In this paper, I will discuss three arguments which have been advanced by three of the most important recent analytic philosophers: Willard Van Orman Quine, Hilary Putnam, and Michael Dummett.  Each argument is central to the views of the philosopher in question, and each leads to sweeping and, to my mind, highly implausible conclusions concerning the content of our thoughts about the world.  The philosophers in question claim, of course, that these implications should be accepted, but few others have been willing to follow them in this.  At the same time, however, there has been no very widespread agreement on where and how the arguments go wrong.  My view is that they are best viewed as reductions to absurdity of their premises and of one underlying premise in particular.[1]  But just which premise is at fault is not, perhaps, immediately obvious.  I will have more to say about that after we have had an initial look at the arguments.


I begin with Quine’s argument for the famous thesis of the indeterminacy of translation.  Though it has, as we shall shortly see, an enormously wider application, the indeterminacy thesis is first developed by Quine in application to the situation of radical translation: the situation in which a linguist is attempting to translate a completely unknown language, unrelated to his own, and is therefore forced to rely solely on the observed behavior of its speakers in relation to their environment.  Quine’s claim, in brief, is that while such a radical translator can perhaps succeed, in principle at least, in translating (i) observation sentences and (ii) truth-functional connectives in a determinate, non-arbitrary way, the possibility of determinate, non-arbitrary translation does not extend to the rest of the unknown language.  While the sentences which fall outside these bounds can indeed be putatively translated in a way which will be consistent with all possible behavioral evidence, any such possible translation will, he argues, be only one of indefinitely many different alternatives, all of which are equally satisfactory from a behavioral standpoint and between which only an essentially arbitrary choice is possible.[2]

A simple and by now familiar illustration will help to make clearer the basic thrust of the thesis.  Quine imagines a (putative) word in the unknown language, ‘gavagai’, which is observed to be uttered in the presence of rabbits (or to which the native speakers respond affirmatively when rabbits are present).  His claim is that although the translator can perhaps determine that ‘gavagai’ has something to do with rabbits, he will be unable to determine (on a purely empirical, behavioral basis) whether ‘gavagai’ should be translated into English as ‘rabbit’, or alternatively, for example, as ‘temporal stage of a rabbit’ or as ‘undetached rabbit part’ or as ‘fusion of all rabbits’ (in Goodman’s sense) or perhaps even as the universal ‘rabbithood’.  Which of these translations is chosen will of course have a bearing in turn on which native locutions are to be equated with other English locutions, such as numerals, expressions for identity and diversity, etc.  But Quine’s claim is that it will always be possible to adjust the translations of these other locutions in such a way as to preserve any of the alternative translations of ‘gavagai’ [WO 51-4, 71-2].  (This example is probably misleading in one respect, in that it suggests a much narrower range of variation among acceptable translations than Quine actually seems elsewhere to advocate; but even this degree of variation is sufficiently problematic to raise the issues I want to consider here.)

Quine’s argument for this conclusion concerning radical translation is complicated, obscure in important respects, and raises a number of difficult interpretive issues.  While I do not have time here to enter into a detailed discussion, I think that in the end the argument is best construed as a challenge to Quine’s opponents to show how determinate radical translation is possible, given only behavioral evidence, and I think it is fair to say that this challenge has not been met.  Thus I regard Quine’s initial thesis, the thesis of the indeterminacy of radical translation proper, as reasonably firmly established.  Restated so as to bring out the key point a bit more clearly, that thesis says that no appeal to purely behavioral evidence, no appeal to how the native words are actually used in concrete situations, suffices to determine a unique translation of the native language, so that widely differing translations are, on this basis, equally acceptable.

This initial result, though no doubt a bit surprising, is hardly startling in itself.  The natural response, from an intuitive standpoint, might seem to be that of course the native speaker means one determinate thing by ‘gavagai’, whatever he in fact has in mind, and that the translator is merely unable to tell what that is.  Quine’s conclusion, however, is much more radical and intuitively paradoxical: insofar as such indeterminacy of translation exists, he claims, there is simply no right answer, no fact of the matter as to what the native speaker really means [WO 26-27, 73].  For, it is claimed, the indeterminacy extends not only to our knowledge of the native speaker’s meaning, but to that meaning itself and even to the state of mind of the native speaker which embodies it.[3]  Thus the view seems to be that when the native says ‘gavagai’, he means something having to do with rabbits, but no particular, determinate thing: his thought is somehow intrinsically indeterminate between the various alternatives.

