You are on page 1of 21

Mind Association

Fanciful Arguments for Realism

Author(s): Alan H. Goldman
Source: Mind, New Series, Vol. 93, No. 369 (Jan., 1984), pp. 19-38
Published by: Oxford University Press on behalf of the Mind Association
Stable URL: .
Accessed: 13/09/2011 23:39
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact
Oxford University Press and Mind Association are collaborating with JSTOR to digitize, preserve and extend
access to Mind.
Mind (i 984) Vol. XCIII, I9-38
Fanciful Arguments for Realism
i. Throughout the long history of the realism-idealism debate, the
realist has encountered notorious difficulty in specifying clearly his
notion of independence, when he claims that the world is in-
dependent of our perceptions and concepts. Michael Dummett has
suggested a way that interprets realism as a- semantic and epi-
stemological, as much as a metaphysical, thesis.1 For him the
realist's claim is that meaning must be defined in terms of truth
conditions, and that truth here must be contrasted with verification
or warranted assertibility. The realist takes his assertions to state
that their truth conditions are satisfied, whether or not we are in a
position to tell that they are. The independence in question is the
irreducibility of truth to warrant, coupled with the analysis of
meaning in terms of truth.
Realism in this semantic sense must be thoroughgoing in order to
have any plausibility. We cannot be realists about truth but not
meaning. We cannot, as some contemporary philosophers seem to
suggest,2 analyze meanings as given by justification conditions and
then continue to contrast justification with truth. This leads to the
absurd result that to assert that x is F is to assert that x is probably F
but may not be. Since these assertions are not equivalent, if
meaning is to be given by justification conditions, then truth must
reduce to warranted assertibility as well. If we accept the antecedent
but not the consequent, we must also renounce the equivalence in
all but observational contexts between 'It is true that p' and 'p',
since the latter would assert only justification conditions and the
former truth. The semantic realist denies the antecedent as well as
the consequent.
Dummett, of course, has challenged the ultimate viability of the
more thorough semantic realism. The basic problem for him is to
I Michael Dummett, 'Realism', in Truth and Other Enigmas (Cambridge, MA:
Harvard University Press, I978); see in the same collection, Preface, 'The
Reality of the Past,' and 'Truth'.
2 See, for example, John Pollock, Knowledge and Justification (Princeton:
Princeton University Press, I974),
connect the realist notion of truth to the practice of language users,
to show how it helps to explain linguistic practice. This demand is
legitimate given the supposed link between meaning and truth. To
provide a theory of meaning is to show what a language user knows
when he understands sentences in the language. If that knowledge is
knowledge of the conditions under which the sentences are true, it
must still be manifest in their use. For understanding must be
taught, acquired and judged in the context of public communi-
cation. Meaning, and truth as its explanans, cannot transcend all
manifestation in practice, or we could never know whether others
understand us (not even perhaps whether we understand our own
Dummett's challenge addresses the intelligibility or possibility of
realism, not the truth of the position traditionally construed. He has
not really provided a translation of the metaphysical thesis, but at
most clarified a necessary condition for its truth. That meaning is
analyzed in terms of truth (as opposed to warranted assertibility)
conditions entails that we intend our terms to refer to a deter-
minately structured, independent reality, that we assert the exist-
ence of such states of affairs. It does not show that these states of
affairs obtain or that our realist concepts are instantiated.
But while not sufficient for its truth, the truth conditions analysis
of meaning and the contrast between truth and assertibility are
necessary for the intelligibility of the realist's thesis. Dummett's
challenge must therefore be met before the traditional issue can be
2. Several more preliminary remarks are in order regarding
another formulation of the semantic realist's thesis by Dummett.
For him the truth conditions analysis equates with belief in the
principle of bivalence, the idea that each statement is determinately
true or false. Several paths lead to this equation, the clearest
through the conception of the realist theory of meaning along
Tarskian or Davidsonian lines. A Tarskian theory of truth entails
all sentences of the form 'S is true iff p', where S names the sentence
that replaces p, or p translates the sentence named by S. If a theory
of this form is to serve as the realist's theory of meaning, it is clear
that he must accept the principle of bivalence. Otherwise, the
designated sentence in the Tarskian schema might fail to be true
when the sentence on the right is not false. The equivalence will
then fail.
Bivalence might also seem on intuitive grounds to be a require-
ment for the realist. The crucial cases for his position appear to be
those in which there is no overriding evidence for a statement or
class of statements, or those in which statements are conceived as
possibly false despite the balance of evidence as projected in-
definitely into the future. Here he claims that such statements are
true or false independently of the evidence. Lack of evidence does
not imply lack of truth or falsity, and positive evidence does not rule
out falsity. Concentration on such cases can easily lead to the
generalization of the realist's claims in regard to them to the
principle of bivalence, to the claim that all statements are deter-
minately true or false (whatever the evidence fo^r them might be).
