Facts and the Function of Truth

Huw Price

Full text of First Edition (with minor typographical corrections)

1
Preface
Some years ago I worked on the philosophy of probability. For a long time I thought about
probabilities – what kinds of things they could be, and so on. Then it struck me that this might be the
wrong approach. We didn’t have to assume that there must be something that probabilistic
judgements are characteristically about, and so concern ourselves with the nature of that distinctive
subject matter. Instead we could interpret probabilistic judgement in terms of a characteristic way of
talking and thinking about other things. At the time it seemed quite a revelation. Of course I soon
realized that others had expressed the same view about probability, as about several other topics. But
in the case of probability, at least, the view was not well developed. I went on to write a thesis (Price
1981) that argued its advantages over other views of probability, and defended it against certain
objections raised against similar ‘non-factualist’ views in other areas.

By contemporary standards the unusual feature of this view of probability is that it takes a linguistic
approach to metaphysical issues. Non-factualist theories were a common and characteristic product of
the mid-century linguistic movement in philosophy; but these days that movement is widely felt to be
discredited. Contemporary metaphysics regards itself as free of the peculiar concern with language
that characterized that period. So as I eventually came to appreciate, a defence of such a treatment of
probability (or indeed other topics) calls for some defence of the linguistic stance. More importantly,
it calls for an elucidation of the cluster of notions whose proper application non-factualists are
concerned to dispute – notions such as assertion and statement of fact. In this respect, as in others, the
linguistic approach needs to be grounded in the philosophy of language.

I found that such a grounding had not in practice been obtained. This was not simply because it had
usually not been sought. Contemporary philosophy of language has appeared to have reason to
dismiss non-factualism – a reason that turns on the role of the notion of truth in Fregean linguistic
theory. In effect, non-factualism and contemporary philosophy of language have wanted to draw the
limits of fact-stating discourse at different places. This conflict has not had the attention that it
deserves, I think. The linguistic approach to difficult areas of philosophy has more to be said for it
than its reputation suggests. And the philosophy of language has much more to learn from the
encounter than the history would indicate.

This book is in part an attempt at mediation. I try to show that with a proper understanding of what an
account of statement of fact should amount to the conflict can be resolved. Neither side escapes
unscathed. The resolution turns, in particular, on the rejection of a shared assumption about the
appropriate goal of an elucidation of the notion of statement. But both sides benefit from the new
perspective thus required. The major benefit is a particularly fruitful perspective on the nature and
origins of truth.

I appreciate that the approach may seem misguided. Many philosophers will feel that the conflict
described is long since decided in favour of the philosophy of language. I suspect that others, on both
sides of the fence, will not have noticed that it ever existed. I can only urge such people to read the
book – or at least to read chapter 1, which contains a more detailed guide to the nature and origins of
the conflict, and the structure and benefits of the resolution I propose.

In order to make it a little easier to read the book, I have starred a number of sections of a more
technical or incidental nature. These may be omitted without serious discontinuity. Chapter 1
mentions a further possibility, which involves beginning at chapter 6.

The book was written during my tenure of an Australian National Research Fellowship, in the School
of Philosophy at the University of New South Wales. The Editor of Mind has kindly allowed me to
reproduce parts of ‘Truth and the nature of assertion’ (Price 1987), which appear mainly in chapter 2.
I am also indebted to Philip Pettit, whose assistance and encouragement long since transcended the

2
merely editorial; and to Peter Menzies and Phillip Staines, for their insight and patience in many
conversations on this material.

3
1
Introduction
1.1 WHY READ THIS BOOK?

Few areas of contemporary philosophy have not been touched by the suggestion that their distinctive
problems are in part a product of some sort of misconception about language. One view of this kind
has been particularly widespread. It is the claim that in virtue of their grammatical form, certain uses
of language are mistakenly construed as descriptive, or fact-stating. Perhaps the most familiar cases
arise in ethics. Moral philosophers have often argued that despite the syntactical resemblance that
ethical utterances bear to genuinely descriptive uses of language, such expressions should not be
interpreted as statements of fact. There are actually several versions of this view, differing as to what
they take it that ethical judgements really are. One at least of these is common currency among non-
philosophers. Known to philosophers as ‘emotivism’, this version holds that moral claims are ‘value
judgements’ – expressions of attitudes of approval, disapproval, or something of the kind. The Prime
Minister might say for example that the Soviets have the best possible system of government, with the
exception of all the rest. An emotivist would interpret the remark as an expression of the PM’s own
political preferences, of her marked antipathy to the Soviet state, rather than as a factual claim in
comparative political theory.

We shall need a name for the general philosophical strategy that emotivism here exemplifies. I shall
call it ‘non-factualism’. ‘Anti- realism’ would be a little less ugly, and is perhaps more widely used in
this context. However, it is widely used for other programmes as well. ‘Non -cognitivism’ suffers
from the opposite problem, being largely confined to the ethical case. A non-factualist thus holds that
there are certain utterances whose grammatical form is that of statements of fact, but whose linguistic
role is actually quite different.

Further philosophical consequences will usually be held to flow from this linguistic insight. If nothing
else, it allows the non-factualist to dismiss previous philosophical concerns about the nature of the
‘facts’ in question. An emotivist will say for example that the traditional philosophical problem of the
nature of good and evil rests on a misconception about the language of morality. As I said, ethics is by
no means the only branch of philosophy whose practitioners have been attracted to non-factualism.
Aesthetic judgements, mathematical statements, theoretical sentences in science, Lockean secondary
qualities, knowledge claims, meaning ascriptions, indicative and subjunctive conditionals, and
probabilistic and modal judgements: all of these topics have been given non-factualist treatments, and
readers may well be able to add to the list.

This book is not about any of these topics as such. Its initial concern lies rather at the next level of
generality, with the foundations and hence the viability of non-factualism itself. In particular it is
concerned with the nature of the distinction between factual and non-factual uses of language. I am
going to argue that the view of this distinction normally taken for granted by non-factualists is
radically mistaken. Non-factualists have taken the issue to concern the distribution in language of a
property that we may call ‘statementhood’. Their claim is then that this property is less extensive than
we ordinarily assume. Certain classes of utterance – moral judgements, or whatever – appear to
possess it but actually lack it. Once the issue is couched in these terms the important questions
become ‘What does the property of statementhood consist in?’ and ‘How do we determine that a
particular kind of discourse possesses or lacks the property so construed?’ In other words, non-
factualism then calls for an analysis of statement of fact – preferably, of course, an analysis that
serves to support the view that there are fewer statements in language than we initially suppose.

4
The main task of the first part of the book will be to challenge this analytic perspective. I shall argue
that of the available ways of cashing out the notion of statement of fact, none provides what the non-
factualist evidently requires: a means to exclude the possibility that there are actually more statements
than the non-factualist holds there to be. Indeed, we shall see that none of the available analyses can
exclude the possibility that all linguistic utterances are fact-stating.

More in a moment on the structure of this argument. I want to make it clear first of all that the book is
not against non-factualism as such. On the contrary, I am going to defend a version of non-
factualism, albeit an unorthodox one, and argue that it vindicates the non-factualists’ concern with
language as a route to enlightenment on a range of philosophical topics. At the very least it vindicates
the intuition that there is something unusual about the language of morality, probability, conditionals
and a range of other topics with which non-factualists have been concerned.

In particular, therefore, the book is not aligned with the influential view that in virtue of some rather
simplistic mistakes about language, non-factualism itself is misconceived. As I shall explain, this
view stems particularly from the Fregean tradition in modem philosophy of language. Like non-
factualism itself, this tradition is committed, by and large, to the existence in language of a distinction
between fact-stating or assertoric discourse and discourse of other kinds. The conflict arises as to
where this distinction should be drawn. A non-factualist wants to narrow the class of genuine
statements. A Fregean, on the other hand, is inclined to do exactly the opposite – to construe the
bounds of assertoric discourse as widely as possible. We shall see that this inclination is grounded
particularly in the role that truth plays in Fregean linguistic theory. Its effect is that non-factualism,
the epitome of philosophy in the linguistic style, comes ironically to be rejected by the philosophy of
language itself.

In criticizing the usual presentation of non-factualism, how then do I fail to align myself with its
Fregean critics? By rejecting the principle that non-factualists and Fregeans have in common, namely
that there is a real property of statementhood, or some such, whose distribution in language can be a
matter for dispute. As I said, I argue in part I that however we try to characterize this supposed
property, we are left with no basis on which to show that it applies to some utterances but not to
others. In particular, we cannot exclude the possibility that all utterances are statements: an admission
that would be as unacceptable to (most) Fregeans as it would be to non-factualists.

I shall describe the conflict between non-factualism and Fregean semantics in more detail in the next
two sections. Its effect, I think, has been to stunt the development of an adequate theory of the
linguistic notions on which non-factualism depends. Both sides suffer as a result. For its part, non-
factualism has not died out. It remains popular in traditional strongholds, such as ethics and aesthetics,
and it continues to win converts elsewhere. As always, however, it lacks adequate conceptual
foundations. On the other side, I think that contemporary philosophy of language has suffered from a
failure to take seriously the non-factualist's concerns. The notions non-factualism appeals to –
factuality, statement, truth, and so on – are central to an understanding of language. By focusing on
some of the hard cases, non-factualism offers a particularly fruitful perspective on these notions. To
dismiss non-factualism is to miss this perspective.

The book thus approaches the philosophy of language by way of linguistic philosophy. The needs of
non-factualism, as viewed in the light of part I, lead in part II to a novel account of truth. One of the
advantages of this account is its ability to deal with these difficult cases, but its interest and
application is more general. Its novelty, by contemporary standards, lies in the fact that it is
explanatory rather than analytic. By this I mean that the question it addresses is not so much ‘What is
truth?’ as ‘What use is the concept of truth, to creatures like us?’ In other words, it seeks to explain
our possession of the notion of truth in terms of its function in a linguistic community. It is a
genealogical theory. It asks why we should ever have developed such a notion.

5
I shall have more to say later on about the distinction between analytic and explanatory theories. In
case naturalists in the audience are already tempted to leave, however, I emphasize now that the
proposed account will be thoroughly naturalistic. For any concept in common use there are (at least)
two possible naturalistic accounts of how it got there. One approach is to analyse the concept
concerned in naturalistic terms, and hence explain its use in terms of some general account of our
reaction to our natural environment. But another – often overlooked, I think – is to find some non-
referential function for the concept in the lives of natural creatures such as ourselves. The latter
approach is most clearly appropriate in cases in which we have reason to deny referents to the
concepts concerned – religious concepts, or those of certain former scientific theories, for example. It
is not excluded in other cases, however, and the proposed account of truth will be of this kind.

In view of the wide importance of the notion of truth in contemporary philosophy, this may seem a
case in which a solution outgrows the problem from which it stems. Some readers will inevitably feel
that in grounding such an account of truth on the needs of non-factualism, I use a small tail to wag a
very large dog. Others, more favourably disposed to the general approach of part II, will feel that the
dog is quite capable of wagging itself.

To the former objection I would reply that if the argument of part I goes through then it is not simply
non-factualists who need a new approach to truth and related notions. Their Fregean opponents should
be equally concerned by the failure of the usual ways of construing assertoric discourse. That said,
non-factualism provides a valuable guide to and constraint on alternative approaches. One of the
advantages of the account of truth proposed in part II is that it does make very good sense of non-
factualist intuitions. Moreover, non-factualist concerns are a powerful antidote to another non-
analytic approach to truth which has had significant support in recent years. This approach reacts to
the perceived failure of attempts to analyse truth by dismissing the problem, recommending instead
that we change the subject and stop worrying about truth. In drawing our attention to the limits of
truth, non-factualism does much to counter this unhealthy temptation to let a sleeping dog lie.

I have more sympathy with the latter objection – the view that the proposed account of truth might be
motivated without reference to the needs of non-factualism. I would simply encourage the reader
whose interest is more in the dog than the tail to feel free to begin at part II. Chapter 6, sections 1 to 5
of chapter 7 and chapter 9 deal largely with the general account of truth. They make little reference to
non-factualism and the concerns of part I, and might be read independently. The remainder of chapter
7 and chapter 8 apply the general account to the non-factualist cases, and might be read next. Part I
would then recommend itself to the reader who found the orientation implausible, but was curious to
see where I had gone astray. This approach to the book aside, I have starred a number of sections to
indicate that they may be omitted without serious discontinuity.

Who then might profitably read this book? First, I think, anyone with an interest in non -factualism in
any of the areas in which it has been app lied. I imagine that most such readers will take it that they
have a reasonably clear conception of what distinguishes fact-stating discourse, and therefore as to
why a non-factualist interpretation might be judged appropriate in any particular case. Part I aims to
shake such confidence, and part II to show what might replace it. Secondly, I recommend the book to
those with an active disinterest in non-factualism – those in particular who take its misconceptions to
have been exposed by contemporary philosophy of language. Here too part I aims to undermine such
prejudices. And thirdly and no doubt most numerously, I recommend it to anyone with an interest in
truth. The account of truth here offered is novel and potentially important. The route will be
unfamiliar to most readers, but I think the result will repay their efforts.

The remainder of this chapter is in four sections. The next, section 1.2, outlines the historical
background to the conflict between non-factualism and contemporary philosophy of language. In
section 1.3 I explain how the conflict can be thought of as a disagreement concerning the extension of
the property of statementhood in language, and therefore how I propose to attack the common
assumption that there is such a property. Section 1.4 is then a brief guide to the argument of part I,

6
and section 1.5 some remarks about orientation of the alternative programme to be developed in part
II.

1.2 NON-FACTUALISM AND FREGEAN SEMANTICS

With the benefit of hindsight, at any rate, it is clear that non- factualism predates the linguistic
movement of the mid-twentieth century. In ethics it traces its origins at least to Hume, who may be
regarded as an early emotivist. Hume seems to have taken a similar view of causal judgements:
namely that they do not refer to a realm of causal facts, but rather express our expectations, formed on
the basis of observed regularities. But the great popularity of non-factualism in this century – indeed
the recognition that Hume should be seen in these terms – seems largely associated with the linguistic
tum. In those halcyon days after the war the non-factualist techniques were applied to a wider and
wider range of traditional philosophical topics.