This is surely paradoxical enough.  But the most crucial point is that while Quine develops his argument mainly in relation to the situation of radical translation, he makes it quite clear that its significance is not restricted to that rather unusual situation.  Consider once more the ‘gavagai’ example, but now imagine that we are the native speakers, and that someone else is trying to decide between analogous choices in his language for the translation of our locution ‘rabbit’.  Quine argues that just as we cannot determine on an empirical basis whether ‘gavagai’ means rabbit, rabbit-stage, undetached rabbit-part, rabbit fusion, or rabbithood, so also the radical translator of our language will be unable to decide in a non-arbitrary way between an analogous set of alternatives.  But then, Quine claims, just as in the original case, there is simply no fact of the matter as to which of the alternatives captures what we really mean when we say ‘there is a rabbit’; and just as in the original case, the alleged indeterminacy extends as well to the correlative state of mind.

It is this result which I regard as an unintended reductio ad absurdum of the premises which lead to it.  For surely I do in general have something more definite in mind when I use the word ‘rabbit’ than Quine’s range of equally acceptable alternatives would permit, whether or not someone else can determine from the outside what it is; and any argument which leads to the denial of this entirely obvious fact must be unsound.

How Quine would reply to this sort of objection is reasonably clear: he would say that my “having something definite in mind” reflects only my adoption of what he calls the “homophonic” translation of my own idiolect, i.e. translating ‘rabbit’ as ‘rabbit’:

Staying aboard in our own language and not rocking the boat, we are borne smoothly along on it and all is well; ‘rabbit’ denotes rabbits, and there is no sense in asking ‘Rabbits in what sense of “rabbit”?’[4]

But this response fails to address the basic problem.  For to believe the sentence ‘“Rabbit” denotes rabbits’ is, from a Quinean standpoint, quite compatible with having nothing in mind which would qualify, from an intuitive standpoint, as a determinate referent for ‘rabbit’.  As Quine himself says in another place, “‘Caesar’ designates Caesar and ‘rabbit’ denotes rabbits, whatever they are . . . .”[5]  Moreover, the “homophonic” translation of my own idiolect is in no way mandatory: “Reference goes inscrutable if, rocking the boat, we contemplate a permutational mapping of our language on itself . . . .”[6] and such a “permutational mapping” is, on Quine’s view, always a theoretically available option.  Thus it is apparently only by ignoring alternatives which I know to be in fact available that I can seem to myself to know what I mean even in Quine’s sense.  And this is surely absurd.


I turn now to a second, not entirely unrelated, argument, advanced by Putnam.  Putnam’s aim is to refute a view which he calls metaphysical realism, roughly the two-part view (a) that the world exists in itself--in Kant’s phrase, exists an sich--and has whatever character it has in a way which is quite independent of what humans or cognitive beings of any other kind say or think about it; and (b) that our statements and beliefs, by virtue of their meaning, at least purport to describe this independent, an sich reality.

An immediate corollary of metaphysical realism as thus understood is what Putnam calls a “radically non-epistemic” view of truth: whether what we say or believe is true depends entirely on whether the content of our statements or beliefs agrees or corresponds with the way the world really is and not at all on what evidence we happen to have.  Of course, we may hope that our evidence leads us to an accurate view of the world and may, perhaps, even have good reasons for thinking that it does.  But it would from this standpoint nonetheless be possible, even if perhaps unlikely, for a comprehensive view of the world which was epistemically ideal to be largely or even entirely false--where by an “epistemically ideal” view is meant one which agrees with all the evidence we have or ever could have and which also satisfies ideally whatever methodological standards are appropriate for views of this kind.  And it has seemed to most philosophers that it was relatively easy to describe at least some of the cases in which this possibility would be realized: cases, for example, in which our beliefs and perceptions are produced by a Cartesian evil demon; or, in a more modern, science-fiction version, cases in which we are all disembodied brains, floating in vats of nutrients, and fed beliefs and sensations via electrode stimulation which depict a wholly unreal world.  In such cases, it seems, our picture of the world would be totally or almost totally false, in spite of being in perfect agreement with our evidence and methodologically flawless.

Now metaphysical realism is a heady doctrine at best, and it is easy to portray it in a way in which it might seem itself to be absurd or at least highly dubious.  Though my own belief is that, when properly understood, metaphysical realism is both true and obvious, I can quite well understand some degree of sympathy for an argument which aims to refute it.  What I will claim to be absurd about Putnam’s argument, however, is not the denial of metaphysical realism per se, but rather the way in which this denial is defended and the view which is claimed to be the only alternative.