But despite such considerations, bivalence is neither necessary
nor sufficient for realism, semantic or otherwise. First, regarding
sufficiency, consider phenomenalism as a form of anti-realism
regarding physical objects. The truth values of statements about
objects are not logically independent of statements of their total
evidence on this view, but the principle of bivalence may be
satisfied. It will be satisfied if the subjunctive conditionals in the
phenomenalist translations of physical object statements are
allowed only when supported by evidence. The point here is more
easily made, however, in relation to other domains of discourse. In
the philosophy of law, for example, Ronald Dworkin has argued
that every question of kIw has a determinate answer; every
statement of the form 'It is the law that p' is determinately true or
false.1 But truth here, of course, is of the coherence variety. A
statement of law is true if it is more consistent with the body of
established law than is its negation (not if it corresponds to some
independent reality). One might well adopt a similar position with
regard to moral judgments and the data base of settled moral
convictions. In both cases bivalence can hold while the hallmarks of
realism-the distinction between truth and warranted assertibility,
the notion of correspondence, the conceivable falsity of a univers-
ally endorsed, maximally consistent theory-are absent.2
Regarding the necessity of the principle for the realist, bivalence
can fail for reasons relating to portions of our language, more deeply
to our perceptual and conceptual abilities, rather than to any lack of
I Ronald Dworkin, Taking Rights Seriously (Cambridge, MA: Harvard
University Press, I 977), chs. 2-4.
2 Compare Simon Blackburn, 'Truth, Realism, and the Regulation of Theory',
Midwest Studies in Philosophy, V (I980), pp. 353-37I.
independence or determinateness in the world. The realist's
position, as characterized by the criteria mentioned at the end of the
previous paragraph, is not threatened by vagueness in our terms or
concepts, although vagueness renders problematic the assignment
of truth values to certain statements. Realism entails belief in a
determinate reality, but not in a fully determinate language with no
'open concepts'. As others have pointed out, the vagueness of many
predicates is no accident or oversight to be remedied by simple
stipulative adjustment. It typically characterizes, for example, not
only observational predicates like colour terms, but ordinary kind
terms as well. To use Quine's latest version of the sorites paradox,'
if we imagine subtracting one molecule at a time from the surfaces
of a table, we will be unable to draw a precise line at which 'x is a
table' is no longer true of the object. Thus bivalence is problematic
for a range of cases. Here it is clear that the problem lies not in any
lack of determinateness in the realist's world. The world may be
completely determinate in every space-time region and yet fail to
determine 'a is a table' as true or false, because the concept of a table
cannot be sharpened to the same degree.
There are reasons irrelevant to our point here for doubting that a
Tarskian truth theory can serve as a theory of meaning. The main
one is that the equivalence relation in the test sentences seemingly
needs to be stronger than material equivalence in order to guarantee
that the sentences on the right side translate those named on the
left.2 Davidson seeks to maintain extensionality via constraints on
the generation of the theory, creating relations among the Tarskian
sentences that achieve the required guarantee.3 What I have noted
here is that such a theory, with its implied principle of bivalence, is
not necessary for the semantic realist. But the analysis of meaning in
terms of truth, and the contrast between truth and verification or
warranted assertibility, remain necessary. We may now turn to
Dummett's arguments against such notions of meaning and
W. V. 0. Quine, 'What Price Bivalence ?', Journal of Philosophy, 78 (i98i), pp.
2 Even entailment or equivalence in all possible worlds will not by themselves do
the trick, since necessary truths can differ in meaning.
3 See, for example, Donald Davidson, 'Reality without Reference', in Mark
Platts (ed.), Reference, Truth and Reality (London: Routledge and Kegan Paul,
I980); see also Davidson, 'Truth and Meaning', Synthese, 17 (I967), pp.
3. Dummett suggests three related arguments against the realist's
concepts of meaning and truth. All three derive from the following
claims: (i) that meaning is what a language user knows when he
understands a statement; (2) that a theory of meaning must provide
a systematic account of such understanding; and
that linguistic
understanding and the knowledge that underlies it must be
manifest in practice or communicative use. Dummett questions the
compatibility of the realist's analysis of understanding, i.e. knowl-
edge of meaning, as knowledge of evidence-transcendent truth
conditions with the thrust of these three claims.
His first argument derives more specifically from considerations
regarding language acquisition. He points out that we learn to use
sentences correctly by being taught to make assertions with them in
conditions taken to establish the truth of those assertions.' If
understanding is what is acquired through this training process,
then it becomes problematic how we could understand what it is for
an assertion to be true independently of knowing what evidential
conditions count in its favour.
Dummett's second argument derives from his demand that the
realist provide an account of the knowledge of truth conditions that
is to constitute understanding of meaning. According to him,
stating truth conditions in Tarskian form cannot give us a grasp of
the designated sentence's meaning, since we must already under-
stand the meaning of the latter sentence in order fully to grasp the
sense of the Tarskian sentence that contains it.2 The same question
might be raised regarding knowledge of assertibility conditions,
since we might want to represent such knowledge by suitably
transformed Tarski-type sentences. But there are several important
differences here. First, non-trivial statements of assertibility con-
ditions will be possible for a broader class of sentences than in the
case of truth conditions, since evidence for assertions is more often
statable in terms other than those of the assertions themselves.
Second, and more important, assertibility conditions can be taught
and expressed directly by assertions in those actual conditions. This
will not be possible for truth conditions in the differentiating cases.