In the 1960s, however, non-factualism acquired a reputation as a cheap approach to subtle

philosophical problems: too quick to claim new territory, and too slow to justify its existing claims.
The reputation is not entirely undeserved. The non-factualists of the 1950s do seem to have paid too
little attention to the foundations of their own position. But it also owes something to the fact that the
approach had the misfortune to coincide with what, from the non-factualist's point of view, was a
particularly inhospitable period in the philosophical study of language itself. For non-factualism is at
odds with a view that at this period philosophers of language had increasing reason to uphold – the
view that assertion or statement of fact is the central and primary use of language, rather than one use
among others of equal significance.

At first sight, indeed, a reader might gain the impression that there is no place for the notion of a non-
fact-stating use of language in contemporary philosophy of language. Writers will refer, for example,
to the task of specifying the truth conditions of ‘the sentences of a language’, as if there were no such
thing as a sentence without truth conditions – as an utterance whose point is not to state that certain
conditions obtain. As it stands, however, this may be said to reflect a benign idealization, not
inconsistent with a more pluralistic view of natural language. The real basis of the doctrine of the
primacy of assertion lies deeper, I think, in the Fregean use of the notion of truth.

Once truth is part of a philosopher’s repertoire, there are at least two routes to the primacy of
assertion. One turns on the fact that a systematic approach to the theory and taxonomy of language
will need to distinguish the ‘genuine’ uses of languages, to which the theory should be expected to
apply, from various sorts of pseudo-linguistic noise – for example the kind of noises we make for dog
rousing, baby settling and the like. A natural suggestion is that the genuine uses of language are the
sentential utterances. (Thus Davidson 1984, p. 60: ‘In defining sentencehood what we capture,
roughly, is the idea of an independently meaningful expression.’) In order to apply this suggestion we
have to know what to count as a sentence. If the notion of truth is already at hand, it may suggest the
following solution: a sentence is the bearer of a truth value. The significant utterances are those that
are capable of being true or false: in other words (or so it seems), the assertions, or statements.

In practice, this route to the primacy of assertion is overshadowed by another. The notion of truth not
only promises to delineate the raw material for a systematic theory of language; it also offers an
approach to problem of meaning – an approach developed since the 1960s in the influential work of
Davidson and his followers. (See particularly Davidson 1984, essays 1 to 5; and Evans and McDowell
(eds) 1976.) Davidson proposes that a systematic theory of meaning for a language – a theory that
relates meaning to syntactic structure, in such a way as to show how the meanings of complex
sentences depend on those of their constituents – should be a systematic theory of the truth conditions
of the sentences of the language concerned. He suggests also that this truth theory be modelled on
Tarski's definition of truth for formal languages. Such a theory is bound to accord a central place to
those linguistic expressions that can be said to have truth values (and hence truth conditions). At first
sight, indeed, there seems to be room for nothing else.

7
It may seem then that whichever route it takes to the doctrine that assertion is the primary use of
language, a Fregean theory of language should be embarrassed by the existence of utterances such as
questions and commands. These would appear to be complete, independent, meaningful uses of
language, and yet not to be assertions – to lack truth values, for example. However, there are at least
two ways in which a Fregean theory might deal with these cases. One way, obviously, would be to
claim that questions, commands and the like do have truth values after all. Questions and commands
would thus be construed as special kinds of assertions – perhaps as reflexive assertions about their
speakers. As we shall see in chapter 3, this option has been explored by some of the proponents of the
truth-conditional approach to general semantics.

On the whole, however, advocates of Fregean general semantics have preferred a less radical solution,
turning on the Fregean distinction between sense and force. Roughly, the idea is to confine the
significance of the notion of a truth condition to the determination of a certain core factor in the
meaning of an utterance (its sense); and to regard the complete meaning of an utterance as a product
of this factor and another (its force). Thus, the meaning of the question ‘Has the dog been fed?’ will
be explained as a product of the interrogative force (here indicated by the interrogative mood) and of
a sense that the sentence shares with its assertoric transform, ‘The dog has been fed.’

This solution enables a Fregean theory of meaning to accommodate non-assertoric utterances. Indeed,
in principle it allows for the assertoric mode of utterance to be treated as simply one among many.
Each mode calls for an account of its peculiar pragmatic significance. The assertoric mode might be
said to be distinguished by the convention that in using it, speakers aim to utter true sentences; the
imperative, by the convention that speakers utter sentences they require their audience to make true;
and so on. In principle, the list could be very long indeed.

In practice, however, the theorist has a strong incentive to peg the boundaries of the class of assertions
as widely as possible, at the expense of non-assertoric varieties of force – to expand the semantic
component of a theory of meaning at the expense of the pragmatic. Faced with a new linguistic
expression, the theorist has to decide whether its contribution to the meaning of a sentence consists in
a modification of sense or an effect on force. The former answer has the advantage, for a Fregean, that
it brings the case within the scope of the core component of the general theory of meaning. It avoids
the need for a further addition to the supplementary theory of force. The latter answer, on the other
hand, is a threat to the simplicity and comprehensiveness of the Fregean approach.

Most seriously, however, the latter answer would undermine the Fregean’s claim to provide a
systematic account of meaning. As Davidson emphasizes, the great advantage of the truth-theoretic
approach is that it promises to account for the finite-based ‘creativity’ of language use – the ability of
speakers to use and interpret a potentially infinite number of novel utterances. It does so by linking
syntactic structure to conditions for truth. The significance of sentence-forming components is
explained in terms of their contribution to the truth conditions of sentences in which they occur. On
this view then, the open-endedness of language simply mirrors the open-endedness of possible truth
conditions. This elegant solution would be crippled by the admission of a dimension of linguistic
creativity not matched to a range of truth conditions. If sentences expressing moral judgements do not
have truth values, for example, then the infinity of novel utterances we can generate from moral
sentences (using the ordinary logical connectives, say) cannot be matched to a corresponding infinity
of truth conditions; and our capacity to understand such sentences cannot be explained in terms of
such a correspondence.

So if creativity is to be confined to the semantic part of a theory of meaning, non-assertoric

expressions can be tolerated only on the condition that they do not proliferate. Like questions and
commands, non-assertoric expressions must not significantly combine with each other (or indeed with
assertoric sentences) to generate further such expressions. However, this condition clearly excludes
the cases of interest to non-factualists: expressions that behave like fact-stating sentences, at least to
the extent of participating in the usual modes of supra-sentential complexity. Truth-conditional

8
general semantics thus has a powerful incentive to oppose non-factualism in general, and to dispute its
claims in individual cases. The same applies, incidentally, if truth is replaced with a notion such as
‘assertibility’; the resulting theory will have the same reason to confine creativity to what it regards as
the non-pragmatic component of a theory of meaning.

As a distinctive product of the linguistic movement in philosophy, non-factualism was thus

overwhelmed by developments in the philosophy of language itself: by the growing interest in the
project of a general theory of meaning, and the growing popularity of a Fregean approach to that
project. This conflict has not had the attention it deserves, I think. Proponents of truth-conditional
general semantics, and indeed of other Fregean approaches to a general theory of meaning, have
shown a degree of professional blindness to the attractions of non-factualist treatments of particular
topics; while on their side, non-factualists often seem to have been unaware of the problem of
reconciling their theories with a workable approach to general semantics.

There are signs of renewed interest in non-factualism, of a new respectability. At least one source of
this interest lies within the philosophy of language itself: in recent interest in Wittgenstein's later
views on meaning, and in particular in the suggestion (notably by Kripke 1982) that the notion of
meaning itself might require a non-factualist treatment. However, it seems to me that without proper
attention to its foundations, and to its place in a general semantic programme, the approach will prove
no more durable at the end of the century than it was in the middle. This book, then, is in part an
investigation of these foundations, in the light of the conflict just described.

1.3 THE NATURE OF STATEMENT: CONFLICTING CONSTRAINTS AND A SCEPTICAL

PROGRAMME

As I noted earlier, some Fregeans think that all discourse is assertoric (and hence, in effect, that all
language is fact-stating); but most prefer to distinguish assertions from utterances such as questions
and commands. The more common view is thus that the fact-stating/non-fact-stating distinction
roughly corresponds to the English indicative/non-indicative distinction.

A major advantage of this view has been thought to be that it gives us an account of grammatical
mood. Mood is a striking feature of natural languages, whose existence we might expect a theory of
language to be able to explain. It has been widely assumed that an adequate explanation would come
in two parts. The first would identify some property characteristic of just those speech acts for which
we do employ the indicative mood; while the second would consist in the claim that, by convention,
the indicative mood is used to ‘mark’ the performance of a linguistic act with this particular property.
This view of mood – the marker view, as I shall call it – thus takes the indicative mood to be the
conventional sign of a certain property, or factor, in the meaning of an utterance. The following
remarks are representative; they come from a recent text on mood by the linguist F. R. Palmer.

It is undoubtedly the case that most, perhaps all, languages have a clear way of indicating that
the speaker is making a statement that he believes to be true. This is what may be called the
Declarative – the grammatical form that is typically used for such statements … In languages
that have systems of mood, the indicative … typically indicates a declarative. (Palmer 1986,
pp. 26–7)

Where a language has an indicative and an imperative mood, these are the formal
grammatical markers associated with the notions of statement and [command], though there is
no exact one-to-one correspondence. (1986, p. 24)

As Palmer notes, the marker view needs to be qualified, in practice, to cope with various additional
grammatical devices. In English, for example, stress and inflexion enable us to do many things with
indicatives normally done with non-indicatives, and vice versa. However, a proponent of the marker
view will take this to mean that in practice the marks are just more complicated than the initial

9
account recognized. For present purposes we may ignore such refinements. As long as we don 't
criticize the marker view on the grounds that it misses these complexities, we won't go wrong in
criticizing the simpler version.

In drawing the bounds of assertoric discourse at the limits of the indicative mood, most Fregeans thus
align themselves with the marker view; and marker theorists will often characterize their position in
terms of the Fregean force categories. The views thus have a common interest in the elucidation of
these categories; in particular, in an account of assertion in terms of which the status of a given
utterance can be established. For both views want to defend the claim that the non-indicatives are not
assertions; and to be able to show, pace the non-factualist, that all indicatives are assertions.

The marker view will be the second front for the argument of the next four chapters. Our main target
will be non-factualism, as usually presented. However, I want to show that the untenability of this
position can be traced to its dependence on the analytic presupposition – on the view that there is a
real property of statementhood, about whose distribution in language we are liable to be misled. I
think that expressed in other terms, non-factualism contains important insights about language. So I
shall need to exclude an alternative response to the argument – that of rejecting non-factualism, while
retaining its usual analytic footings. The marker view makes this response particularly appealing. In
other words, we have to show that the failure of orthodox non-factualism is not attributable to the
success of the marker view.

Fortunately, the same style of argument works on both fronts. In effect, indeed, we shall be able to
encompass both targets in a single advance, arguing first that the available analyses do not enable the
non-factualist to hold a line short of the marker view itself; and secondly that they do not enable the
marker theorist to hold the line at the indicatives (and so to exclude the possibility that all discourse is
factual discourse ).

I emphasize that the conclusion we are aiming for is not that all discourse is factual discourse; merely
that neither non-factualists nor marker theorists can exclude this possibility. The argument will thus
be a sceptical critique of the analytic assumption these views have in common – a demonstration that
in neither case can analysis serve its intended purpose, in justifying a particular division of language
into fact-stating and non-fact-stating parts.

1.4 GUIDE TO PART I

The structure of the next four chapters is as follows. In view of the importance of the notion of truth
in general semantics and to the proposal of part II, chapter 2 treats non-factualists and marker
theorists in tandem, examining the suggestion that assertions be distinguished in virtue of their
connection with the properties of truth and falsity. Both sides have found this idea appealing. I argue,
however, that without a prior understanding of the nature of assertion, we cannot restrict the notion of
truth in such a way that not every utterance has the appropriate connection with these properties.

With truth out of the way, it is then convenient to separate our targets. Chapter 3 concentrates on the
marker view, arguing by way of reductio that if there were an underlying property of the kind the
view requires – a property marked by the indicative mood – we could not exclude its possession by
non-indicatives. This shows, in effect, that the line cannot be held at the limits of the indicative mood.
If all indicatives are fact-stating, then there is no way to exclude the possibility that non-indicatives
are also fact-stating. The remaining task is then to show that the line cannot be held within the
indicative mood – that of the various possible accounts of the nature of statement, none suffices to
show that some indicatives are not statements of fact.

In chapters 4 and 5, with this end in mind, we take the side of the non-factualist’s opponent in
particular cases. Chapter 4 argues that in two important and arguably representative cases (probability
and moral judgement) such an opponent forces the non-factualist to rely on a psychological

10
characterization of statement of fact – on the claim, essentially, that statement is the expression of
belief. Chapter 5 then examines some possible bases for the required distinction between beliefs and
other propositional attitudes. We find that the attempt to make good this distinction leads the non-
factualist once more to appeal to truth – in effect, to the proposal rejected in chapter 2. The chapter
concludes with a look at some recent arguments that seem to favour non-factualism about conditionals
– confirming that here, too, non-factualism lacks the basis it requires.

1.5 THE EXPLANATORY TURN

One option at this point would be to blame the problem for the failure of the analytic solution – to
conclude that there is simply nothing of philosophical interest to be said about a distinction in
language of the kind non-factualism requires. Such ‘quietism’ has an evident appeal: to dismiss a
problem is to save oneself the trouble of finding a solution. It is the relief the lapsed theist might feel
in being able to put theological problems aside. But of course, this advantage is like the claimed
psychological benefits of religious faith itself. It is a motive for wanting a particular conclusion to be
true, but not a reason for believing it so. In the present case there seem to be two good reasons to
remain at least open-minded. Quietism cannot explain the significance of mood, or the continuing
attraction of non-factualist approaches to particular philosophical topics.