Putnam’s basic thesis is that metaphysical realism, in particular the claim that a global theory which is epistemically ideal might nonetheless be false, is itself not just false but indeed unintelligible.[7]  From an intuitive standpoint, for such an epistemically ideal theory to be false would be for its terms (or their thought-analogues) to refer to or stand for pieces of an sich reality in such a way as to make false claims about that reality.  But, Putnam argues, in any case that might seem to fit this picture the supposedly false theory can simply be interpreted, or reinterpreted, in such a way as to avoid this result.  And since there will, he argues, be no reason to regard such an alternative interpretation as any less correct or acceptable, the claim that the theory in question might really be false collapses.

It might be thought that a reinterpretation which can accomplish this initially surprising result would have to be extremely subtle or complicated, but in fact Putnam’s method of interpretation could hardly be simpler.  One simply maps the various terms of the theory onto the world in such a way that they combine to make true statements.  Thus, for example, if the ideal theory contains statements in which a certain predicate is applied to certain singular terms (names or descriptions), one simply makes sure that the class which is fixed as the extension of the predicate in the proposed mapping includes the individuals who are fixed as the extensions of those singular terms.  Of course this will have to be done in such a way as to preserve the truth of those statements which express observational evidence and the inferential connections between those statements and other components of the theory, but since the evidence statements are themselves equally subject to reinterpretation, this restriction is much less severe than it might seem.  (The claim that even observational statements are reinterpretable in this way is one respect in which Putnam’s view is even more radical than Quine’s.)

Perhaps the best way to appreciate the radically counter-intuitive character of Putnam’s claim is to look at some of his own examples.  Consider, he suggests, the parts of our own total theory of the world which intuitively depict cats.  There is nothing, he argues, to exclude interpreting the relevant terms in such a way as to yield the result that the term “cat” refers to dogs (and the term “dogs,” to cats) [RTH 218].  Indeed, perhaps even more surprisingly, there is an interpretation according to which the term “cat” refers to cherries and the statement, e.g., that a cat is on a mat says that a cherry is on a tree [RTH 33-35].  And, Putnam insists, both of these interpretations, along with many others which may be as intuitively bizarre as one likes, have just as much claim to be correct or accurate as does the intuitively obvious interpretation according to which “cat” refers to cats.

If these claims are correct, then it is obvious that the metaphysical realist view that the ideal theory might be false is untenable.  For any supposedly false statement in the ideal theory, e.g. one which might be thought to say falsely that a cat is on a certain mat, can just as well be interpreted as a true statement, e.g. to say truly that a cherry is on a certain tree.  There is thus no way for such a statement to be definitely, determinately false, as the metaphysical realist claims that it might be.  (It is worth noticing in passing that it is not at all clear why this result applies only to the ideal theory.)

Now the natural response to these claims is surely to insist that such alternative interpretations, even if acceptable in some narrowly logical or technical sense, can be ruled out on the grounds that they do not accurately reflect the interpretation of the terms in question which is intended by those who actually employ those terms, by us.  There are, Putnam suggests, several different possible versions of how such an argument might go.  It might be claimed, for example, that such an interpretation does not capture the true relation of reference between terms in question and the world; or that under such interpretations, the terms and the objects in the world which are mapped onto them fail to stand in causal relations of the right sort; or that such interpretations fail to adequately reflect the relations between the terms in question and what we see or observe when we employ them.  But, responds Putnam, no such appeal can genuinely establish the correctness of the intuitive interpretation over the bizarre alternatives.  For the terms in which this sort of appeal is couched are, he argues, just more terms or signs in the global theory, equally subject to such reinterpretation.  Thus the interpretation according to which “cat” refers to cherries can perfectly well include a correlative reinterpretation of the term “refer” itself, together with equally compatible reinterpretations of the terms “cause” and “observe.”  And the same will also be true, he insists, for the internal, mental correlates of these terms.

The proposed conclusion on the basis of all this is not, of course, that the epistemically ideal theory must be true after all, at least not if truth is understood in the way in which the metaphysical realist understands it, viz. as correspondence with an sich reality.  It is rather that the very idea of a system of language and thought standing to an sich reality in the definite referential or descriptive relation which the idea of correspondence presupposes is unintelligible and must be abandoned.  This means giving up the idea that we have conceptions of cats and of cherries which pick out definite and different sorts of an sich reality, or indeed which pick out any sort of an sich reality at all.