Dummett's third, closely related argument expresses the demand
that understanding, knowledge of truth conditions according to the
I Compare Dummett, 'Truth', ibid., p. 362.
2 Dummett, Frege: Philosophy of Language (London: Duckworth, I973),
P. 458.
realist, be manifest in use.1 If the realist's notion of understanding
as transcending knowledge of verification conditions is to be
genuinely different from its verificationist counterpart, it must
make a difference to the practice of language users. But if using
sentences correctly is using them to make warranted assertions, it
again becomes problematic how such use could justify ascription of
understanding that transcends conditions of warrant.
Dummett himself suggests in several places that the notion of
truth derives from a broader notion of linguistic correctness in
certain contexts in which appeal to verification or warranted
assertibility seems insufficient to determine correct use. In the
formation of certain compounds, the assertibility of the com-
ponents does not imply the assertibility of the whole sentences.
These non-assertible-functional compounds require an auxiliary
notion to account for assertibility conditions themselves. Dummett
sometimes suggests that the notion of truth plays this role.2 Robert
Brandom has systematized those examples that support this line of
argument.3 In general, he points out, whenever one clause of a
compound expresses a state of the speaker and the other ascribes
that state to him, the compound may not preserve assertibility.4 As
a simple case, consider 'I believe that p', 'p', and 'If I believe that p,
then p'. Sentences formed from the first two schemas will have the
same assertibility conditions (leave aside the complication that a
person may know that he believes that p for stupid reasons), but
assertibility does not transfer to the compound. Here we require a
stronger criterion for the second clause, and the notion of truth
suggests itself. If p is true, and not simply assertible by this speaker,
then the compound will be true.
Unfortunately the realist's conception of truth is not required
here-a notion of warranted assertibility broader than that of
verifiability by particular speakers at particular times will do. That
the former is not necessary becomes clear when we consider the
same problem as it arises when p is a moral judgment. It does not
follow from my believing that an action is right that it is right,
although the assertibility for me of the components may be the
same. This does not establish realism as a meta-ethical theory,
I Dummett states the argument, among other places, in 'What is a Theory of
Meaning (II)', in Evans and McDowell, ibid., p. 70.
2 Dummett, Frege, pp. 420-42I, 449-450.
3 Robert Brandon, 'Truth and Assertibility', Journal of Philosophy, 73 (1976),
4 Ibid., p. I44.
however. We need only distinguish between the speaker's warrant
at a given time and the convergent and maximally coherent
judgment of the moral community at the limit of moral develop-
ment. Verification is certainly distinct from warranted assertibility
in the Peircean sense, since the latter may not be accessible to
current speakers. But the realist's notion of truth also is distinct at
the other end of the spectrum from Peircean assertibility. It
transcends this broader nonrealist concept, since for the realist even
convergent and maximally coherent theories can be false.
Thus, to account for the failure of assertibility perservation in
compounds such as that mentioned above, we need only posit a
linguistic division of labour in which first person warrants for self-
ascribing certain states override, while some form of collective
warrant or criterion of coherence overrides in expressions of the
contents of those states. Certain other contexts in which we contrast
truth with justification can be accommodated in the same way by
the anti-realist, in terms of the contrast between narrower and
broader conditions of warrant. Conceptual change often results
from criticism of beliefs about warrant, as warrant is supposed to
generate truth. In the clearest cases we discover that procedures of
verification lead to conflicting judgments. The demand to abandon
some of these procedures, or to order them so as to remove the
inconsistency, appears to derive from the concept of a single
transcendent truth at which we aim.' But once more it can derive
instead from the goal of consistent and convergent theory itself.
The distinction between narrower verification and ultimate collec-
tive warrant again allows us to avoid appeal to realist notions of
truth and meaning. The challenge remains to show how these
concepts are necessary to an account of linguistic practice, how they
are manifest in use.
4. In fact two questions of Dummett remain to be answered: 'How
is it possible to acquire understanding in the realist's sense?' and
'How is such understanding manifest in use?' We may begin with
the former. Dummett himself suggests, only to dismiss, that we
may acquire understanding beyond the ability to verify through
imaginatively extending our grasp of verification procedures them-
selves by analogy. By imagining creatures in more advantageous
Compare David Papineau, Theory and Meaning (London: Oxford University
Press, I979), p. 9I.
positions in space-time or with superior intellectual capacities for
acquiring corroborating evidence, we come to understand what
would make our unverifiable sentences true, hence in what their
truth consists.' He dismisses this first fanciful response by the
realist, however, on the ground that we could not manifest such
imaginative grasp of meaning in use. Perhaps the point is that our
use of sentences must continue to be governed by whatever actual
justification procedures we have.
It does not appear to be necessary to continue the argument in
this direction, since I do not believe that the realist needs to extend
by analogy his grasp of verification for each unverifiable sentence he
may claim to understand. The demand first of all oversimplifies the
way that understanding of sentences is built up from understanding
of terms. Terms can be learned originally from their contributions
to verifiable assertions, or from their functions in explanatory
models grasped partly by analogy, or from combinations of other
terms learned in these ways. They can be recombined into
sentences and theories that, in Quine's terms, face the tribunal of
experience only holistically. Even the broader theories face this
tribunal not alone, but in relation to competing alternatives. The
latter point is important here, because a theory thought to be
verified or justified by evidence at a given stage of inquiry can be
replaced by a theory more explanatory of that evidence without the
former's becoming thereby unintelligible. An outmoded theory
may lack conceivable evidence or further justification, all con-
ceivable evidence being reinterpreted in the terms of the new
theoretical vocabulary. But we may nevertheless understand the old
theory (e.g. that of phlogiston) as well as did our predecessors, for
whom it was justified.