Part II therefore explores a different response. This concentrates on the assumption that non-
factualism needs to be couched as a claim concerning the distribution of a certain property in
language. I think that the prevalence of this assumption among non-factualists is particularly a
product of their native tendency to set aside linguistic appearances. Non-factualists want to say that
appearances are deceptive, that assertions cannot readily be distinguished in virtue of their syntactical
form. This is needed to explain how more traditional philosophical theories can have been led astray.
And there must be something in the picture for these theories to have been wrong about – in other
words, it seems, some underlying property that syntax can mislead us about. Ask what property this
is, and one is committed to what I have called the analytic programme.

The solution, I think, is to recognize that there is room for deceptiveness at the syntactical level itself.
An expression that behaves as an assertion in most respects and circumstances might none the less
behave differently in minor respects or unusual circumstances. If the differences are then
overshadowed by the likenesses, it is easy to see how interpreters might be led astray. With the
deceptiveness of linguistic appearances explained in these terms, non-factualism doesn’t need to
appeal to linguistic categories underlying usage. So the non-factualist no longer owes us an analysis
of statementhood. Non-factualism simply doesn’t need to appeal to any such property. True, it ought
to explain the syntactical anomalies to which it does now appeal – the features of usage that
distinguish moral judgements, probabilistic claims, or whatever. As we shall see, however, these
anomalies can be explained without being taken to reflect any underlying semantic distinction.

What are these anomalies in usage, and what is the standard pattern from which they depart? We shall
see that both involve the notions of truth and falsity. Dealing first with the ordinary use of these
notions, I shall propose that we explain it in functional and evolutionary terms. The main idea will be
that in virtue of the normativity of truth and falsity, speakers are encouraged to resolve disagreements.
This has long term survival advantages. On the whole, it improves the behavioural commitments with
which language users meet the world. As I said earlier, this is not intended to be an analysis of truth.
It is a genealogical theory, an explanation as to why a language community might come to possess
such a notion. It is comparable for example to a biological view of the origins of morality – the view
say that human societies tend to evolve such a value system in virtue of the communal advantages that
flow from its regulative effects.

Returning to non-factualism, we shall see that this account of truth leaves room for certain borderline
cases – cases in which the use of truth and falsity departs from the standard pattern. These variations
turn out to be explicable in similar terms to the standard pattern itself. Moreover, they appear to

11
correlate well with the cases in which non-factualism has seemed attractive. The explanatory
perspective thus provides a sympathetic reconstruction of non-factualist intuitions – while vindicating,
in a sense, the popular conception that truth is at the basis of the fact-stating/non-fact-stating
distinction. It also finds a place for the non-indicatives.

It will be necessary to defend this account against the charge that it merely offers another analysis of
the nature of statement. An obvious source of such a charge will be the fact that the account professes
to explain variations in a pattern of usage across language. Hence it needs apparently to appeal to
some more basic differences between utterances, or possible uses of language. Could not these
differences themselves be made the basis of a characterization of fact-stating discourse? The reply to
this will turn particularly on holistic considerations – the crucial point being that the relevant
underlying difference need not show up in the individual case.

We began with the conflict between non-factualism and the Fregean programme for general semantics
over the scope of the fact-stating/non-fact-stating distinction. In the final chapter I shall indicate how
this conflict is resolved by the proposed treatment of truth. I shall locate the proposed account in
relation to some recent discussions of realism and objectivity. And I shall finish with some comments
on the significance of the explanatory perspective in other areas – particularly its relevance to the
problem of the status of folk psychology.

12
PART I

A Sceptic’s Guide to the Matter of Fact

13
2
The Place of Truth
Truth has often seemed to be the key to an analysis of fact-stating discourse. Marker theorists and
non-factualists have both been tempted by the idea that an utterance is genuinely assertoric or fact-
stating just in case it stands in a particular relation to the properties of truth and falsity – for example,
just in case it bears a truth value, or is capable of doing so. In this chapter I want to show that truth
cannot be invoked in this way.

The argument will turn on the demands placed on truth by the requirement that it provide the base
property for an analysis of the notion of statement of fact. I cannot emphasize too strongly that I am
not suggesting that truth has no role to play in an adequate account of fact-stating discourse. On the
contrary, truth will be central to the theory to be developed in part II. But that theory will not be an
analysis of statement of fact. Instead it will offer a non-reductive explanation of the characteristic
marks in usage of the distinction between fact-stating and non-fact-stating discourse – including,
perhaps most characteristically, the use of the notions of truth and falsity.

The present chapter goes as follows. I begin by clarifying the demands on the notion of truth that flow
from its attempted use by marker theorists and non-factualists to elucidate the notion of assertion.
With reference to familiar theories of truth, I then illustrate some of the ways in which a theory of
truth can fail to meet these demands (fail to provide a substantial theory of truth, as we shall say). I
offer some general arguments against the possibility of a substantial theory of truth, and conclude by
examining a specific proposal: the approach suggested by David Wiggins in a paper called ‘What
would be a substantial theory of truth?’ (1980). Wiggins’s approach – I think the most explicit attempt
in the literature to tailor a theory of truth to these purposes – has several features of interest, from our
present point of view. For one thing it is grounded in the truth-theoretic approach to general
semantics, whose constitutional aversion to non-factualism we noted in chapter 1. Wiggins is aware of
this tension, observing that the notion of truth he derives from truth-theoretic considerations – a
notion applicable to the indicatives as a whole – leaves little scope for the more restricted notion the
non-factualist requires. In showing that such a notion of truth cannot in fact be counted substantial, we
shall defend non-factualism against this objection and clear the ground for its promised reconciliation
with general semantics. For another thing, Wiggins’s theory sets useful targets for any alternative
account of truth. Wiggins identifies properties or ‘marks’ that are plausible characteristics of any
satisfactory account of truth. We shall see that his own derivations of these marks are illicit, given the
intended role of a substantial theory of truth. But the marks themselves will set a standard for our own
explanatory theory of truth. We shall want to be able to explain why truth has these particular
characteristics.

2.1 TWO WAYS OF APPEALING TO TRUTH

In attempting to characterize assertion in terms of truth, marker theorists and non-factualists must
both aim for more than a merely descriptive account of a distinction found in usage. Neither side can
simply observe that in practice we apply the notions of truth and falsity selectively – that only certain
kinds of utterance are treated as capable of possessing these properties. In the marker theorist’s case,
this is because the marker theory itself is an attempt to explain a distinction we observe in usage. It
may be illuminating to describe this distinction further – to note that in practice the use of the
indicative mood correlates with that of ‘true’, ‘false’ and related terms. But to gloss the problem in
this way is not to provide a solution. The ordinary use of truth is as much in need of explanation as the
indicative mood itself.

14
In the non-factualist’s case a merely descriptive characterization would be inadequate for a different
reason. The non-factualist’s interest in usage is not so much to explain it as to escape it. Non-
factualism urges us to draw distinctions that are missed by ordinary usage. Its characteristic claim is
that we are misled by usage into mistaking the real nature of a certain kind of discourse (typically
with philosophical consequences). True, the non-factualist might say that we are misled only by
insufficient attention to usage – that if we concentrate on the details, we do find evidence of the
distinctions in question. But evidence of what? Even in this case the non-factualist assumes that the
usage distinctions in question reflect the deeper distinctions on which the philosophical moral
depends. The point is made by Michael Dummett. In his (1958) paper on truth Dummett considers the
suggestion that an appeal to the use of ‘true’ and ‘false’ can settle non-factualist claims, for example
about the status of ethical statements. ‘At one time’, he says,

it was usual to say that we do not call ethical statements ‘true’ or ‘false’, and from this many
consequences for ethics were held to flow. But the question is not whether these words are in
practice applied to ethical statements, but whether, if they were so applied, the point of doing
so would be the same as the point of applying them to statements of other kinds, and, if not, in
what ways it would be different. (1958, p. 144)

Non-factualists and marker theorists should therefore not be construed, in trying to characterize
assertion in terms of truth, as merely labelling a distinction they claim to find in ordinary usage. Their
appeal is not to the ordinary use of ‘true’, ‘false’ and related terms, but to the properties of truth and
falsity. Their common claim is that an assertion (or statement of fact, or whatever) is an utterance that
stands to these properties in some distinctive way.

At this point the possible views begin to diverge, in two senses. First, the non-factualist and the
marker theorist clearly apply this characterization in different ways. The marker theorist takes
standing in the specified way to truth and falsity to be the property conventionally marked by the
indicative mood (and as we shall see, offers parallel accounts of the other moods); whereas the non-
factualist claims that not all indicatives stand in this way to truth and falsity. Secondly, there are a
number of views of the precise role of the properties of truth and falsity in a characterization of this
general kind. Is it that assertions have these properties, that they are capable of having them, that they
are governed by the convention that we try to ensure that they have one of them (i.e. truth), or some
yet further connection?

These various elaborations have this much in common, however: they all depend on truth and falsity,
and hence on the availability, at least in principle, of some account of the nature of these properties.
The point is a trivial one. Unless the property of truth is subject to some theoretical constraints there is
nothing to exclude its application not merely to the utterances the non-factualist or marker theorist
wants to regard as non-fact-stating, but to any object whatsoever. Without a theory of truth, it is not
simply that we cannot exclude the possibility that imperatives or ethical judgements have truth values
(or stand in some other important relation to truth); we cannot exclude the corresponding possibility
for trees, rocks and clusters of galaxies. (It is no good appealing to usage. We have seen that non-
factualists and marker theorists both need more than that.)

Of course, not just any theory of truth will do. Again, one only needs to consider some trivial
possibilities. Let us say that if a theory of truth can be invoked to characterize fact-stating discourse in
the way that either non-factualists or marker theorists require, then it is a substantial theory of truth. (I
think the term was first used in this way by Bernard Williams 1973, p. 202). I want to show that there
can be no such theory.

2.2 TWO WAYS FOR TRUTH TO BE INSUBSTANTIAL

One of the most explicit versions of a truth-based formulation of the marker view is that of Stenius,
who says that ‘we may ... formulate the rule for the indicative in the following way: Produce a

15
sentence in the indicative mood only if its sentence-radical is true (1967, pp. 267–8). By ‘sentence -
radical’ Stenius, following Wittgenstein, means a sentence considered as devoid of assertoric force.
Thus:

Consider the following sentences:

(1) You live here now.

(2) Live (you) here now!
(3) Do you live here now?

These three sentences have something in common, which ... I call the sentence-radical; what
is different in each of them I call the modal element. The sentence-radical signifies the
descriptive content of the sentence, the modal element signifies its mood. (1967, p. 254)

Stenius suggests the following rules for the use of the imperative and interrogative moods: ‘React to a
sentence in the imperative mood by making the sentence-radical true’, and ‘Answer the question by
“yes” or “no”, according as its sentence-radical is true or false.’

Let us focus on the possible justification for a theory of this kind. If we want to appeal to a semantic
property to explain the significance of the indicative mood, we need to be sure that we can tell
whether an utterance has that property, without appealing to its grammatical mood. A theory such as
Stenius’s cannot simply take for granted that certain sentence-forming expressions are mood
indicators, and others are modifiers of descriptive content. Rather it needs to justify the particular
dividing line it claims to find in language.

This point is liable to be overlooked, if – in the grip of the orthodox view – we fail to notice that there
are alternatives. But consider negation, for example. Why not think of ‘It is not the case that ...’ as a
mood indicator? It will be governed, of course, by this rule: Produce a sentence in the negative mood
only if its sentence radical is false. Or on the other side, why not think of the imperative as a content
modifier (so that imperative utterances have assertoric force)? We cannot rule out these possibilities
on the grounds that negatives but not imperatives are indicatives, if what we want to explain is the
significance of the indicative construction itself.

These considerations place the following constraint on a substantial theory of truth. It must not take
for granted that only indicative utterances have truth values, or stand in the requisite relation to the
properties of truth and falsity. Yet at the same time it must exclude the extension of these properties to
non-indicatives. (The non-factualist will want the bounds to be narrower still.) We shall see that there
is a tendency to avoid the latter mistake at the cost of the former.

The second major constraint on a substantial theory of truth is merely an instance of a general
requirement for acceptable analysis: an account of a base property cannot itself depend on the very
notion that property is intended to elucidate. A substantial theory of truth cannot rely on the notion of
assertion, for example. In particular, it cannot rely on the otherwise appealing intuition that a sentence
is true just in case it would be correct to assert it. For that would reduce the supposedly distinctive
claim that assertions are governed by the convention that speakers aim to speak the truth, to the truism
that assertions are governed by the convention that speakers aim to make them correctly. Any
convention-governed activity is governed by that convention.

Stenius is bothered by this sort of point, and suggests that

there are two concepts of truth: (i) the concept of truth ... which refers to the sentence-radical
– I shall call this kind of truth descriptive truth; (ii) a concept of truth referring to a sentence
in the indicative – I shall call this kind of truth modal truth.

16
 When stating that somebody ‘speaks the truth’ one does not mean that he is producing a true
sentence-radical – for one does not say that somebody is speaking the truth if he presents a
true sentence-radical in the imperative or interrogative mood .... The expression ‘speaking the
truth’ thus refers to the modal concept of truth, and means that one is following the rule for
the indicative correctly. This moral rule is often motivated by the argument that by not
following it one is undermining the communicative function of language, which morals must
safeguard. (1967, p. 268)

As it happens, of course, to use the indicative correctly is to utter a sentence radical which is true in
the descriptive sense; and so the two kinds of truth seem co-extensive. On Stenius’s view this seems
to be a kind of accident, stemming from the rule which happens to govern the indicative mood. But
there is a danger here: the more the modal notion of truth is detached from the actual use of the
indicative mood, the more it needs to be explained why the modal notion should not extend to other
moods. Why, for example, should we not say that someone ‘acts the truth’ when he or she obeys a
command, thus following the rule for the imperative correctly? The modal notion of truth thus floats
free of the indicative mood, while we are still in need of a substantial theory of the descriptive notion
of truth.