The alternative, for Putnam, is “internal realism.”  Within our global theory, we can still talk about the reference of our terms; within that theory, “cat” refers to cats and “cherry” refers to cherries.  But if it is asked whether we have any determinate grasp, beyond such trivial, disquotational principles, of what sort of thing a cat or a cherry is, the answer must apparently be that we do not.  Of course, we can also say such things as that cats are furry and that cherries are red; but since “furry” and “red” are equally subject to reinterpretation, being able to make these remarks still involves having no determinate conception at all of what sorts of things we are speaking (or thinking) about.[8]  Quine’s remark quoted earlier in relation to his own view is equally appropriate here (I have adjusted the example):”[‘Cherry’] designates [cherries] and [‘cat’] denotes cats, whatever they are . . . .”

It is this result which I regard once again as a clear though unintended reductio ad absurdum of the premises which lead to it.  For surely I do have something reasonably definite in mind when I use the word “cat,” definite enough at least to exclude cherries, and hence any argument which leads to the denial of this obvious fact must be unsound.[9]


I turn now to Dummett.  The argument which I will consider, though aimed at roughly the same general conclusion as Putnam’s, is quite different in its details from the arguments considered so far.  It arises in the course of a general discussion of truth condition theories of meaning: theories which hold that a speaker’s understanding of the meaning of a sentence consists in a knowledge of the condition under which the sentence is true.  Dummett’s challenge to such a theory is to explain what this knowledge of the truth condition of a sentence amounts to in the case of a particular speaker.[10]

One obvious possibility is that the knowledge in question is explicit knowledge: the ability on the part of the speaker to state the truth condition explicitly in verbal form, in some way other than by merely repeating the sentence at issue.  But this sort of answer will not work in general.  For any such verbal ability obviously presupposes a prior knowledge of the meaning and hence of the truth conditions of the sentences employed in giving it.  Thus if the same question is asked about that knowledge, one must eventually come to a level where the knowledge of truth conditions takes some form other than a capacity for verbal restatement [WTMII, 80].

The only alternative, according to Dummett, is that the knowledge in question is what he calls implicit knowledge, by which he means the sort of knowledge which is manifested in behavior without the person in question being able to offer a verbal formulation [WTMII, 70-71].  It is implicit knowledge in this sense which is involved in various practical abilities, from the ability to ride a bicycle to the ability to distinguish grammatical sentences of one’s native language without being able to formulate explicit principles of grammar.  Such knowledge is, on Dummett’s view, to be identified with the capacity to exhibit behavior of an appropriate sort; and hence, he argues, it can be justifiably ascribed only when it is actually manifested in such behavior.[11]

But what sort of behavior would manifest implicit knowledge of the truth conditions of a sentence?  The only possible answer, suggests Dummett, is that such knowledge would be manifested in an ability to recognize that those truth conditions obtain--or, in more complicated cases, to employ some appropriate procedure leading to such a recognition [WTMII, 80-81].  And the problem which arises immediately is that natural languages contain, or at least seem to contain, very many sentences for which no such recognition is possible and no such procedure exists, either in general or at a particular time: sentences which are not effectively decidable [WTMII, 81-2].  The most obvious examples here are statements about the past or about the future for which no adequate evidence exists, and also certain statements in mathematics, but there are many others as well.

At this point, it will be clearer to focus on a particular example.  Consider then the sentence, “Caesar ate eggs for breakfast the morning before he crossed the Rubicon,” which seems intuitively to have as its truth condition a certain event which may or may not have occurred in 49 B.C.  Assume for the sake of the argument, as might well be the case, that the truth of this sentence is not effectively decidable, i.e. that there no longer exists any adequate evidence of any sort, either recorded in writing or in some other way available, as to what Caesar ate on the morning in question.  It seems intuitively obvious, nonetheless, that we do know what condition would have to be satisfied (or rather would have to have been satisfied) for this sentence to be true.  But what, on Dummett’s account, can this knowledge consist in?

It may be that I can in fact verbally formulate the truth condition for this specific sentence.  But, as Dummett is quick to point out, a sentence whose truth condition is explicitly stated can be undecidable only if one of the sentences employed in stating that condition is itself undecidable [WTMII, 81].  It is thus a consequence of the existence of sentences like the one in question that there are at least some undecidable sentences whose truth conditions are known, if at all, in some way other than by possessing the ability to state them verbally, and we may as well simplify the discussion by supposing that our original sentence has this status.