Examples abound within science and without, including those
within our most extensive domain of fanciful discourse, religion
and myth. Religious explanations of empirical phenomena once
may have competed as serious scientific contenders; they may have
been justified by the evidence in relation to available alternatives. At
present, however, an atheist can refuse to accept any conceivable
evidence as verifying the existence of a deity, while at the same time
intelligibly denying that existence. He can also deny the possibility
of conclusive falsification, while continuing to understand the
Dummett, 'What is a Theory of Meaning (II)', in Evans and McDowell, ibid.,
pp. 98-99; Frege, p. 465.
assertion of existence. Evidence here would consist in observations
best explained as divine manifestation, but in this case some theory
of hallucination might be preferable as a last resort alternative.
Whether or not the latter would be explanatorily preferable, and
whether or not the atheist is able to picture a creature better situated
to gather verifying or falsifying evidence, his religious talk need not
lack sense.
5. If it is not necessary to extend our notion of justification or
evidence by fanciful analogy in order to derive realist concepts of
understanding and truth, how are these concepts derived and
distinguished from their nonrealist counterparts? It is through
extending our primitive notion of error, not verification, that we
arrive at the realist's conception of truth as independent of
evidence. For the nonrealist error remains conceivable only in
relation to experiences and beliefs taken to be veridical: fallibility
entails corrigibility. For the realist the truth value of a mistaken but
justified assertion does not change when it is shown to be false.
It would have been false even had the counterevidence never
materialized. The realist extends this concept by analogy first into
one of possible error despite the weight of all conceivable evidence
and then into that of possible global error, possible falsity of our
ultimately corroborated theories. Thus he is committed to the
intelligibility of global scepticism, while the nonrealist must deny
not only the force, but the sense, or the sceptic's pervasive doubts. 1
Traditionally these doubts have been made seemingly intelligible
to us with the aid of some fanciful creatures-evil demons and,
more recently, brains in vats. The latter are to illustrate fully
coherent belief systems that are uniformly false in ascribing
properties to nonexistent objects, objects that exist only in the
minds of the brains. The former are to make a similar point, that our
own beliefs, even when maximally coherent or verified, may be
themselves systematically false. But these prodigal offspring of
prior epistemologists may be inadequate for their tasks, either
sceptical or realist.
There are several possible replies to these modern philosophical
One might assume a realist position for certain propositions (by holding that
their truth conditions transcend verification procedures) without countenanc-
ing the possibility of global scepticism. But once truth as correspondence is
distinguished from coherence, it certainly appears logically possible that most
of our coherent beliefs might fail to correspond. See below pp. 33-34.
myths; all would argue that they fail to demonstrate the possibility
of maximally coherent, false beliefs. To begin with the brains in
vats, if these are programmed to have beliefs shared by the scientists
who maintain them, their beliefs might be taken to refer to ordinary
objects via deviant causal chains (interpreted nonrealistically)
through the beliefs of the scientists. The former could then be
deemed true or false of these objects by means of the usual
coherence tests. If the scientists rather programme wildly false but
internally coherent and verified beliefs into their brains, two further
nonrealist responses might be offered. The beliefs of the brains
might be deemed false in this situation by failing to cohere with
those of the scientists and other inhabitants of their shared world.
Or the nonrealist, if wary of this union of the world of the scientists
and that of their brains, could hold that the brains occupy their own
worlds, those defined by their sets of beliefs. Within those worlds
their beliefs could be held true when coherent or verified, and true
of the objects in those worlds as their beliefs define them.
If the point of the brains in the vats case is rather that we might be
such and hence cannot trust our own coherence tests for truth, we
assimilate this case to that of the evil demon. The initial reply to this
spectre is that we have good reason to dismiss him, since the
explanation for the nature of our beliefs that appeals to him as their
cause is far less satisfactory on many counts than that which appeals
to ordinary objects, whether or not realistically construed. The
rejoinder of the realist here is that the mere possibility of the demon
illustrates once more that coherence does not entail truth. A satis-
factory reply to the sceptic need not bother the realist, given their
different uses of the demon's services. But the final reply of the
nonrealist here can echo that of the previous paragraph. For the
nonrealist our willingness to countenance the demon need not shake
our faith in the truth of our maximally coherent sets of beliefs.
Within our world, the world defined for us by those sets, such
beliefs remain true of the objects they specify. We can imagine them
false only by introducing new beliefs with which they fail to cohere,
by imagining them falsified. If we were to accept belief in the
demon, we would have to alter other beliefs in the prior system. But
the suggestion that the demon might exist although never to be
countenanced by our beliefs or revealed in our experiences can be
dismissed as incoherent by the staunch nonrealist.