In summary then, there are two main ways in which a theory of truth can fail to be substantial. It can
‘get the grammar wrong’, either illicitly assuming that only indicatives stand appropriately to truth or
failing to exclude the possibility that non-indicatives so stand. Or it can appeal to the sorts of notions
it is supposed to elucidate (or, not quite such a serious sin, to notions that would do a better job by
themselves). We shall see that the first disability is endemic to the proposal to analyse assertion in
terms of truth, though it is liable to be disguised by the second.

It is easy to find both kinds of insubstantiality in popular accounts of truth. This is not an objection to
the accounts concerned, of course. It simply means that they cannot be applied, as in general they
were not intended to be applied, in the special role of a substantial theory.

Thus traditional correspondence and coherence accounts of truth are likely to embody the second
form of insubstantiality. To explain truth in terms of ‘correspondence to the facts’ is to take for
granted the very notion we are supposing that the non-factualist wants to elucidate (and so restrict) by
means of a substantial theory of truth. For example, it makes the issue as to whether there are genuine
moral truths – whether moral claims can properly be said to be true or false – depend on that of
whether there are moral facts. If we could already settle the latter issue, we wouldn’t need to
characterize fact-stating discourse in terms of truth. Similarly, it is unlikely to be helpful to try to
explain truth in terms of coherence with a system of beliefs. In the moral case, that leads us back to
the question whether moral attitudes are genuine beliefs. But given the notion of belief, we would be
able to use it to characterize assertion (in terms of expression of belief), and the excursion through
truth would be redundant.

The first kind of insubstantiality is likely to be displayed by any account of truth that trades on the
equivalence property of truth: on the fact that for any (‘appropriate’) sentence P, the sentences P and
‘P is true’ are in some sense equivalent in meaning. The most familiar such accounts are the
‘redundancy’ and ‘disquotational’ theories of truth (Ramsey 1978, chapter 2; Quine, 1970, chapter 1),
for which this property characterizes truth. A less familiar case is Strawson’s (1949) ‘performative’
account of truth. This turns on the application we make of the equivalence property – on the fact that
by saying ‘That’s true’ we can endorse an utterance made by another speaker. The novelty of the
approach lies in taking this endorsing or confirmatory speech function to be the primary characteristic
of truth; or as the view would think more accurate, of the use of the term ‘true’. (This view is also
known, rather aptly, as the ‘amen’ theory of truth.)

The insubstantiality of such theories stems from their reliance on the assumption that the equivalence
property involves a restriction of truth to indicative sentences. It is not in fact the case that for any

17
sentence P, P and ‘P is true’ are equivalent in meaning. Manifestly, ‘Feed the dog’ does not mean the
same as ‘“Feed the dog” is true.’ Hence the restriction to ‘appropriate’ sentences. But which are the
appropriate sentences? Why, those that grammar permits us to substitute for P in the construction ‘It
is true that P’; in other words (for our purposes), the indicatives. So the equivalence property
distinguishes the indicatives only by fiat. A theory of truth which rests on the equivalence property
either fails to pick out the indicatives (or any sub-class thereof) at all, or does so only because the
restriction of truth to such a class of utterances is assumed from the start. But if the latter, then we can
again invoke the sceptical challenge: Why this restriction? Such a notion of truth does not therefore
provide the base property that is supposed to be characteristic of assertion and marked by the
indicative mood.

The fact that Strawson’s theory is insubstantial in this way is noted by Bernard Williams, in the paper
in which he introduces the idea of a substantial theory of truth. Williams says that he doesn’t see ‘how
on such a theory it could be more than an accident of language that “is true” signified agreement with
assertions rather than agreement with anything else’ (1973, p. 203). In practice we cannot endorse an
apt question or an appropriate command by calling it true. Of course, we can do it in less direct ways,
but so we could in all the cases in which we can just say ‘That’s true.’ The amen theory does not
explain why we should be allowed the conversational convenience in some cases but not in others.

The above examples suggest that a substantial theory of truth would be an unusual theory, by existing
standards. The common approaches seem to be insubstantial in one or both of the ways we identified.
Could there be such a theory? I think we can show that there could not.

2.3 TWO ARGUMENTS THAT TRUTH CANNOT BE SUBSTANTIAL

The first argument generalizes the sceptical challenge. Marker theorists and non-factualists both claim
to find a division in language between fact-stating and non-fact-stating uses. A substantial theory of
truth would permit a particular account of this distinction. It would allow the theorist to say that the
fact-stating uses are distinctive in standing in a certain (yet to be specified) relationship to the
properties of truth and falsity. Given such a proposal, the sceptic challenges the theorist’s warrant for
assigning any given utterance to one side of the division or other.

For simplicity let us focus on the claim that marker theorists and non-factualists will have in common,
that non-indicatives do not stand in the designated way to truth. How could such a claim be justified?
Consider an example. On what grounds do we say that sentences such as those in this paragraph do
not have the property supposedly marked by (or confined to some sub-class of) the indicatives? In
other words, how do we know that these sentences are not fact-stating?

Ordinarily we might appeal to their mood, but what is now at issue is our right to the kind of theory
that would allow such an appeal. Perhaps then we should appeal to the intuition that in uttering such
sentences we do not express beliefs, make any claims about the world, and the like. However, to
depend on claims of this sort is to render our theory of truth insubstantial in the second of the two
ways we distinguished earlier. A substantial theory of truth cannot rely on alternative
characterizations of fact-stating discourse.

The last alternative seems to be to appeal to ordinary usage – to the fact that the sentences concerned
are not ordinarily ascribed truth values. We have noted, however, that in distinguishing fact-stating
from non-fact-stating discourse, marker theorists and non-factualists both want more than a merely
descriptive characterization of ordinary usage. Among other things a substantial theory of truth is
supposed to provide an analytically grounded explanation of usage. But it seems that there can be no
evidence for such a theory, other than the merely linguistic phenomena it is intended to explain.

18
It might be objected that this is a common predicament for scientific theory. It seems that the grounds
for acceptance of a theory often consist simply in the fact that it offers the best available explanation
of a range of phenomena. Such cases lack a crucial feature of the present linguistic case, however. In
trying to account for distinctions in usage in terms of the fact-stating/non-fact-stating distinction,
marker theorists and non-factualists both seem to have regarded the relevant features of usage as
responsive to an underlying distinction. The marker theory, in particular, seems to involve that the
choice of grammar be guided by at least an implicit recognition of underlying linguistic categories.
The sceptical argument challenges the possibility of such recognition (given, in the present case, that
these categories are supposed to rest on a substantial theory of truth). There is no parallel to this
recognitional requirement in the scientific cases, and hence no easy answer to the sceptic by analogy
with those cases.

The role of implicit recognition of the underlying categories is well illustrated by Stenius’s version of
a truth-based marker theory. We saw earlier that if assertions are to be characterized in terms of the
convention that they aim at truth, truth must be considered a property of the members of a class of
unasserted sentences: Stenius’s sentence radicals. Making an assertion is then to be thought of as
uttering a sentence radical, while indicating by means of its (indicative) mood that one takes it to be
true. There are parallel conventions for the other moods. However, such conventions require that
speakers be able to distinguish the sentence radicals in advance of the choice of mood. The sceptic
argues that this is impossible (at least in terms of a substantial theory of truth). Our only guide to the
limits of the class of sentence radicals is the indicative mood itself. So the convention that supposedly
governs that mood turns out to presuppose an understanding of the significance of mood.

The second and perhaps more familiar argument against the possibility of a substantial theory of truth
also turns on the fact that such a theory makes truth conceptually prior to assertion. If fact-stating
constructions are characterized as sentences that bear a certain relationship to the properties of truth
and falsity then the distinction between truth and falsity comes into question. For example, couldn’t
we imagine a group of speakers who followed the convention that they aimed to utter false sentences.
This convention seems no less useful than the ordinary one for purposes of communication. If one of
these speakers says ‘I’ve made the tea’, then knowing the convention we know what he means. He
means that he has not made the tea. However, what he said was supposed to be false. (We were
imagining that he follows the convention of trying to utter false sentences.) But the truth value of a
sentence depends on what it means. If he meant that he had not made the tea, then if we agree we
should say that the sentence he uttered was true, not false. ‘For a sentence is true if things are as we
say they are by means of it; and if by “p” we mean “~p”, and things are as we mean them, then “p”
is on the new interpretation true and not false.’ The sentence comes from the Tractatus (4.062).
Dummett (1973, p. 318) refers to it in support of the contention that assertion cannot be analysed in
terms of truth. Indeed, in his (1958) paper on truth Dummett criticizes ‘Frege’s theory of truth and
falsity as the references of sentences’ on the grounds that it ‘leaves ... quite out of account’ that

it is part of the concept of truth that we aim at making true statements .... Frege indeed tried to
bring it in afterwards, in his theory of assertion – but too late; for the sense of the sentence is
not given in advance of our going in for the activity of asserting, since otherwise there could
be people who expressed the same thoughts but went in instead for denying them. (1958, p.
143, my italics)

The two arguments of this section show the absurdity of Stenius’s ‘descriptive’ notion of truth – of
the idea that sentences could be considered to have truth values (or indeed meanings) in advance of an
understanding of their conventional linguistic role. On the contrary: a sentence can be taken to have a
significant truth value only in virtue of the understanding that its conventional use is such as to admit
a standard of right and wrong, of correctness and incorrectness. (This is a minimum requirement;
presumably not just any standard will do.) The true sentences are those whose utterance would be
correct, according to this standard. In other words (now taking for granted what we are trying to
explain), a true sentence is true in virtue of the fact that it would be correct to assert it – which reduces

19
the claim that in making assertions we aim to speak the truth, to the truism that in making assertions,
we aim to do so correctly.

Dummett criticizes Stenius in similar terms, but seems to think that the point concerns only the
distinction between truth and falsity:

Stenius insists that the ‘sentence-radical’... itself has a ‘directed’ sense, independently of the
assertion which can be expressed by attaching the assertion sign to it. Admittedly, the
‘sentence-radical’ has ‘true/false poles’, in that its sense consists in the distinction between
conditions under which it is true and conditions under which it is false: but we can determine
the ‘direction’ of this sense ... only by considering its use when assertoric force is attached to
it, because it is only by reference to its role in assertions that we can identify one set of
conditions as those in which it is true and the other as those in which it is false. (1973, p. 327)

However, the sense of a sentence radical seems equally dependent on its role in assertion. Dummett’s
own views on truth and sense suggest as much: if ‘the notion of truth has first to be explained’ in
terms of ‘a distinction between correct and incorrect assertions’ (1978, p. 20), and ‘sense consists in a
distinction between conditions under which [a sentence radical] is true and conditions under which it
is false’ (1973, p. 327), then a grasp of sense already depends on a conception of potential assertoric
use.

2.4 TRUTH THEORIES: INSUBSTANTIAL THEORIES OF TRUTH

Let us turn now to the promised example: David Wiggins’s (1980) attempt to construct a substantial
theory of truth. ‘What would be a substantial theory of truth?’ is Wiggins’s contribution to a
Festschrift for P. F. Strawson. He begins, appropriately, with some remarks about Strawson’s views
on truth. In particular, he claims (1980, pp. 190–1) to find in Strawson’s (1950) article on truth the
view that ‘of proper concern to the theory of truth ‘ is ‘the task of elucidating the nature of a certain
type of communication’: the ‘empirically informative’ or ‘fact-stating’ use. He thus counts Strawson
as an early advocate of the need for what Williams was to call a substantial theory of truth: ‘I do not
doubt that what [Williams] felt he needed was the very same thing that Strawson had envisaged in
1950’ (Wiggins 1980, p. 192).

Strawson does refer to the task just mentioned. However, when he asks ‘why should the problem of
Truth ... be seen as this problem of elucidating the fact-stating type of discourse?’ his ‘answer is that it
shouldn’t be .... The problem about the use of “true” is to see how this word fits into [the fact-stating]
frame of discourse. The surest route to the wrong answer is to confuse this problem with the question:
What type of discourse is this?’ (1950, pp. 142–3). At that time, at any rate (there is no hint of it in his
reply to Wiggins in van Straaten 1980, pp. 294–6), Strawson’s view thus seems to have been that the
problem of truth arises only within the fact-stating form of discourse, and hence depends on, or
assumes, an understanding of that form of discourse.

Wiggins also claims Dummett as an early advocate of Williams’s project, quoting from Dummett ‘s
(1958) paper on truth. But we have seen that in that paper and elsewhere Dummett rejects the idea
that assertion can be explained in terms of truth. He again emphasizes the point in a postscript to the
(1958) paper, first published in 1972:

What has to be added to a truth definition for the sentences of a language, if the notion of
truth is to be explained, is a description of the linguistic activity of making assertions; and this
is a task of enormous complexity. What we can say is that any such account of what assertion
is must introduce a distinction between correct and incorrect assertions, and that it is in terms
of that distinction that the notion of truth has first to be explained. (Dummett 1978, p. 20)

20
We shall see in chapter 3 that Dummett prefers a quite different account of the distinctive features of
assertion.

History aside, Wiggins’s approach is to consider what must be true of truth if the predicate ‘is true’ is
to figure in a truth theory adequate for radical interpretation. More precisely, the idea is to look at
theories in the style of a Tarski truth theory for a language L, yielding equivalences of the form

(2.1) S is f if and only if P.

Such a theory is to be regarded as specifying the meaning of the L-sentence S. There are many such
theories, but some of them make better sense than others of the speakers of L. The constraints on f
imposed by the anthropological constraint of ‘making sense’ are the marks of the notion of truth.

I am going to argue that the distinction that a substantial theory of truth is supposed to help us with –
between assertoric or fact-stating uses of language, and all the rest – is already built into the truth-
theoretic approach to meaning, as Wiggins employs it. Wiggins gets out only what Davidson and
others have already built in. The project of radical interpretation, even when constrained by the
structural insights we have from Tarski, does not in itself yield either the Fregean distinction between
sense and force, or (what would come to the same thing) a substantial theory of truth. These have to
come from somewhere else.