But now the problem is obvious.  The knowledge of the truth condition of such a sentence is not, we are supposing, explicit knowledge.  But neither can it be implicit knowledge in the sense explained by Dummett, for there is apparently no behavior which would count as recognizing the truth condition of an undecidable sentence and thus no basis for ascribing to anyone such an implicit knowledge.  Thus the conclusion to which we seem to be driven is that, contrary to our intuitive convictions, we do not have such knowledge at all.  Dummett takes this to show that the whole project of giving an account of our understanding of meaning by appeal to a knowledge of truth conditions is fatally flawed and suggests, fairly tentatively, that a verificationist account of meaning should be adopted instead.  But my present concern is not with this more theoretical result, implausible though I believe it to be, but rather with the prior claim that we do not in fact know what historical condition would have to be satisfied for the sentence about Caesar to be true.  For this claim is once again, I submit, plainly and undeniably absurd.


Assuming that I am right in regarding these arguments as reductios, albeit quite unintended ones, what is the premise which should in each case be rejected?  Returning to Quine, I have already said that I regard the initial claim about the indeterminacy of radical translation proper as acceptable, so the focus must be on the argument which leads from the claim that behavior evidence does not fix what the native means to the much more radical result that all meaning and even all thought is intrinsically indeterminate.  Here the situation is a bit murky.  The crucial move is clearly the rejection of any appeal to what the speaker actually has in mind to render his meaning determinate, but there is more than one possible reason which a Quinean might give for this rejection.  Quine himself might well want to invoke a general behaviorist view of mental states, which would automatically rule out an appeal to any internal awareness not reflected in behavior; such a strong version of behaviorism seems to me obviously untenable, and I will not consider it further.  But it seems clear that for many Quineans the key underlying premise is the idea that thought or belief is itself essentially linguistic in character, that, as Harman puts it, belief is an attitude toward a sentence--whether a sentence of a natural language like English or rather, as I shall generally interpret the view, a sentence of a distinct, internal “language of thought,” which we might call “mentalese.”

If this view were correct, if thought itself were nothing more than the internal manipulation of sentences or sentence-like objects, then the original indeterminacy thesis would apply directly to thought as well--assuming, as seems plausible, that the behavioral evidence available to a radical translator of mentalese is, though no doubt more elusive, not different in kind from that available in the original situation of radical translation.  And if this is accepted, then Quine’s seemingly absurd conclusion, e.g. that I have nothing determinate in mind when I use the word “rabbit,” does indeed seem to follow once the original claim about radical translation of public language is accepted.  Once thought itself is construed as linguistic, there is nothing but the way in which the language of thought is used, nothing but what amounts to more linguistic behavior, to fix a definite meaning or content; and such an appeal to behavior, as the original case of radical translation seemingly shows, is insufficient to yield a determinate result.  This both establishes directly the conclusion that thought is indeterminate and also of course undercuts any appeal to thought to avoid the indeterminacy of meaning for public language.[12]

Turning to Putnam, there can of course be no objection to the idea that the sentences of a public language can be treated in the model-theoretic way that he suggests: that they can be taken, as it were, as an uninterpreted calculus and mapped onto the world in an indefinite variety of ways.  The symbols on their own have no intrinsic meaning.  What is implausible is the idea that there is no further constraint of the sort that Putnam refers to as “the intended interpretation” which can rule out such alternative mappings as incorrect.

What form might such a further constraint take?  The answer seems obvious, especially when we reflect that the constraint in question must also be somehow accessible to me if it is to account for the intuition that I know that I mean cats and not cherries by the word ‘cat’.  The obvious suggestion is that such a constraint is provided by the subjective mental conception, the idea or content of thought, that I and presumably others conventionally associate with the term ‘cat’.  It is surely not at all immediately obvious that this mental conception is itself open to model-theoretic re-interpretation, and it seems to preclude correctly applying the term ‘cat’ to cherries.

Putnam’s response to this rather old-fashioned sort of suggestion is quite clear: it is that ideas or thoughts are merely more symbols, “thought-signs,” lacking any intrinsic meaning of their own and thus just as subject to reinterpretation as the words of a public language.    To be sure, he speaks of mental representations as including things such as images that do not seem at first glance to be language-like.  But he proceeds to argue at length [RTH 1-5] that such elements have no intrinsic connection to any sort of extra-mental reality, but instead, like words, function only as signs.  It is for this reason that an image which we would regard intuitively as an image of a cat can just as well be interpreted as a representation of a cherry.