A reply similar to this last offered to the evil demon and brains in
vats is suggested by Putnam's recent anti-realist argument. Putnam
considers the global case of a complete theory or set of beliefs that
meets all the operational and aesthetic, coherence and pragmatic
constraints on theories judged true.1 According to him we can
simply choose the referents of this theory so as to make it come out
true of those referents under our ordinary (Tarskian) concept of
truth. When we so map a model of the theory onto objects in our
world so that all such criteria are satisfied, we cannot make sense of
the claim that this theory might be false. This would amount to the
claim that our mapping does not pick out the real intended objects
as they are related according to the real meanings of our terms, that
it is the wrong mapping. But why is not the interpretation that maps
onto those objects (and properties) that make the theory come out
true the correct one?
In the brains in vats and evil demon cases, Putnam's argument
would imply the suggestion offered by the nonrealist above: the
maximally coherent beliefs of the brains and demon-victims are
true of the objects in the worlds accessible to them, what we from
the outside would consider their phenomenal worlds.2 There may
appear to be a problem here with certain of their coherent beliefs,
for example the belief that objects continue to exist unperceived and
unconceived; but possibly such apparently realist beliefs could be
given some Millian interpretation that would make them true also
(or they could be shown to be false in Dummett's fashion by being
shown to be inconsistent with the brains' genuine understanding of
object terms). There is yet a third option, that of maintaining that
these propositional attitudes fail to refer and therefore lack truth
value. Here the lack of entailment from coherence to truth lacks a
realist ring, since it does not imply the independence of the world to
which reference is made from evidence or theories. Then too, under
this interpretation we might hesitate to call the attitudes in question
beliefs at all. In any case, the freedom we have in assigning or
refusing to assign referents means that the myth fails to force a
realist interpretation upon us. It fails to drive the wedge between
warranted assertibility and truth.
I Hilary Putnam, 'Realism and Reason', in Meaning and the Moral Sciences
(London: Routledge and Kegan Paul, I978).
2 In a more recent book Putnam has made this argument more explicit by holding
that the brains could not refer to real objects because they do not causally
interact with them. See Putnam, Reason, Truth and History (Cambridge:
Cambridge University Press, I 98 I), chs. I-2.
6. Use of the notion of global error has not succeeded in demon-
strating the requirement for a realist concept of truth, since the
nonrealist appears free to deny the intelligibility of that notion. The
intelligibility of evil demon and brain in vat stories does not affect
that freedom, since the nonrealist can assign sense to certain
interpretations of them. Perhaps our realist protagonist will do
better to return to the more local level for the time being. To
demonstrate a lack of entailment between verification or coherence
and truth, we seek a belief that is conclusively verified, that remains
forever consistent with others and immune from falsification, but
that is false.
To illustrate the possibility, let us add a new creature to the
metaphysical zoo, or rather borrow one from recent literature in the
philosophy of mind. Imagine then a person S and his physical
replica R, an atom for atom duplicate.' In this story shortly after R
materializes, S dematefializes without a trace. R's seeming mem-
ories now match S's former genuine memories. Specifically, R
remembers having had an unhappy childhood, which, of course, he
never had. It appears, nevertheless, that his belief must meet all
pragmatic and coherence tests of truth: it can be verified but not
falsified. Complete evidence here will fail to entail truth; it appears
to be the realist's notion of correspondence, or lack of it, that
determines the falsity of R's belief.
An initial nonrealist response to this story might be to say that in
it S simply becomes R, so that R's belief is true of his, i.e. S's,
childhood. This move can be blocked, however. First, if we are
allowed to use material criteria of personal identity, we can point
out that there has been no physical process of division in this story.
If such criteria are rejected, we can point out that there need not be
even deviant causal chains between S's childhood and his replica's
belief. We can imagine that the replica just happened to materialize
as a physical duplicate. The causal criterion of memory cannot be
rejected, although the non-realist will provide his own interpre-
tation of cause to accommodate it. There clearly is a difference
between a genuine memory and one which accidentally, though
accurately, represents a past occurrence. The required connection
Replicas have recently made their way from the literature of science fiction to
that of philosophy. See, for example, Stephen Stich, 'Autonomous Psychology
and the Belief-Desire Thesis', The Monist, 6i (1978), pp. 573-59I; Jaegwon
Kim, 'Psychophysical Supervenience', Philosophical Studies, forthcoming. My
replica serves a purpose different from those of his predecessors.
is not present in the story I am telling, making R's belief false
although seemingly consistent with all evidence.
A second reply might accept the falsity of R's belief, but hold that
it is its failure to cohere with our understanding of the situation that
leads us to declare it false. This reply will not work either. Beliefs
and the propositions they express are true or false within possible
worlds. R's belief is false within his world and not simply in relation
to ours. For the reply to be sound, his belief about his childhood
would have to be true in his world although false in ours. But his
belief is false in his world, although coherent with all other relevant
beliefs in that world. Hence this reply is of no avail.