Wiggins’s strategy does not consist simply in a Tarskian definition of truth. Tarski’s approach takes
for granted the notion of translation. His ‘condition of adequacy’ is that a truth theory should yield
sentences of the form

(2.2) S is true if and only if P

where P is a translation in the metalanguage of the object language sentence to which the description
S refers. Whereas for Davidson and the truth theorists who follow him, the priorities are in a sense
reversed. The T-sentences are now expected to yield interpretations of the sentences of the object
language, and the theory is constrained, roughly speaking, by the requirement that it make good
anthropological sense of the speakers of the object language. Moreover, it is not crucial to this
programme (as it was to Tarski) that the predicate the theory invokes be the predicate ‘is true’. John
McDowell has repeatedly emphasized this point. McDowell describes what he calls ‘the best version
of the truth-conditional proposal ... along these lines’:

We may reasonably set ourselves the ideal of constructing, as a component of a complete

theory of meaning for a language, a sub-theory which is to serve to specify the contents of
(for instance, and surely centrally) assertions which could be made by uttering the language’s
indicative sentences .... [A] direct assault on that task would be to look for a sub-theory which
generates, on the basis of structure in the object-language sentences, a theorem, for every
appropriate sentence, of this form: ‘s can be used to assert that p’. Now there is a truistic
connection between the content of an assertion and a familiar notion of truth ...; the
connection guarantees, as the merest platitude, that a correct specification of what can be
asserted, by the assertoric utterance of a sentence, cannot but be a specification of a condition
under which the sentence is true. A radical proposal at this point would be as follows: as long
as the ends of the theorems (think of them as having the form ‘s... P’) are so related that,
whatever the theorems actually say, we can use them as if they said something of the form ‘s
can be used to assert that p’, it does not actually matter if we write, between those ends,
something else which yields a truth in the same circumstances; our platitude guarantees that
‘is true if and only if’ fits that bill, and this gives a more tractable target than that of the direct
assault. (1981, pp. 228–9)

21
McDowell here takes for granted the association between assertoric force and the indicative mood. In
speaking of the indicative sentences of an arbitrary language, he clearly has in mind the assertoric
sentences. The idea is thus that the truth theory (or rather, what McDowell says may as well be
regarded as a truth theory) yields an account of the descriptive content, or sense, of what are
effectively the sentence radicals of the language in question. For a complete theory of meaning, we
need to add to this content-specifying sub-theory a theory of force – an account of the various
different sorts of things that speakers of the language do with sentence contents (including, of course,
asserting them).

As we noted in chapter 1, the extent of this additional theory of force depends on how many varieties
of force there are reckoned to be in the object language in question. This suggests that other things
being equal it would be better to treat differences of meaning as differences of sense than as
differences of force; for this way they are incorporated into the theory of sense, and don’t have to be
dealt with later. Every effort should be made to handle a difference of meaning within the sort of sub-
theory that McDowell describes, before consigning it to the assortment of leftovers which will have to
be dealt with when that sub-theory is complete. Of course, if we had some criterion for whether a
difference of meaning was a matter of sense or a matter of force, there would be no question of trying
to accommodate it in the sub-theory; it would be in or out, according to what the criterion told us. But
at the moment we have no criterion. A substantial theory of truth is supposed to give us one.

Moreover, McDowell’s care in distinguishing a theory whose primary role is to state the contents of
sentences of an object language from a theory of truth, removes what may have seemed a barrier to
accommodating certain sorts of differences of meaning within a theory of this general form. The
standard ‘filling’ for a T-sentence – ‘is true if and only if – suggests that what goes into the blanks
must at the left hand end designate the sort of object language sentence to which the predicate ‘true’
applies; and at the right hand end be the sort of metalanguage sentence that can form the antecedent or
consequent of a material conditional. But McDowell’s insistence that the standard filling is not
obligatory should alert us to the possibility of a filling that does the same work, without imposing
such a grammatical restriction; and that hence enables the same process of meaning specification to
extend to what are traditionally regarded as matters of force.

McDowell suggests the filling

(2.3) ... can be used to assert that ....

Of course, he has already restricted the theory to assertions – to ‘the language’s indicative sentences’.
Except in that it actually hints at this restriction, 2.3 is no different from

(2.4) ... can be used to say that ....

However, 2.3 and 2.4 might still seem to restrict the kinds of sentences we can use to fill the blanks.
On the right-hand side, in particular, we are still effectively restricted to indicative sentences; for
English grammar won’t allow anything else to fill the gap in this sort of ‘that’ clause.

At this point, however, we can appeal to Davidson’s own (1968) ‘paratactic’ analysis of this kind of
grammatical construction. On this analysis the logical form of the sentence ‘Ramsey said that truth is
redundant’ is

(2.5 Truth is redundant. Ramsey said that.

This allows that the clause which appears after ‘that’ in the original sentence is used, rather than
mentioned, and hence makes sense of the intuition that in translating this remark about Ramsey into
another language, we should not leave this clause in English. (We are not quoting Ramsey.)

22
The present relevance of this proposal is that it frees us from grammatical restrictions on the nature of
‘that ...’ clauses.

(2.6) Make truth redundant. Ramsey said that.

makes as much sense as 2.5. So indeed does

(2.7) Is truth redundant? Ramsey said that.

Of course, in this case we would be inclined to say ‘Ramsey asked that’, but this is incidental. The
important point is that just as in 2.5, the imperative and interrogative sentences in 2.6 and 2.7 are used
rather than mentioned. We are not reporting Ramsey’s exact words. We are samesaying, as Davidson
(1968, p. 140) puts it, in our own words.

Applying the same analysis to 2.4, the ‘filler’ we derived from McDowell, we find that the theorems
of resulting sub-theory can be regarded as having the form

(2.8) S can be used to say this: P,

and that there is now nothing in the form of these theorems to prevent the sub-theory from extending
to the non-indicative sentences of the object language. If the object language contains imperatives, for
example, there is now nothing to prevent the sub-theory giving us their meaning in the same way it
does for indicatives: by using a metalanguage sentence, and telling us that this sentence says the same
thing as the given sentence of the object language. Note that the fact that the metalanguage sentence is
still used rather than mentioned means that the theory is still distinguished from a translation manual
in the way that truth theorists have been keen to stress. (See, for example, Evans and McDowell 1976,
p. x.)

It now appears that the preferred form for what was supposed to be a theory of the content of the
sentences of the object language, makes no distinction between differences in content (or sense) and
differences in force. The tasks initially consigned to the separate theory of force are in fact taken care
of by the so-called sub-theory. The significance of the force indicators of the object language – its
imperative and interrogative moods, for example – is explained in the same way as that of sense-
modifying expressions: by the use of corresponding constructions in the metalanguage. The
supplementary theory, originally thought of as a theory of force, survives, at most, in a force-neutral
account of the activity of saying; a mere gloss to the main theory, whose main task would be to
distinguish the full-blooded use of sentences from other sorts of appearances in writing and in speech.

This need not mean that there is no longer a place for the notion of force; only that the distinction
between force and sense need not be reflected in the radical interpreter’s methodology, as usually
supposed. As long as we draw such distinctions in the metalanguage, the process of interpretation will
draw corresponding distinctions in the object language. But the process makes no intrinsic demands
on such distinctions – which suggests that we can’t found an elucidation of the fact-stating/non-fact-
stating distinction (or, what is supposed to come to the same thing, a substantial theory of truth) on the
requirements of radical interpretation.

If this conclusion seems surprising, I think the impression stems from a distorted view of the aim of
radical interpretation. Radical interpretation is most naturally seen as the programme of systematic
translation of the sayings of others into the home language. ‘Sayings’ here means sayings of all kinds
– as many kinds as we need, to understand what our subjects are saying. From this point of view the
introduction of a Fregean sense/force distinction looks rather odd – a case of changing horses in mid-
stream. So from the radical interpreter’s natural perspective there is an odd discontinuity in a truth-
theoretic theory of meaning. We have seen however that truth-theoretic general semantics shows a

23
constitutional tendency to concentrate on assertoric uses of language. This emphasis obscures the
discontinuity, because it obscures the fact that a truth-theoretic theory of meaning leaves us some way
short of the interpreter’s goal. Saying comes to be equated with asserting, and the radical interpreter’s
programme comes mistakenly to be identified with the task of specifying the contents of assertoric
utterances.

I want to mention two objections to the above conclusion – two considerations that might be thought
to show that a radical interpreter does have methodological grounds to distinguish indicatives from
non-indicatives. (I discuss these and other objections at greater length in Price 1987.) The first stems
from a popular view of the empirical grounds of a theory of radical interpretation: the view that ‘the
evidential base ... will consist of facts about the circumstances under which speakers hold sentences of
their language to be true’ (Davidson 1974, p. 320). It might be claimed that the radical interpreter’s
theory will need to embody a sub-theory of the kind McDowell describes, in order properly to reflect
its empirical base.

The objection is easily turned, however. The available ‘facts about circumstances under which
speakers hold sentences of their language to be true’ cannot exceed the available facts about
circumstances under which speakers are prepared to utter these sentences, with serious intent. Adding
the truth predicate doesn’t help to distinguish the literal utterances from all the rest; and doesn’t, for
example, help us to decide whether reluctance to make an utterance is due to politeness or disbelief.
The subject who is too polite to say, ‘You linguists all look the same to me’, will be too polite to
assent to the suggestion that we linguists all look the same to him.

On the other hand, to insist on framing the evidence in terms of ‘holding true’ is to restrict one’s gaze,
quite unnecessarily, to those sentences that can be held true. It is not clear that such a restriction is
possible – it depends on being able to distinguish (what will be interpreted as) the indicative sentences
at a very early stage. But even if it is possible, there can be nothing to be gained by it – far better to
collect facts about the circumstances in which sentences of all kinds are seriously uttered. The facts
about ‘holding true’ can then be distilled, if necessary, when we’ve distinguished the indicatives – and
the residue will do us nothing but good. (The interpreter’s standard question will thus be, not ‘Would
you assent to this: P?’, but ‘Would it be appropriate to say this: P?’)

Secondly, the truth theorist might object that the justification for restricting one’s initial meaning-
specifying theory to the indicatives lies in the advantages of expressing its theorems in the predicative
form exemplified in 2.2. This point is best coupled with another. The main thing that recommends a
truth-theoretic approach to a theory of meaning is its ability to deliver plausible recursive clauses for
at least some of the sentence-forming operations of natural language; while at the same time keeping
faith with the intuition that an adequate theory of meaning will give us the meanings of expressions of
the object language by using, rather than mentioning, expressions of the metalanguage. It might be
thought that these virtues justify what otherwise seems a needless complexity: the restriction of the
meaning-specifying theory to indicative sentences, and the consequent introduction of a separate
theory of force.

This argument depends on the assumption that it is impossible to express the theorems of our
extended sub-theory in the predicative form of 2.1. In other words, it turns on the impossibility of a
predicate¢ such that for any sentence P, ‘P is f’ can be used to say that P – i.e. such that

(2.9) ‘P is f’ can be used to say this: P.

However, in a sense we have already encountered such a predicate, in Strawson’s amen theory of
truth. We saw that the amen theory does not explain why ‘true’ should not be used to endorse any sort
of utterance whatsoever. It is thus insubstantial in the first of the ways we distinguished in section 2.2.

24
We noted that the same can be said of any theory of truth that relies on the equivalence property – on
the fact that for any (appropriate) sentence P, P and ‘P is true’ are in some sense equivalent in
meaning. However, it is precisely this feature of truth that is exploited by the present objection. For
imagine that we could generalize truth to the non-indicatives, its significance being given by the
generalized equivalence principle:

(2.10) For all sentences P, P is true if and only if P.

The theorems of our general meaning-specifying theory could then be written in the predicative form
of 2.2, and the advantages of such a form could not be the grounds for a discontinuity in a programme
of radical interpretation. Our generalized truth theory would do what an orthodox truth theory does,
but do it more generally. Certainly, we could carve the orthodox theory out of the general one; but
only if we already knew what to cut away.

So the objector needs a reason why we shouldn’t generalize the equivalence principle, and hence
apply ‘true’ to non-indicatives. As we saw, however, such a reason will call for an independent
characterization of the indicatives, in order to account for the required restriction. This defeats the
avowed aim of a substantial theory of truth. There is surely some reason why we don’t use ‘true’ in
this way. But unless we know more about truth than the equivalence property tells us, it is bound to be
a reason to which a substantial theory of truth is not allowed to appeal.

In a footnote to the passage quoted above, McDowell remarks that ‘it is a philosophical issue whether
there are respectable purposes for which a stronger notion of truth is required’ – a stronger notion, in
other words, than that guaranteed by the ‘platitude’ that ‘ ... is true if and only if ...’ is a suitable filling
for the theorems of the meaning-specifying sub-theory. McDowell notes that non-factualists will want
a stronger notion, but says that he is ‘inclined to suppose that this is a matter not so much of an
alternative notion of truth as of a characteristically philosophical misconception of the only notion of
truth we really have: one which the platitude in fact suffices to determine’ (1981, p. 229, n. 9). We
find, however, that the platitude does not account for the ordinary use of truth. The truth-theoretic
approach to truth is insubstantial in the first of the ways we earlier distinguished. It picks out the
indicatives (or any smaller class) only to the extent that it takes for granted that such utterances are the
only bearers of truth. So McDowell’s platitude does not suffice to determine the fact that truth is not
applied to the non-indicatives; and the non-factualist is not alone in wanting something stronger.

2.5 THE MARKS OF TRUTH

Wiggins claims to derive certain ‘marks of a theory of truth’ from the anthropological constraints on a
Davidsonian programme of radical interpretation. However, we have seen that that programme does
not yield the kind of distinction that a substantial theory of truth is supposed to underpin. If we
already have the distinction, interpretation will reflect it; but it needs to come from somewhere else.
What then of Wiggins’s marks of truth? Are they properties of truth that, like the equivalence
property, could equally extend to non-indicatives? Or do they depend on some prior conception of the
nature of fact-stating discourse – on the sort of conception that a substantial theory of truth is
supposed to provide? The answer, I think, is the latter. Wiggins allows himself the notions of
assertion and belief.

