregean platonist is less than a realist seems to imply that the Russell of 1912 was something less than a realist about the physical world. But that seems wrong. (Of course he was something less than a realist in the old epistemological sense of the term, where realism was a thesis about the immediate objects of perception. But that sense of "realism" is not Dummett's concern.) On the face of it, Russell's "causal descriptivism" together with the theory of descriptions is a paradigm of metaphysical realism. The fact that the view offers a "non-face-value" construal of the language and that attributions of reference to names for material objects plays no fundamental role in the semantics seems not to count against this at all.
Striking for its absence in Dummett's discussion of Fregean platonism is any attempt to connect the semantical features of the view upon which he fastens with the metaphysical imagery associated with realism and its rivals. If there really are abstract objects and if statements about them can be altogether true, then the claim that the Fregean platonist is something less than a realist must amount to the claim that abstract objects as he conceives them are somehow dependent on us or are less than brutely there. But how exactly does this follow from his concession that the notion of reference to abstracta plays no basic role in his theory of meaning? There is no implication that just because our semantic access to them is indirect, abstract objects must be "creatures of language"; no suggestion that they have only the properties we can detect in them, since it is perfectly consistent with the view that verification-transcendent truths about the concrete should generate verification transcendent truths about the abstract according to the translation scheme. Indeed it is consistent with the view that the abstract objects we manage to refer to in this indirect way should have sui generis properties and relations of which we can frame no conception at all. It is characteristic of the Fregean to insist that abstract entities exist altogether independently of us and that our linguistic practice serves only to name what is "already there"; and perhaps the imagery here is not strictly forced upon him. The point remains, however, that Dummett has done nothing to show that the Fregean's semantics precludes this sort of imagery, much less that it commits him to the alternative rhetoric that defines the anti-realist end of the spectrum.
I have run through one set of familiar problems for The Case. I rehearse them only because they seem to remain unanswered in Dummett's latest discussions of the problem, as do the problems mentioned above for the approach to the philosophy of language upon which the position depends. I do not mean to suggest that Dummett does not speak to objections. His approach is nothing if not dialectical. The trouble is that the objections are nearly always voiced anonymously and never correspond exactly to those of any flesh-and-blood interlocutor. In my view it would be a positive service for Dummett to speak more directly to his critics on the subject he regards as his greatest contribution. Dummett's writing is notoriously demanding. By responding directly to critics who are doing their best to engage him Dummett cannot fail to shed light on the absolutely central issues he has done us the favour of raising.
GIDEON ROSEN Philosophy Department Princeton University Princeton, NJ 08544 USA