There is a point nevertheless to emphasizing the special point of
view from which we are considering this highly artificial case. In
defining the possible world as we choose, we have special access to
the correspondence relations between beliefs and facts within it. In
our world we rather share the fate of our science fiction counterparts
in lacking such direct access. Correspondence to fact cannot be our
criterion of truth, if by 'criterion' we understand a test that can be
applied to determine the applicability of a term or concept. We are
limited in our world, as opposed to those fanciful worlds of which
we can acquire a transcendent view, to reliance upon coherentist
criteria. The realist will then be obliged to link up the criterion of
coherence with his correspondence notion of truth itself, if the latter
is to be of use. I shall return to this difficulty briefly below.
A final nonrealist response here might be to imagine the
possibility in the replica's world of future verificationist tests that
might show his belief to be false. Many such conceivable tests, we
have seen, can be ruled out in the definition of the fanciful case. But
perhaps not all. Ultimately it seems that the nonrealist simply can
take whatever criterion of difference we apply in saying that R is
distinct from S and imagine a possible test for that difference.'
Since this test will express our notion of identity and nonidentity,
any attempt to rule it out will render unintelligible the claim that R
replaces S and therefore has false beliefs about his past. If our
criterion is physical identity, for example, then we can imagine a
molecule tracer that will show R's physical makeup to be non-
continuous with S's. The falsity of R's belief about his childhood
will then be held to derive from its lack of coherence with the
outcome of this conceivable verificationist test.
I This possibility was pointed out to me by Eddy Zemach.
Several final replies are in order. First, since the test in question is
unavailable (in R's world), we can maintain that the notion of truth
and falsity applicable to the case remains realist. Dummett permits
the realist to imagine nonactual verifying tests while retaining
his concept of truth. The second point, addressed to stronger
definitions of semantic realism as well, is that we may question
whether the conceived test in itself would demonstrate the falsity of
R's belief. Imagining that such cases have never occurred pre-
viously in the replica's world, that all other evidence continues to
point to the truth of his belief, overall maximization of coherence
would seem to call rather for dismissing the reliability of the test or,
as a last resort, interpreting its results as illusory. Again it is only our
appreciation of the lack of correspondence to (fanciful) fact that
leads us to interpret the conceived test as veridical; the realist
notion, here applicable from the transcendent viewpoint, does all
the work. Coherence alone still calls for assignment of truth, despite
the introduction of a single inconsistency that must and would be
7. The point of introducing this more localized counterexample to
the idea of verifiability or coherence as truth was to block the
countermoves to the traditional holistic fanciful cases. First, unlike
the brains in vats, there is no question of the replica's belief about
his childhood being true via some deviant causal chain, since he had
no childhood, and since S's childhood is causally unconnected.
Second, there is no lack of coherence with beliefs of others in his
society, as was the case with the brains. Finally, and most im-
portant, the nonrealist cannot take the replica's beliefs to define a
world for him within which they are true. We cannot assign
phenomenal referents here, since the reference of terms like
'childhood experience' will be fixed by uses of such terms to refer
successfully by other members of R's linguistic community. His
belief cannot be taken to refer to his imaginary experience, since
there will be a distinction in his world, as in ours, between
imaginary and genuine experiences and childhoods. We cannot
then map the belief onto an object that would make it come out
true, despite the belief's meeting all operational and verificationist
The intelligibility of this new fanciful case reveals the fallacy in
the antirealist argument of Putnam. In the global case, as Putnam
argues, appeal to a causal theory of reference cannot help the realist,
since 'cause' can be given a globally nonrealist interpretation too.1
But in the local case causal relations exist in R's world between
genuine experiences and later memories of them and references to
them that do not obtain for him. This fact, together with the
linguistic division of labour and the fact that he intends to use his
terms as others use them, fixes the interpretation of his belief in a
way that makes it come out false despite meeting all coherence tests.
The moral for Putnam's argument is that our 'theories' (taken in the
broadest sense in which a set of beliefs and linguistic practices
together constitute a theory of the world) can build in structure
beyond the ways in which they can be tested. There may be
intentional structure in a theory (or between a particular theory and
other facets of linguistic practice) that prevents the one-to-one
mapping of its model onto the world from succeeding. This lack of
congruence between relations among real objects and such internal
relations among terms of the theory may not show up operation-
ally.2 In the local case the mapping that makes the belief or set of
beliefs true may not capture their intended referents. Here we can
make sense of the latter notion, as Putnam argues we cannot in the
global case, because reference is determined by practice outside the
local belief or theory that borrows and assumes the fixation of
referents for its terms. Putnam's linguistic division of labour allows
for this possibility.3
Once we admit a requirement for the realist's notion of truth to
account for the local case, we should not think that the nonrealist
can maintain his ground on the global level. Once we admit that
maximally coherent beliefs can be false for failing to correspond,
once coherence fails to entail truth, the entailment cannot hold for
global theories either. The realist can once more infer to the
possibility of global error by generalizing the conceivable diver-
gence of coherence from truth shown on the local level. If truth is
verification-transcendent correspondence, then there appears to be
no priori reason why our maximally coherent sets of beliefs cannot
I Putnam, 'Realism and Reason', pp. I26-I27; see also his 'Models and Reality',
Journal of Symbolic Logic, 45 (I980), pp. 464-482.