It might be suggested that the conclusion of Putnam’s argument could be avoided while still retaining such a linguistic or symbolic  view of the intrinsic character of thought by invoking some sort of causal theory of meaning or reference.  The basic idea would be that it is actual causal relations, not whatever might be called “causal relations” within a particular linguistic system, which prevent “cat” from being interpreted as referring to cherries and require instead that it be interpreted as referring to cats.  But there are two problems, to my mind insurmountable, with such a suggestion, even apart from the fact that it seems applicable in any very straightforward way only to observation terms.  First, it ignores the fact that the term “cause” in the language of the external observer who is making the suggestion is just as open to reinterpretation as the terms in the original language whose interpretation is at issue.  Thus it is unclear that anyone is in a position to specify the relation which is supposed to do the job in a way that is not equally subject to Putnamian reinterpretation.  Second, even if this suggestion could somehow be made to work, it would not alter the fact that from my own internal perspective, my statements and thoughts concerning cats (as I would intuitively regard them) could just as reasonably be construed as statements and thoughts concerning cherries.  Even if such an interpretation can be ruled out from an external perspective which is unavailable to me (or indeed to anyone), it does not remove the absurd consequence that I, from the inside, have no idea what I am speaking or thinking about.[13]

In the case of Dummett, the reliance on the linguistic or symbolic conception of thought is if anything even clearer.  The basic structure of the argument depends on the assumption that the dichotomy between explicit and implicit knowledge of truth conditions, between knowledge which can be explicitly stated and that which consists in its behavioral manifestations, is exhaustive.  But someone who was not already bewitched by Dummett’s Wittgensteinian perspective would be almost certain to reply that this dichotomy errs by ignoring a third alternative: the possibility of a subjective, mental grasp of the truth conditions of a sentence, one which is not necessarily capable of being linguistically formulated in some way other than the trivial and unhelpful formulation which uses the very sentence in question nor of being fully manifested in behavior.  And it is precisely this sort of understanding which I seem intuitively to have in the case of a simple undecidable sentence like the one in question.  Though Dummett nowhere to my knowledge considers such a response very explicitly,[14] it is abundantly clear that he would reject it by appealing to the view, which he does discuss in many places, that the very idea of language-independent thought is unacceptable.  And, of course, if thought were just more language, equally subject to the same problem, then the dichotomy would be restored and the suggested response would collapse.  Thus, once again, rejecting such a view of thought seems to be the only way to avoid clear absurdity.

There is, however, one other, less radical response to Dummett’s argument that is worth considering.  It might be proposed that his conclusion can be avoided by accepting the idea that one’s knowledge of the truth condition of an undecidable sentence consists in a capacity to recognize that its truth condition is satisfied, while rejecting the claim that such a capacity can be reasonably ascribed only on the basis of behavior that manifests it.  Thus the idea would be that I do have the capacity to recognize the truth condition for the sentence about Caesar’s pre-Rubicon breakfast, i.e. that if I were somehow transported back in time to the appropriate occasion, I would be able to tell by direct observation whether the sentence was true.

Though a full discussion of the issues raised by this proposal is impossible here, there are several reasons for thinking that it does not really solve the main problem.  First, even allowing time travel, it is unclear that I would be able to tell that my trip back into time had reached the right destination, viz. Caesar’s breakfast table in 49 B.C.  Second, even if one rejects Dummett’s essentially verificationist view that a capacity can be justifiably ascribed only on the basis of manifesting behavior, there surely still needs to be some reason for thinking that such a capacity is possessed by someone who ex hypothesi will in fact never manifest it.  The only very obvious suggestion is that such a recognitional capacity might be ascribed to a person on the basis of the manifested ability to recognize the obtaining of suitably analogous sorts of situations.  And the basic difficulty with such an approach is that being in a position to judge with authority that such an analogy obtains seems to presuppose a grasp of the unverifiable situation which goes beyond an unexercised recognitional capacity.