Thus ‘the first mark of assertibility (and so of truth)’ consists in ‘assertibility’s being the primary
dimension of assessment for sentences, or that property which sentences have normally to be
construed as aiming to enjoy’. If there were no such norm, Wiggins says, ‘interpretation would either
be impossible ... or unjustifiable’ (1980, p. 205). Wiggins’s point seems to be that unless we take
utterance to be constrained by such a norm, we shall be unable to take seriously anything that our
speakers say. But only the implicit assumption that by ‘utterance’ we mean ‘assertion’ can justify
Wiggins’s way of stating the constraint. Certainly, we need to assume that by and large our subjects

25
are speaking seriously – that they mean what they say – but only in the case of assertions does this
amount to the assumption that they are aiming to speak the truth.

As for Wiggins’s remaining marks of truth, they make essential use of the claim that ‘language ... par
excellence ... is the vehicle for the communication and expression of belief (1980, p. 205). Thus: ‘The
argument for [the third mark) ... is similar to the argument for the second mark in turning upon the
notion of belief (pp. 208–9); ‘... the fourth mark [is] simply... a condensation of the first, second and
third ...’ (p. 211); and as for the fifth mark, ‘it is a norm of rational belief that one is committed to
conform one’s beliefs to this requirement; and the interpretation of beliefs as beliefs has to see belief
as answerable to it’ (p. 212).

As a result, Wiggins’s theory of truth is insubstantial in the second of the two senses of section 2.2.
Given the notions of belief and expression of belief on which Wiggins here relies, we could quite well
characterize the fact-stating use of language in terms of expression of belief. The basic distinction
would be that between those utterances whose standard function is to express a belief and those that
do something else. Far from looking for a substantial theory of truth, we ought then to expect our
account of truth to flow from the same source as our theory of assertion. The obvious approach would
be to regard truth as primarily a property of belief, and thence a derivative property of expressions of
belief. The correctness of an assertion would thus be a matter of the correctness of the belief it would
normally express.

The argument of this chapter should alert us to a potential hazard for an account of this kind. We have
seen that assertion cannot be characterized in terms of truth. On the contrary, an account of truth and
falsity as linguistic properties seems bound to depend on an elucidation of the fact-stating type of
discourse. In offering such an elucidation we must be careful not to depend on a linguistic notion of
truth. In particular, therefore, we cannot rely on a characterization of belief that itself depends on a
notion of truth as a property of linguistic entities. This notion should arise from, not underpin, our
psychological theory. If we want to analyse the fact-stating/non-fact-stating distinction in terms of a
distinction between beliefs and other sorts of psychological state, we must be prepared temporarily to
forego the linguistic notion of truth. This point will be very much in evidence in chapter 5, where we
shall be looking in more detail at attempts to find a psychological basis for the notion of statement of
fact.

26
3
The Insignificance of Mood
Marker theorists and non-factualists agree that there is a non-trivial distinction between fact-stating
and non-fact-stating uses of natural language. They differ, however, as to where this distinction
should be drawn. We have seen that if the marker theorist tries to settle the issue by appealing to truth,
the attempt backfires. A notion of truth that secures the indicatives tends also to encompass the non-
indicatives. We shall see now that this is the inevitable predicament of the marker view. No analysis
of fact-stating discourse that includes the indicatives can exclude the non-indicatives. This will enable
us to put the marker theory and the non-indicatives to one side. We shall be able to concentrate, in
other words, on whether there is a distinction to be drawn within the indicatives.

Our goal is thus a generalized sceptical challenge to the marker view – an argument that non-
indicatives cannot be shown not to be assertions. Since this would be entailed by a demonstration that
non-indicatives are assertions, our course runs parallel to that of the ‘universal factualist’, who holds
that all discourse is fact-stating. However, we’ll see that the path leads not to universal factualism but
to our own less ambitious destination. It does not establish that non-indicatives are assertions; but it
does provide a plausible reduction of any purported demonstration that non-indicatives are not
assertions.

It will be helpful to have in mind some of the characteristics that have been thought to distinguish the
non-indicatives, other than their standing to the properties of truth and falsity. So let us again refer to
the views of Dummett, whom we have seen to be a critic of the attempt to analyse assertion in terms
of truth. In his discussion of assertion in Frege: Philosophy of Language, Dummett mentions four
main alternative characteristics. He says that assertions are the expressions of mental states with inner
consequences (so that we can make sense of an internal assertion – a judgement – but not for example
an internal command); are not typically associated with definite consequences (as are questions and
commands); can be used to deceive; and can be used (perhaps stripped of assertoric force) as
compone n ts of complex sentences. (Dummett 1973, chapter 10).

Not all of these features could plausibly be offered by way of analysis of assertoric discourse (or are, I
think, so intended by Dummett). The last property is clearly grammar-dependent; and hence, as we
saw in chapter 2, no use to the marker theorist, whose aim is to explain the grammar. At best this
criterion amounts to a further characterization – a rather obvious one – of the distinction in usage that
calls for explanation.

The suggestion that only assertions can be used to deceive also seems an unlikely basis for an analysis
of assertion. If true, it surely rests on some more fundamental property of assertion. However, it
proves a useful focus for our present concerns. In asking how non-indicatives can be used to deceive,
we shall be led to a notion of the informational content that such an utterance could be considered to
convey. From there it will be a short step to the claim that a non-indicative can be construed as an
assertion with such a content – or rather, cannot be shown not to be such an assertion.

The attempt to make sense of non-indicative deception will lead us in particular to the proposal that
non-indicatives are equivalent to explicit performatives (these being themselves assertoric). This
proposal – advocated in the interests of general semantics by David Lewis (1972) – has been
criticized on a number of grounds. I shall defend it against the better known objections, but argue that
it is nevertheless untenable. Once recognized, however, the main problem is easily solved. I shall
argue that the improved account meets our purposes (if not the universal factualist’s); and confirm
that in so far as they need to do so, the assertoric paraphrases suggested by the account do indeed have
the features Dummett takes to distinguish the non-indicatives.

27
3.1 NON-INDICATIVE DECEPTION

The universal factualist has two possible responses to the claim that assertions are distinctive in
allowing deception: to argue that some assertions (and in particular non-indicative assertions) cannot
be used to deceive; or to try to show that non-indicatives can be used to deceive. Let us begin with the
latter possibility.

In support of the claim that assertions can be characterized in this way, Dummett argues that non-
indicatives show no parallel to Moore’s paradox. He concedes that ‘the asking of a question is in a
sense the expression of a desire to know the answer: so one may ask a question to deceive someone
into thinking that one wants to know the answer.’ But he counters that this

is not ... part of the convention which is learned when one learns to use and answer questions:
it merely results from the fact that a desire to know the answer provides the usual motive for
asking a question. This is shown by the fact ... that there is nothing self-defeating in asking a
question and at the same time letting it be seen that one has no interest in the answer (if, e.g.,
one is an official whose duty involves the posing of certain questions); whereas the making of
an assertion while letting it be seen that one does not believe it to be true is intrinsically self-
defeating. (1973, p. 356)

At best, however, this shows us that a question does not express an interest in knowing its answer. It
does not show that there is not some other attitude whose disavowal will frustrate an interrogative
speech act. And on the face of it, any linguistic act can be defeated by an appropriately chosen attitude
report. One simply needs to add that the utterance was not intended seriously, or sincerely. This takes
advantage of the most general linguistic convention of all, namely that one means what one says.
There may be various ways of signifying that a previous utterance should not be taken as governed by
this convention, but there is always at least one way: thanks to Quine, one need only add that the
utterance should be counted a mention rather than a use. True, not all effects of an utterance can be
cancelled in this way. Some effects are insensitive to a speaker’s intentions: for example, effects on
voice-operated machinery. Others are immediate, so that the disavowal after the fact is bound to come
too late. (Think of crying ‘Wolf!’ – or for that matter, of the effects of ‘Think of a wolf!’) Clearly,
however, such cases do not distinguish the non-indicatives.

Does the possibility of insincerity allow non-indicatives to be used to deceive? It might be objected
that pretending to sincerity is not the same as lying. The liar pretends to sincerity, but not all
pretenders to sincerity are liars. A person who utters an insincere non-indicative misleads his audience
about his own sincerity, but does not lie. Whereas a person who lies misleads her audience in two
ways: about her own sincerity, and (if she is believed, at least) about the facts. To lie is to try to
mislead in the second sense. In the case of commands, for example, the claimed contrast would thus
be that a person who obeys a command is in no way at fault for doing so, even if it turns out that the
command was insincere. Whereas a person who unknowingly accepts an insincere assert ion may end
up, not at fault for accepting it – that was the appropriate thing to do – but at fault in possessing a
false belief. It is a virtue of this difference, it seems, that lying misleads about more than a speaker’s
sincerity.

However, it is easy to see how obedience to insincere commands might be inappropriate. In general
obedience will involve some disadvantage, or behavioural cost; but this will be regarded as preferable
to the disadvantages of disobedience – in other words, to being punished for failure to comply.
However, the relevant social conventions might well recognize the possibility of insincere commands,
and administer punishment accordingly. Disobedience to an insincere command would thus go
unpunished. Hence we would have a worldly basis for the state of being misled by an insincere
command, in the consequent performance of an unnecessary and otherwise undesirable action.
Moreover, we could match such a state to a false belief. We could say that someone wrongly believed
that he was required to perform some action. This suggests that we might interpret a command as an

28
assertion with content of this belief – for example, that we equate ‘Fill in this form’ with ‘You are
required to fill in this form.’

To forestall an objection, let me clarify the role of the doubtful (and certainly contingent) proposition
that insincere commands do not establish real obligations – that a person is not at fault for failing to
comply with what turns out to be an insincere imperative. The suggestion does not require that actual
commands conform to this convention. The point is rather that since the giving of commands clearly
could be constrained in this way, we can make sense of a worldly basis for imperative deception. In
terms of that basis we can then explain, if necessary, the fact that some or all commands do not permit
this sort of deception. I shall come back to this point. But briefly: we can say that some or all actual
commands are governed by the convention that they be self-justifying – the convention that their own
utterance be sufficient to establish the requirement that makes their suggested paraphrase true.

There would be other object ions to any serious proposal to paraphrase imperatives as reports of
obligations. As yet, for example, such a report does not capture the first-person character of a
command. Perhaps the paraphrase could be refined to cope with such objections, but for present
purposes there is a more serious concern. Even if reports of requirements can be made to do duty for
imperatives, we shall be left with the other non-indicatives. Insincere commands may mislead us
about our obligations, but what about insincere questions, requests, wishes, and so on? A case by case
argument is no use to us. We need a general guarantee that some such transformation will always be
available. I want to show that the marker theory gives us such a guarantee. This will deliver the
promised reductio. It will show that from the marker theorist’s point of view, universal factualism
remains a viable alternative.

3.2 HOW TO CONSTRUE THINGS IN WORDS

The general strategy relies on some familiar point s about linguistic communication. Utterances have
various effects on the social environment of speakers and audiences. Among these effects are some
that depend on linguistic understand ing – on the fact that audiences are able to interpret what
speakers say. Effects of this kind are convention governed. They rely on the meaning conventions in
force in the com m unity in question. Interpretation of an utterance can be thought of, in part, as a
process of acquiring beliefs about these convention-governed effects. These beliefs are not often
explicit; but it would seem that any of them could be elicited by an appropriate enquiry, at least from
a subject with the necessary conceptual resources. Learning to understand a language seems in large
part a matter of learning to form such beliefs in the right circumstances.

I take this much to be uncontroversial. The difficulties arise when we try to fill in the details. In
particular, the interpretation of a single utterance will in general give rise to a string of beliefs. The
proper characterization of some of these beliefs is far from obvious. Interpretation presumably
involves the application of the interpreter’s knowledge of meaning, and hence the proper form of
interpretative beliefs will depend on that of a theory of meaning for the language in question. A theory
of meaning will bring with it an account of the corresponding interpretative beliefs.

The marker theory, for example, claims that mood distinctions mark significant meaning categories. If
so, then the process of interpretation should recognize the factor of meaning marked by mood. Hence
on this theory we might think of the main elements of the interpretative string as the following four
beliefs: that someone has spoken (rather than simply made a noise); that a certain speech act (a
command, an assertion, or whatever) has been performed; that the speech act concerned has a
particular content; and that one is oneself thereby affected in some way (for example in being thereby
obliged to perform some action).

The present significance of these points lies in the fact that beliefs can be acquired not only directly,
in virtue of perceived features of one’s environment; but also indirectly, as a result of an assertion by
somebody else. So if the interpretation and hence the effects of an utterance always depend on chains

29
of belief, we might achieve the same final effects in a different way: that is, by simply telling a person
what they would otherwise come to believe at some point in the chain of interpretation. From the
marker theorist’s point of view this is what happens when we replace a command with an assertion of
the existence of a corresponding requirement. In both cases we normally cause our audience to
believe that there is such a requirement, but the initial pathways are different. In one case the belief
concerned is inferred from the observation that we have issued a command; in the other it flows in the
normal way from the observation that we have made an assertion – from the fact that we have said
that there exists the requirement in question. But note that not only do the pathways converge, in
giving rise to the same belief; but also the effects of initial insincerity may be the same, in making this
belief false.

The marker theory suggests a simple and elegant way to adapt this construction to the general case.
We want to break into a chain of interpretative beliefs, so as to assert what a hearer would otherwise
infer. It doesn’t seem to matter where we break in, as long as we can characterize the information the
new assertion needs to convey. Given the marker theory, the obvious target is the point at which the
mood itself is first deciphered. At this point the interpreter comes to believe that he is faced with a
question, a request, or whatever. Why not replace the original utterance with an assertion to this
effect?

The suggestion is thus that we exploit the marker theorist’s account of the significance of mood. That
account entails that the process of interpretation of an utterance requires, inter alia, a belief that a
speech act of a certain kind has taken place – the kind being the meaning category that the marker
theorist associates with the mood of the utterance in question. The marker theorist takes this belief to
determine the further effects of the utterance concerned (or those, at any rate, that depend on its
mood). The suggestion is that we imagine the same belief to be transmitted in another way: not
inferred from mood, but reported in content. Non-indicatives are thus to be paraphrased in terms of
explicit performatives (these being themselves regarded as assertoric). ‘Fill in the form’ will now be
read not as ‘You are required to fill in the form’, but simply as ‘You are commanded to fill in the
form’; or more explicitly, as ‘I command you to fill in the form.’