2 G. H. Merill, 'The Model-Theoretic Argument Against Realism', Philosophy
of Science, 47 (i98o), pp. 69-8I, argued the other side of this coin-that the
world as a structured domain rather than a set of discrete objects may realize
properties and relations that fail to be isomorphic to models of operationally
ideal theories, despite the possibility of one-to-one mappings of objects in the
models onto objects in the world.
3 Putnam, 'Language and Reality', in Mind, Language and Reality (Cambridge:
Cambridge University Press, I 975), p. 274.
fail to be true. Putnam has argued more recently that any set of
operationally acceptable beliefs can be assigned referents so as to
come out true.' In fact any set of beliefs as a consistent formal
structure has a model that can be mapped onto different sets of
referents so as to make the beliefs come out true. All this shows,
however, is that referents for (some) terms in natural languages
must be fixed independently of the formal structures of the
sentences into which they enter.2
The possibility of global falsity can lead us to reinterpret the
global myths again. In the brains in vats case there appeared to be an
option to translate their beliefs as true of phenomenal objects rather
than as false of real objects. But their realist beliefs, regarding un-
perceived objects for example, remained problematic, the Millian
interpretation being stretched and unintended. Maximal internal
coherence of their beliefs, including these anti-phenomenalist
beliefs, would seem to call for counting the majority of them as false
(or as lacking truth value in lacking referents). We can recognize
that maximal coherence calls for assignment of falsity rather than
truth in this case once we are forced by the local case to distinguish
truth from verification. Regarding Putnam's argument that the
brains in vats could not refer to real objects (hence could not believe
they were real brains in vats) because they do not causally inter-
act with real objects, this rests upon a strongly causal theory of
reference whose implausibility is evident from this very example. I
have elsewhere suggested an alternative theory of reference accord-
ing to which we collectively refer to those entities that satisfy core
descriptions from which we would not retreat in different epistemic
conditions.3 The implausibility of the strong causal requirement is
evident here in the implication that language users collectively
cannot refer to what they can describe in detail, even if they intend
to refer to whatever satisfies some of those descriptions.
Once coherence is distinguished from truth, we need not seek to
maximize ascription of true beliefs in ascribing propositional
attitudes to others and referents to their terms. What we are really
after in ascribing propositional attitudes when seeking to under-
stand others is to render their lingauistic behaviour explicable. This
I See 'Models and Reality', ibid.
2 Compare Alvin Plantinga, 'How To Be an Anti-Realist', Proceedings of the
American Philosophical Association, 56 (i982), pp. 47-70.
3 Alan Goldman, 'Reference and Linguistic Authority', Southern 3ournal of
Philosophy, 17
pp. 305-32I.
goal may lead to assignment of mostly true beliefs, and will do so
when we detect nothing out of the ordinary in regard to the
behaviour of the language users in the circumstances. When instead
we have an explanation for their having been misled, we will ascribe
false, but rational and coherent beliefs, as in the brains in vats case.
At the extreme, when behaviour strikes us as incoherent, we will
seek a more direct causal explanation that eschews appeal to
rationality. We may occasionally seek to explain even linguistic
behaviour in this way, when it appears bizarre. Thus, not only may
we conceive of globally false conceptual schemes, we may also
imagine circumstances in which we would ascribe them to other
language users. Understanding others does not require even
agreement with them, let alone truth in the stronger sense. And
given the realist's notion of truth, we can picture as well distortion
in the sources of our beliefs that would render them globally false.
Thus the local case that forces this notion upon us, distinguishing
truth from coherence, appears to defeat the global anti-realist
arguments as well.
The realist's point could perhaps be made less fancifully in
relation to ordinary false statements about the past for which there
is present evidence and no counterevidence. Dummett himself
discusses the counterintuitiveness of the nonrealist's seeming
violation of the truth value link between present tense statements
and corresponding past tense statements uttered at later times when
the evidence has changed. If a present tense statement is false when
uttered, it seems that the later past tense counterpart must be false,
whatever the evidence for it at that time. This link holds not only for
replicas' statements and beliefs, but for our own. Dummett
considers the realist thrust of this point in 'The Reality of the Past'.
The nonrealist, according to Dummett, can accept the truth value
link for any present statement when he contemplates its future
semantic status. At the same time he will maintain that the meaning
of any statement when uttered (or, perhaps better stated, the
assertion made by uttering any sentence) will be a function of the
evidence for its truth at that time.' For the realist, but not for
Dummett, the latter claim will nullify the force of the truth value
connection over time. My motive in making the realist's point by
use of the replica example was to illustrate vividly that the meaning
and truth of a statement at a time is neither a function solely of the
Dummett, 'The Reality of the Past', in Truth and Other Enigmas, p. 373.
evidence for it at that time, nor even of a diachronically broader
notion of coherence. The virtue of the example in this respect is that
it can rule out the possibility of defeating (or in this case saving)
evidence in the past or future. The distinction between truth and
coherence even over time is thereby forced upon us, then to be
8. It remains to indicate more specifically responses to Dummett's
antirealist arguments outlined in section 3. One of his questions,
how we can extend our notion of truth beyond that acquired in
learning contexts in which language use and beliefs correct in
relation to evidence, we have now answered. The task is ac-
complished by extending our notion of error to that of locally
coherent but false belief, and then to the conceivability of globally
false belief. But if these realist notions of truth, meaning and
understanding involve reference to verification transcendent states
of affairs, Dummett's second question appears to become pressing,
namely how we can succeed in making such reference. In
Dummett's terms, what can constitute knowledge of such truth
conditions, comprising our understanding of assertions for the
realist? If we can conceive of states of affairs without conceiving of
their verifying conditions, then we can also conceive how our terms
can refer to such states of affairs. We have only then to think of
reference as a relation of actual satisfaction. But if we- cannot so
conceive of states of affairs, if we cannot know what particular truth
conditions would be apart from knowing what would verify them,
then the realist's notion of understanding collapses.