Third and most basically, if the capacity in question is construed strictly as involving only the ability to recognize the situation when presented with it, then even if it can somehow justifiably be ascribed, it still fails to yield in any obvious way an understanding of the sentence in question on the part of someone not so situated.  If my only grasp of the situation that would make the Caesar sentence true is that it is one that I would recognize if presented with it, then I do not now have any real grasp of what it would be for that situation to obtain.  Whereas if, as seems more plausible, my recognition of the situation in question would be guided by some present subjective conception of what it would be like, of what the sentence says, then we are back to the idea of non-linguistic thought--since to construe such a grasp as itself linguistic or symbolic in character would be to reopen the original problem.


My conclusion, though I obviously cannot claim in this unavoidably sketchy discussion to have fully established it, is that the only way to avoid the indicated absurd conclusions is to reject the linguistic view of thought, and accordingly that this view should be rejected.[15]  I am least confident that this is the only way out in the case of Quine, whose argument is by far the murkiest; but in the cases of Putnam and Dummett, at least, such a conclusion seems fairly secure.  I want to finish with three unavoidably brief remarks, the first concerning the precise import of this result and the other two concerning its larger significance.

First.  Exactly how much is ruled out if the thesis that thought is essentially linguistic in character is rejected?  Clearly this question must be answered by considering what is necessary to avoid absurdities like the ones in question.  Though there is a good deal of room for further discussion, my suggestion would be that any view which regards thought as an essentially symbolic process, as composed of elements whose meaning or content is entirely extrinsic to them, dependent on patterns of usage or on external causal connections or associations, will turn out to be unacceptable for the same general reasons.  The implication, then, is that thoughts or thought-elements must, to some extent at least, be intrinsically meaningful--i.e., that my thoughts of cats must by their very nature refer to cats (or at least to reasonably cat-like entities, excluding both dogs and cherries).  Only in this way would it be possible for us to have the sort of internal grasp of the content of our thoughts which it is intuitively obvious that we do, a grasp which both rules out the sort of reinterpretations which Quine and Putnam propose and allows us to understand what it would be for an undecidable thought to be true.  This, of course, is essentially the view which Putnam dismisses scornfully as a “magical theory of reference.”  I am suggesting, in effect, that what seems to him like magic is preferable to clear and undeniable absurdity.

Second.  One reason why this result is important is historical in character: as Dummett and many others have argued, the linguistic conception of thought has a good claim to be regarded as the defining thesis of the analytic or linguistic tradition in philosophy.  While I have no time to go into details, the basic reason for this is obvious enough.  Most at least of the fundamental philosophical problems grow out of reflecting on various aspects of human thought, thought about reality, about knowledge, about value, etc.  It is only if thought is itself somehow fundamentally linguistic in character that there is any obvious reason, and perhaps any reason at all, for thinking that an exclusive or even primary focus on language is the best way to approach such problems.  Thus if the linguistic view of thought is abandoned, at least much of the rationale for the distinctively analytic or linguistic approach to philosophy arguably goes with it.

Third.  The other reason for the importance of my conclusion is more substantive, though also almost entirely negative in character.  What made the linguistic conception of thought (and with it, analytic philosophy) so attractive was precisely that it was, to a very large extent, the only game in town.  What I am suggesting is that if this view of the nature of thought is rejected, it is far from obvious that there is anything currently available to put in its place.  In particular, having argued as I have that thoughts must be intrinsically meaningful or contentful, I must confess to having no very clear or developed account of how this could be so.

Putnam’s suggestion is that such a view of thought will have to embrace an extreme Platonism by positing that the mind has the irreducible, non-natural power to directly grasp forms or universals.  I, unlike Putnam, have a good deal of sympathy for this sort of possibility--and even more for the general suggestion that an adequate account of thought will have to repudiate recent “naturalist” dogmas and indulge in some fairly hard-core metaphysics.  But while I think that some such approach is very much in the right direction, it is far from obvious at present how it should go in detail.[16]  The clear upshot, if I am right, is that on the question of the nature of thought itself, perhaps the most fundamental philosophical issue of all, we may need to start over almost from the beginning.[17]

Laurence BonJour

University of Washington

[1]    I also believe that a similar diagnosis, involving essentially the same implicit premise, should be made of Kripke’s neo-Wittgensteinian defense of skepticism about meaning.  But there is no room in this paper for an adequate consideration of the issues raised by that discussion.

 [2]   The indeterminacy thesis was first developed in Quine, Word and Object (Cambridge, Mass.: MIT Press, 1960) [hereafter WO], chapter 2.  It has since been elaborated and refined in many other places.  Though a full account of the thesis would involve many further details and ramifications, most of these are inessential for present purposes.