In this way, the marker theory seems to provide the universal factualist with the material for an
assertoric interpretation of the non-indicatives. It allows the factualist to propose that non-indicatives
are merely ‘disguised’ reports of the occurrence of the very speech acts the marker theory claims to
identify. To admit such readings would be to abandon the claim that the indicative mood is the mark
of assertoric force. But how is the marker theorist to exclude them, given the intuitive appeal of the
interpretative model from which they stem?

3.3 PERFORMATIVES: TOO EXPLICIT FOR THE FACTUALIST

The marker theorist might well follow J. L. Austin (1962, pp. 5–6, for example) in denying that
explicit performatives are fact-stating – as of course the above suggestion requires, if it is to give us
assertoric as opposed to merely indicative, substitutes for non-indicative utterances. In our terms,
Austin was a non-factualist: he took it as a discovery about language that it contains this large class of
non-assertoric indicative sentences. On Austin’s view, the expression ‘I command ...’ is not used to
describe one’s own behaviour – to assert that one commands – but simply to command; similarly with
‘I request ...’, ‘I promise ...’, and many others.

Some of Austin’s critics, however, though agreeing that explicit performatives are used for these
purposes, have argued that such utterances are also assertions. One such critic is David Lewis, whose
interest in explicit performatives is much the same as ours: faced with the task of incorporating non-
indicatives into a general theory of meaning, Lewis recommends that they be treated as disguised
performatives, and that these be regarded as assertoric, or declarative. In Lewis’s words, the result is
that

30
 the distinction between declarative and non-declarative sentences becomes a purely syntactic,
surface distinction. The only distinction among meanings is the distinction between those
sentential meanings that can only be represented by declarative sentences and those that can
be represented either by suitable declarative sentences (performatives) or by non-declarative
paraphrases thereof. (1972, p. 208)

Lewis’s proposal has been criticized on several grounds. We shall see that although the critics have
mostly missed the mark, there is a valid objection. However, the problem stems from the
unnecessarily Austinian character of the performative approach. Cured of its fixation with speech
acts, the proposal turns out to be just what we want.

Lewis’s critics have concentrated on the claim that explicit performatives are declarative; and in
particular on the consequence that an utterance of an explicit performative must, as McGinn puts it,

be tantamount to the performance of two speech acts, where the (felicitous) performance of
one (saying or asserting) is sufficient for the performance of the other (commanding, etc.) ....
So in uttering a performative one both performs a certain speech act and declares that one
does. (1977, p. 305)

McGinn’s own objection to this consequence of Lewis’s account is that

no such duality is apparent in respect of straight nonindicatives .... ‘I command you to shut
the door’ mentions me and commanding, but ‘Shut the door’ manifestly does not. So they are
simply not equivalent in meaning. (1977, p. 305)

In advance of a semantic analysis of non-indicatives, however, I think this is premature. The

significance of my utterance of ‘Shut the door’ clearly depends in some way on the fact that it was I
who said it (you may decline to obey, for example, for precisely that reason). Is it obvious that this
significance is not the result of an implicit reference to myself? Similarly, one of the effects of this
utterance is surely that you come to believe that I have ordered you to shut the door. Is it clear that
this should not be explained by saying that I have implicitly stated that I have so commanded you?
(Compare: ‘Bring me pig meat’ mentions pigs and meat. ‘Bring me pork’ manifestly does not. So they
are simply not equivalent in meaning.)

The best attempt to make an objection out of the duality of the explicit performative seems to be that
of Stephen Schiffer. Schiffer takes over from Austin the principle that ‘ordinarily one uses the explicit
performative formula to make explicit the precise act one is performing in the issuing of one’s
utterance’ (1972, p. 107). He then points out that if we assume that performatives have assertoric
force, this principle leads to a vicious regress. To make explicit the fact that ‘I order you to shut the
door’ performs an assertoric speech act, we would have to say ‘I assert that I order you to shut the
door.’ And here the same problem occurs again: to make explicit the assertoric force of this new
assertion, we would need a yet longer explicit performative.

Schiffer’s solution is to say that although performatives are constative, they are ordinarily used
without assertoric force. This will not do for Lewis, who wants to claim, in effect, that we give
commands by asserting that we are doing so. If ‘I order you to shut the door’ did not have assertoric
force, you could justify your inactivity with the remark that I had not told you that there was a
command to obey – that you had understood that I was merely entertaining the thought that I was
thereby ordering you to shut the door. But as Kent Bach (1975) points out, a more plausible solution
to Schiffer’s regress is simply to reject Austin’s view that performatives make explicit the precise
nature of the speech act they perform. Instead, we can say that they make explicit one speech act in
performing another.

31
The real difficulty for Lewis’s account seems to me to stem not from the dual role it accords to the
explicit performative, but from the fact that the roles are not independent. Lewis wants a semantic
analysis of non-indicatives. In effect, his account is meant to be a guide to the interpretation of non-
indicative utterances. It tells us, first, that they have the internal structure of a particular kind of
assertion: namely an explicit performative. The structure thus looks like this:

(3.1) I perform the speech act A.

So the account tells us that if we receive a non-indicative linguistic ‘package’, its outer layer is to be
treated as an assertion that we are being offered another package, another speech act. To understand
the original, we have to unpack the package it claims to contain (unless we do so, after all, there will
be nothing to distinguish one non-indicative from another). But what is this new package? Why, the
very one we started with. Offered a command, for example, we are told to treat it as the assertion that
we are being offered a command. Obligingly we refer to our account of commands, to understand
what it is that the assertion says we have been offered. But it is a command, and so the account tells
us that it is equivalent to an explicit performative – an assertion that we have been offered a
command. The chain of interpretation thus has a loop in it.

This is not Schiffer’s objection. Schiffer tries to build a regress outwards, working on the assertoric
part of an explicit performative. This objection goes inwards, working on the speech act that the
assertoric part reports. The problem stems from the attempt to analyse the speech acts performed by
the non-indicatives in terms that refer to these speech acts themselves – that is, as reports of the
performance of these speech acts. As Charles Hamblin puts it,

the so-called reduction ... [is] a logician’s blind alley; ... it is not really a reduction at all. If
someone did not know what an order was, and needed to be told what meaning to attach to a
given one O, it would be no help at all to tell him that it was a locution that, uttered by me to
you, meant

I order you to carry out order O,

since he wouldn’t understand this any better; O remains embedded in it, unanalysed and
unglossed. (1987, p. 135)

The marker theorist can thus object that the universal factualist has not offered a coherent
assertoric analysis of the non-indicatives. An analysis of meaning is required to account for linguistic
understanding. The explicit performative account cannot do this, because the performative refers to
the very linguistic act whose analysis is in question. The factualist must do more than show that for
any speech act conventionally performed by non-indicative means, we can find an indicative utterance
that could be taken to perform the given speech act. Speech acts being conventional, any indicative
utterance (indeed, any public action) can be made to do such additional duty. But the trick requires a
prior conception of the speech act we propose to add to the conventional significance of the chosen
indicative. So it cannot provide an alternative analysis of the speech act concerned. It is as if we had
tried to show that pocket knives can be thought of as potato peelers, by producing an object that
physically combines the two sorts of implement. What we need to show is rather that an arbitrary
pocket knife can already be thought of as a potato peeler (by showing, of course, that it can be
grasped as such).

3.4 SPEECH ACTS AND SPEECH EFFECTS

The problem for the performative approach to an assertoric treatment of non-indicatives is thus that if
performatives are thought of as reporting speech acts, the resulting analysis is viciously self-
referential. The key to a solution, I think, is to appreciate that what such an assertion reports does not
have to be a speech act at all. It is again crucial that we are dealing with conventional acts. We need

32
the fact that if a social state of affairs is actually the conventional effect of an utterance, then it could
instead be taken to be the conventional effect of almost any action (or indeed of other sorts of event:
there are societies in which conventional obligations are taken to be dependent on random physical
processes, for example).

Thus if convention enables a speaker to impose a requirement by saying ‘Please leave the room’, it
could quite well enable the same speaker to impose the same requirement by other means – including,
I think, a simple mental decision that the person concerned shall be required to leave the room. It
might be objected that imposing the requirement in this way would enable people to become subject
to requirements about which they had no way of knowing. This is true, I think, but hardly
problematic. There are other sources of obligations to which individuals may unknowingly be subject.
(The possibility is reflected in the principle that ignorance of the law is no excuse.) In any case, it
could easily be avoided by treating such requirements as conditional on publicity.

We are liable to be misled here by the very view we are attempting to undermine. If we begin with the
idea that a command is a type of speech act – different from, but on a par with, the act of assertion –
then of course we read ‘I command ...’, if as a report at all, as a report of the occurrence of such a
speech act. But if we are not already committed to a non-assertoric view of imperatives, we don’t
have the same reason to think of commands as a category of speech acts. And hence we are free to
think of ‘I command ...’ as a report of a conventionally generated state of affairs: that very state of
affairs that on the orthodox view results from the corresponding imperative speech act.

Looking at things from the orthodox point of view, we can thus separate the effect of a speech act – in
the case of a command, for example, the requirement or obligation it establishes – from the speech act
which normally produces that effect. Not only does the existence of the effect not depend on the
occurrence of such a speech act; but we are free to characterize the effect in terms that do not refer to
such a speech act. And hence we can report the effect, without explicitly (due to our description of the
effect) or implicitly (due to its implied causal history) reporting another speech act. So in place of a
speech act that produces a conventional effect – and in the process, so to speak, alerts an audience to
this effect – we can always imagine an assertion that there has been such an effect, understood as
produced by the speaker by some other conventional means. We thus replace a speech act with a
report, not of the occurrence of the speech act itself, but of the social state that speech act would
normally produce – roughly, by a report of the relevant change that would have been made to an
audience’s social environment by a performance of the original speech act.

We thus aim to intersect the chain of interpretation where recognition of the nature of a speech act
gives way to response. We want to report just those effects of an utterance that a speaker is normally
able to produce at will, without depending on audience compliance (except with respect to the
relevant meaning conventions). Even at this level, of course, there are cases in which a speaker will
fail to produce an intended effect. Lack of authority will vitiate a command, for example. (Austin
1962 describes many ways in which an intended speech act may be ‘unhappy’, as he puts it.) But
these will simply be the cases in which the relevant reports are false.

I think this proposal gives us the indicative paraphrases we have been looking for. We have shown, in
effect, that for each category of non-assertoric speech acts the marker theorist claims to identify, we
could construct a class of indicative utterances that would serve the same linguistic function. These
would be indicatives in the ordinary sense. At worst we would have to coin some new descriptive
terms to refer to the typical effects of the non-indicative speech acts we were concerned to replace.
The marker theorist cannot object to this procedure. Pressed to explain our terminology we can appeal
to the marker theorist’s own account of the social significance of the force distinctions supposedly
marked by the grammatical moods. In distinguishing questions, commands and other non-assertoric
speech acts, the marker theorist incurs an obligation to elucidate these categories. Such an elucidation
is bound to appeal to supposedly distinctive features of the social role of each of these forms of
communication, and in particular to their distinctive effects on conventional relations between hearers

33
and speakers. These effects (or some of them) provide the subject matter for the factualist’s assertoric
paraphrases.

In summary then, the argument shows that if there were non-assertoric speech acts of the kind the
marker view takes to be associated with the non-indicative moods, the function of these speech acts
could quite well be performed by utterances that the marker view would clearly take to be assertoric.
On what grounds then can the marker theorist maintain that non-indicatives are not simply ‘disguised’
assertions?

I emphasize again that the argument does not show that non-indicatives are assertions. It merely
undermines the marker theorist’s grounds for claiming that non-indicatives are not assertions. So it
supports the universal factualist only to extent that it counts against a contrary view. Indeed, I think its
effect is to call into question the enterprise that universal factualism and the marker view have in
common, of seeking to discern the ‘true’ logical form of natural utterances. For it suggests that the
categories both views employ are sufficiently flexible to prevent any authoritative judgement on how
they should be applied to natural language.

3.5 EXPLAINING THE APPEARANCES

The marker theorist’s best hope of a defence would be to find some significant characteristic of non-
indicative speech acts not shared by the claimed indicative paraphrases. It is a slim hope, however.
For one thing, the argument exploits the marker theory itself, in conjunction with a simple and
plausible model of linguistic interpretation. For another, the factualist’s indicative paraphrases seem
to account in one of three ways for the evidence that appears initially to favour the marker view. The
supposed characteristics of the non-indicatives divide into three classes. Some supposedly distinctive
properties of non-indicatives turn out to be shared by their indicative transforms. Some properties
supposedly lacked by the non-indicatives turn out to be applicable after all, in ways determined by the
indicative transforms. And some properties, as we have seen, turn out to be distinctive but merely
grammatical – the kind of thing that the marker theory claims to explain.

All three possibilities are exhibited by the range of characteristics that Dummett takes to distinguish
the non-indicatives. As we saw, Dummett says that assertions are not typically associated with
definite consequences; are the expressions of mental states with inner consequences (so that we can
make sense of an internal assertion – a judgement – but not for example an internal command); can be
used to deceive; and can be used as components of complex sentences. Taking these in order, the
impression that non-indicatives have definite consequences is confirmed and explained by the subject
matter of their claimed indicative para- phrases. These are assertions about the social environment of
a speaker and a recipient – in general, about features of this environment of immediate relevance to
the recipient’ s actions. The indicative substitute for a command, for example, informs its recipient of
a social obligation (or of a state of affairs of which such an obligation is normally an immediate
consequence). So it is hardly surprising that such utterances have consequences which are unusually
definite, by the standards of assertions in general.