The answer to this problem is that, despite Dummett's argu-
ments to the contrary, such knowledge can be expressed by the
relevant Tarskian sentences (suitably strengthened to guarantee
translation).1 Dummett points out that one must understand the
designated sentence in a Tarskian sentence before one can under-
stand the whole. We therefore cannot impart the meaning of the
designated sentence by using this device. But this point appears
relevant only if we fail to draw the distinction between use of
I As pointed out in section 2, even a suitably strengthened Tarskian truth theory
can only approximate to a theory of meaning, since it ignores the vagueness that
afflicts most predicates. But the ordinary person's understanding of most terms
will not take account of this phenomenon, and so Tarskian sentences can
capture his understanding of expressions using those terms. I ignore here
complications aimed at capturing a philosophically more sophisticated under-
Tarskian sentences to express what a language user knows when he
understands meanings, and use of them to teach meanings.' That
they are unsuitable for the latter use does not imply that they cannot
serve the former. Since the sentence inserted in the right side of the
Tarskian schema is used rather then designated or mentioned, it
does not matter whether it is formed by disquotation. The resulting
Tarskian sentence can express what a language user knows when he
understands the designated sentence, since the former does not
simply state that he knows the sentence in question, but rather its
truth condition. Meanings captured by the Tarskian schema can be
taught originally in relation to verification conditions (or by using
terms learned originally in that way) and then come to be under-
stood in terms of verification transcendent truth conditions, as
the concept of error becomes distinguished from that of lack of
coherence, and then generalized.
Dummett argues further that we cannot use Tarskian sentences
to explain both meaning and truth without circularity. If we explain
truth by noting that ascribing truth to a statement is the same as
asserting the statement, then we cannot explain meaning in terms of
truth.2 Rather than stating the meaning of an assertion, such an
explanation will accomplish no more than simply restating the
assertion itself. This argument is more subtle and may appear more
to the point; but all it shows, I believe, is that knowledge of truth
conditions must sometimes be direct. It does not show that truth and
meaning are not intimately connected, only that not all semantic
notions can be taught purely linguistically without circularity. Any
successful set of complete definitions can be challenged on the same
ground, since all such sets will be circular (or incomplete in
allowing primitives). This is obvious in regard to meaning apart
from this argument about the explanation of truth. Some state-
ments must be taken initially to be directly verifiable, although their
truth too can be questioned later, once the distinction has been
drawn between truth and coherence. Again the requirement relates
to teaching and learning meanings, not to the knowledge that
constitutes understanding them.
The only remaining question from Dummett is how the realist's
notions are manifest in practice. Arguments evidencing a logic that
presupposes the principle of bivalence do not provide the answer.
I Compare John McDowell, 'Truth Conditions, Bivalence, and Verifi-
cationism,' in Evans and McDowell, ibid., pp. 55-56.
2 Dummett, Truth and Other Enigmas, p. 7.
We saw that the principle fails to hold for certain ranges of
predicates (that rarely affect the validity of everyday arguments);
also that it is neither necessary nor sufficient for semantic realism.
The distinction between truth and warranted assertibility is itself
implicit in certain linguistic practices, however. It shows up first in
the way referents are assigned and understood. We may take
ourselves to be speaking of the same unobservable objects, whether
gods or quarks, even if we disagree about what would verify or
falsify their existence or presence. Second, we can accept evidence,
even complete evidence as in the replica case, without taking it to
entail truth. Realism manifests itself in a willingness to acknowl-
edge sceptical possibilities, in a healthy sense of our own fallibility
and limited place and point of view in the world. Dummett
might argue that such practices indicate our general intention
to refer transcendently, but that they cannot specify particular
transcendent objects or states of affairs which we can understand
to be referents. This objection is overcome by construing reference
as actual satisfaction (of certain privileged associated descriptions)'
and construing understanding in the way indicated earlier in this
A more interesting question for the realist who admits that, from
our ordinary nontranscendent point of view, coherence must
remain our criterion of truth is how he can link the criterion to the
concept itself. One way to make this connection is to take an
evolutionary view of belief acquisition, according to which our
beliefs become maximally coherent through a process of accom-
modation to reality as encountered in experience. We cannot
explore this possibility further here, since we shift at this point from
the question of the intelligibility of realism to that of its truth. The
issue here has been preliminary, concerned with the coherence of
our linguistic intentions. Here the fanciful cases suffice to show
that, even in flights of fancy, we are all at heart semantic realists.
FLORIDA 33124,
See 'Reference and Linguistic Authority', ibid.