[3]    Quine would not, of course, use the term ‘meaning’; but I can see no real basis for such a stance and hence no reason to deprive ourselves of this useful formulation.

[4]    Quine, Theories and Things (Cambridge, Mass.: Harvard University Press, 1981), p. 20.

[5]    Quine, “Reply to Robert Nozick,” in Lewis Edwin Hahn & Paul Arthur Schilpp (eds.), The Philosophy of W. V. Quine (La Salle, Ill.: Open Court, 1986), p. 367.

[6]    Quine, Theories and Things, p. 20.

[7]    The most important places where Putnam discusses this view are his paper “Realism and Reason,” reprinted in Putnam, Meaning and the Moral Sciences, pp. 123-140 [hereafter RR]; and his book Reason, Truth and History (Cambridge: Cambridge University Press, 1981) [hereafter RTH].

[8]    Putnam admittedly often writes as though internal realism involves more than this, as though our terms, though entirely indeterminate in their reference to an sich reality, somehow nonetheless refer determinately to an “internal reality,” to something like Kant’s phenomenal world.  I confess to having no clear understanding of what this sort of “reality” could amount to--or, more importantly, of why the same argument could not be run in relation to it.

[9]    I say “reasonably definite” because it does not matter for the present point whether my subjective conception of cats is sufficient to distinguish real earthly cats from, e.g., robot cats or twin-earth cats.  It is enough if it is sufficient to clearly pick out real earthly cats as contrasted with real earthly cherries.

[10]   Dummett formulates versions of this general sort of argument in a variety of places.  In the present discussion, I am relying primarily on his paper “What is a Theory of Meaning? (II)” [hereafter WTMII], in G. Evans and J. McDowell (eds.), Truth and Meaning (Oxford: Oxford University Press, 1976), pp. 67-137.

[11]   In a later discussion, Dummett rejects the phrase “implicit knowledge” as an adequate characterization of such non-explicit knowledge, on the grounds that this phrase should be used only for cases where the individual in question can recognize a formulation of the knowledge in question when presented with it.  He is also much less satisfied with the view that such knowledge is essentially practical in character.  See his book, The Logical Basis of Metaphysics (Cambridge, Mass.: Harvard University Press, 1991), pp. 93ff.  But this does not affect the central claim that the knowledge in question consists in the way in which it would be behaviorally manifested.  Cf., e.g., ibid., p. 104.

[12]   Quine’s original argument is admittedly verificationist in character, and it might be thought that a rejection of verificationism would be enough to avoid the conclusion.  This would be so, however, only if some plausible though non-verifiable basis for determinacy was available, and I am unable to see what that might be, once the linguistic conception of thought is accepted.

[13]   The same sort of objection seems to me to apply to David Lewis’s considerably more sophisticated response to Putnam, which invokes as a constraint on acceptable interpretations of a language that it must respect “objective sameness and difference in nature,” i.e. that it must carve the world at its “objective joints.”  Unless such “objective joints” are somehow accessible to me, I still won’t be able to know what I mean or what I am thinking about.  See David Lewis, “Putnam’s Paradox,” Australasian Journal of Philosophy 62 (1984), pp. 221-36.

[14]   The closest he comes is in The Logical Basis of Metaphysics, where he offers the following reply to the realist: “the realist may contend that what is required for a grasp of the meanings of our expressions is . . . the formulation of the right mental conception of the principles underlying [the rules that govern their use]” [344].  But he proceeds to interpret this suggestion in terms of the recognitional capacities of “a hypothetical being with superhuman powers” [345], where those powers are conceived by analogy with our own, and then objects to it on that basis.  I can see no reason, however, for such a construal, other than the verificationism which is supposed to be a conclusion, rather than a premise, of the argument.

[15]   For a more direct argument against the linguistic or symbolic conception of thought, see my paper “Is Thought a Symbolic Process?” Synthese, vol. 89 (1991), pp. 331-52.

[16]   See the paper cited in the preceding footnote for some initial suggestions.

[17]   An earlier version of this paper was read at the University of British Columbia and a somewhat abbreviated version at the Western Division Meetings of the American Philosophical Association held in New Orleans in 1990; I am grateful to my APA commentator, James Sterba, and to the audiences at both occasions for useful comments and suggestions.  I am also grateful to Ann Baker, Carl Hoefer, Genoveva Marti, and Lawrence Goldstein for further helpful comments and discussion.