The apparent lack of ‘internal’ correlates of non-indicative speech acts falls into the second of the
above categories: the indicative transforms provide such internal correlates. Thus if we think of
commands as assertions of the form ‘You are required to do A’, then their internalizations will take
the form ‘X is required to do A’, where X will normally be a term that refers to someone other than
the speaker. It is true that once this reference is made explicit, we can consider the case in which it is
a self-reference. Externally or internally, judgements of the form ‘I am required to do A’ are perfectly
respectable. Indeed, they have just the form in which the recipient of the indicative substitute for a
command will report his or her newly acquired belief. Whether they have another use, as the
transforms of self-commands, depends on extraneous factors: on whether such requirements can be
self-imposed; and on whether, if so, there is ever a need to report them.

34
Similarly, the indicative transforms provide a sense in which commands can be said to express mental
states with separate internal consequences. If commands express beliefs about requirements, for
example, then such beliefs will have other effects: believing that you are required to complete the
form, I take steps to ensure that if you do not do so (having been properly informed of the
requirement), appropriate measures will be taken. It does not count against this view that the belief
may be the result of my own decision to impose such a requirement, or that the requirement may be
conditional on your actually receiving news of it.

The supposed impossibility of non-indicative deception also falls into the second category. We have
already seen that one can be misled by an insincere command just as by an insincere report of a
corresponding requirement. We noted that the assertoric model can cope with the possibility that
disobedience might be regarded as punishable, irrespective of the commander’s intentions. The basis
of such requirements is bound to be a matter of social convention. One relevant convention might be
that in the absence of explicit indications to the contrary, any utterance of a command or its indicative
substitute shall count as establishing the requirement that makes that substitute true. In limited ways
this sort of convention governs man y assertions. Think, for example, of ‘This is a stick-up’, uttered in
a bank. Deceit is not always easy.

This leaves the suggestion that non-indicatives do not occur as sub-sentential components. As noted
earlier, this is a grammatical restriction, and not therefore a feature of non-indicatives to which the
marker view is entitled to appeal. As such, moreover, it collapses in the face of indicative substitutes
for non-indicative speech acts. For example: ‘If you are required to complete the form, you are
required to do so in black ink.’ The fact that we do not say ‘If complete the form then complete it in
black ink’ is on this view no more than a simple consequence of the rules of non-indicative grammar.
This does not mean that it is insignificant, but that its significance is just that of broader question as to
the reason for the existence of the non-indicative moods.

We shall return briefly to this question in chapter 8. By then we shall have an alternative to the marker
theorist’s claim that the moods are the conventional marks of underlying semantic categories. We
have argued here that such a claim cannot be justified. Indeed, it follows from the argument that the
convention the marker theory calls for would be unworkable. Qua speakers, just as qua theorists, we
lack the mood-independent access to the claimed underlying categories that the convention would
require.

In the main, however, the rest of the book will be concerned with the indicative mood and its internal
structure. The aim of this and the previous chapter has been partly to sanction this concern, by
showing that the issues the non-factualist raises within the indicative mood are not to be swept aside
in the marker theory’s grand account of the moods as a whole. More importantly, however, it has been
to show that there is no refuge in the marker view, in the face of a sceptical challenge to non-
factualism. Such a challenge will be the major task of the next two chapters. We may now appreciate
that what the sceptic challenges the non-factualist to exclude is not merely the marker view. That
outcome, if not as philosophically illuminating as non-factualism is thought to be, would at least be
familiar. Given that the marker theory is itself vulnerable to sceptical challenge, however, the
sceptic’s alternative is not so tame. The threat to non-factualism is rather the bleak possibility that
there is no way to exclude universal factualism – no analysis of statement of fact that entitles us to
draw non-trivial limits to fact-stating discourse. We have seen that for non-factualists, as for marker
theorists, to admit this possibility is to admit defeat. Appealing as it may seem to general semantics,
universal factualism cannot support the distinctive claims of either doctrine. It cannot explain mood in
terms of the limits of statementhood, and it cannot attribute philosophical errors to the misattribution
of statementhood.

35
4
A Retreat to the Head
In this chapter and the next our goal is to embarrass the non-factualist with an abundance of facts.
More precisely, we want to show that the non-factualist’s best arguments are inconclusive, not
excluding factualist interpretations of the topics in question. So again, our intentions are mainly
sceptical: we want to show not that there are more facts than the non-factualist supposes, but that the
non-factualist cannot establish that there are not such facts.

Non-factualism is more a species of philosophical theory than a theory in its own right. Since we
cannot examine every member of the species, I propose to concentrate on three of its major
manifestations: the non-factualist treatments of ethical judgements, probability, and (in chapter 5)
indicative conditionals. Among the advantages of this selection is the fact that on the face of it, they
have little to do with each other. This will add weight to the claim that the striking parallels which
emerge, between probability and ethics in particular, reflect a characteristic pattern of non-factualist
argument.

Broadly, the plan of these two chapters is as follows. In this chapter we want to show that non-
factualists can be forced to depend on a psychological distinction – in effect, on the claim that genuine
statements of fact are the expressions of genuine beliefs. In chapter 5 we shall then challenge their
entitlement to this distinction. We shall argue that beliefs cannot be distinguished from the relevant
non-beliefs without appealing – illicitly, so far as the non-factualist is concerned – either to truth or to
existence of appropriate facts in the world.

First, however, a point of clarification. Such a case against non-factualism may seem superfluous,
given the argument of chapter 3. We criticized the marker view on the grounds that by the standards
of that view, any convention-governed speech act can be construed as an assertion. The marker view
requires that a successful speech act convey, in some sense, the information marked by mood.
Consequently, the marker theorist cannot exclude the possibility that the relevant feature is reported,
rather than merely marked. This argument may seem to count as much against non-factualism as
against the marker theory. If no utterance can be shown not to be an assertion, then in particular
ethical judgements, conditionals, and the like cannot be shown to be non-fact-stating.

However, I think non-factualism fares better than the marker theory. For one thing, it is not clear that
the non-factualist need require that the fact-stating/non-fact-stating distinction be marked in speech,
or accessible in principle to ordinary speakers. If not, then the distinction need not be manifest in the
chain of interpretation, in the way that the marker theory evidently requires. Since the reductio
argument of chapter 3 depends on this feature of the marker view – the sceptic proposes an alternative
chain of interpretation with the same final effect – non-factualism may well be immune.

For another thing, the contents of the assertoric paraphrases on which the reductio relies are
themselves closely based on speech act categories. The idea was to map force distinctions onto a class
of content distinctions (roughly, distinctions within the set of assertions about certain conventionally
generated states of affairs). This means that the taxonomy of non-assertoric speech acts that the non-
factualist claims to find within the indicative mood would survive the adoption of this sort of
universal factualism – albeit on the basis of content distinctions rather than force distinctions. So the
non-factualist’s case could simply be rephrased. The question would be, not whether a moral
judgement should be regarded as an assertion at all, but whether it should be regarded, at face value,
as an assertion about a realm of moral facts. The alternative would depend on the version of moral
non-factualism on offer. It might for example be an assertion to the effect, roughly, that the speaker
has expressed an evaluation of a certain kind.

36
The non-factualist could thus adapt to the possibility of universal factualism, by presenting the issue
in terms of the contents of disputed utterances: granted that moral judgements, like everything else,
are statements of fact, are they really about moral facts, or rather about some utterance-related
conventional state? The non-factualist would still claim to show that certain classes of utterances are
not assertions with the contents they seem at first sight to have. The negative claim will remain
unchanged, though the positive claim will now be that the utterances concerned are assertions with
some other content, rather than that they are not assertions at all. However, the negative thesis is the
basis of the non-factualist’s ‘resolutions’ of philosophical problems in various areas. So the approach
seems essentially intact.

Moreover, our present interest, here and in chapter 5, is in criticizing the non-factualist’s own case for
the negative claim. The issue of what would follow from accepting this thesis will thus be largely
irrelevant. It is true that in principle we might want to argue against the thesis by reductio, and hence
be interested in its consequences. But in fact we shall concentrate our attack on the presuppositions of
the non-factualist’s typical grounds for accepting the thesis. The marker view’s problems are thus of
no immediate relevance. They stand in the background – a warning that we would be unwise to take
the downfall of the non-factualists as a vindication of the marker theory.

4.1 PROBABILITIES AND VALUES: THE NON-FACTUALIST INTERPRETATION

I want to exhibit some similarities between the usual grounds for non-factualism about ethical
judgements and about probability. In particular, I want to show to show that for essentially parallel
reasons, non-factualists in both areas can be forced to appeal to psychology. It will be helpful to begin
with a characterization of what is at issue in each case. These characterizations are far from
comprehensive, and perhaps exaggerate the initial formal similarity between probabilistic and ethical
non-factualism. They will do for our purposes, however; and are not, I think, an artificial source of the
striking parallels we find later on.

Let us begin with the probabilistic case. In English a probabilistic expression often has the syntactical
form of a sentential operator: operators such as ‘It is probable that ...’, ‘There is a 60 per cent chance
that ...’, and ‘Probably, ...’ itself, for example. Let us call a sentence formed by attaching such an
operator to a suitable sub-sentence a simple probability ascription, or SPA. I have argued elsewhere
(Price1983b) that SPAs correspond to the so-called single case applications of probability, much
discussed in the philosophical literature. For the present, however, the relevant question is whether a
SPA should be construed as a statement of fact. Does a probabilistic operator modify the content or
the force of a sentence to which it is attached?

Most philosophical accounts of probability take for granted the factualist answer. ln effect, they are
attempts to characterize the distinctive content of a SPA – to tell us what SPAs are characteristically
about. There are several candidates: class ratios (finite or infinite, actual or hypothetical), logical
relations, propensities, objective chances or probabilities, degrees of rational belief, and even a
speaker’s own degrees of confidence. As the list suggests, there is a wide range of opinion on what
should be expected of an account of probability (and especially on the extent to which one should aim
to reduce that notion to others).

The non-factualist, however, says that all these views are misconceived. SPAs have no common
content. Literally, there is no more something that all sentences of the form ‘Probably P’ are about,
than there is something that all sentences of the form ‘Is it the case that P?’ are about. ‘Probably P’ is
about what P is about. It differs in that whereas an utterance of P is normally an assertion, ‘Probably
P’ is normally a different sort of speech act – what may be labelled a qualified or partial assertion.

As in many cases, the linguistic distinction on which the non-factualist here relies is conveniently
elucidated in psychological terms. It assumes a familiar model of assertion, belief and relevant action:
an assertion that Q is typically an expression of a (full) belief that Q; and such a belief is typically

37
displayed in a person’s acting as if Q whenever, in the light of her desires, she thinks it makes a
difference to the outcome of her action s whether it is the case that Q. The factualist subsumes SPAs
under such a model, regarding them as assertions that express (full) beliefs with some characteristic
content. The non-factualist, on the other hand, takes SPAs to express ‘partial’ beliefs, or degrees of
confidence. ‘Probably P’ is thus the characteristic expression not of the full belief that it is probable
that P, but of a strong partial belief that P. A partial assertion (hence the term) is thus the expression
of a partial belief. Moreover, because full beliefs are the limiting cases of partial beliefs, an ordinary
or ‘full’ assertion is a special sort of partial assertion.

For the moment, not wishing to prejudge the question of the priority of psychological over
metaphysical presentations of non-factualism, we should treat this version as merely expository. In
particular, we want to hold open the possibility that the belief/non-belief distinction might need to be
founded on the availability of relevant domains of facts in the world.

Turning now to the ethical case, we can characterize the issue in similar terms. For present purposes
we can confine our attention to a particularly simple sort of ethical judgement: those of such forms as
‘It is a good thing that ...’, ‘It is disgraceful that ...’, and so on, in which the evaluative expression can
be regarded as a sentential operator. Following the pattern of the probabilistic case, I shall call this a
simple ethical ascription, or SEA.

The competing views are then as follows: the factualist holds that SEAs are assertions, and hence that
ethical sentential operators are content modifiers; the non-factualist that the role of the sentential
operator is rather to signal a non-assertoric force. A psychological elucidation again suggests itself.
The factualist says that SEAs express ethical beliefs – beliefs about moral facts or states of affairs.
Whereas the non-factualist takes SEAs to express some sort of distinctive moral attitude – approval or
disapproval, say – towards the non-moral state of affairs referred to within the scope of the operator
concerned.

Not all the arguments for non-factualism about SPAs run parallel to corresponding arguments about
SEAs. For example, a probabilistic non-factualist would do well to help himself to some of the
objections his factualist opponents raise against one another. Thus, he might well endorse the standard
criticism of (most) frequency accounts of probability, concerning single case probabilities (the so
called ‘problem of the reference class’ – see Mellor 1971, chapter 3, for example).

Similarly, the non-factualist will do well to concur with several serious objections to naïve
subjectivism – the view that SPAs report a speaker’s own degrees of confidence. One such objection
is the phenomenological one: in considering the probability of snow this afternoon, I don’t consider
my own degree of belief in snow this afternoon. How could I, since the purpose of deliberation is to
lead me to some such degree of belief? Another objection points to ordinary disagreements about
SPAs. Clearly these are not disagreements about what each of the participants currently believes. The
argument gets its point from the fact that the participants both know what they each believe, and
therefore know that they disagree. (In this case, clearly, there is a parallel argument against ethical
subjectivism.)

However, arguments of this sort merely encourage the factualist to retreat to a stronger position. The
strongest is the simplest: renouncing the attempt to provide a reductive interpretation of SPAs, it
settles on the claim that SPAs are about, simply, probabilities. Probabilities are objective features of
world, and hence a respectable subject matter for statements of fact.

It is easy to imagine a corresponding factualist position in the ethical case. And in the face of
doctrines as bare as these, the non-factualist’s options are rather limited. There is little for an
objection to latch on to – and little, similarly, to distinguish the probabilistic and ethical cases. I think
that in each case the non-factualist has two main avenues to explore. The first develops a
metaphysical argument, while the second appeals to the functional role of ethical or probabilistic

38
judgement in our lives. I want to show that the probabilistic and ethical cases do indeed run parallel
here, and that in both cases both avenues lead to the head.

4.2 ODD FACTS: THE ARGUMENT FROM QUEERNESS

In the ethical case, more perhaps than in any other, non-factualists have been prepared to appeal to the
apparent peculiarity of facts of the disputed kind: ‘If there were objective values, then they would be
entities or qualities of a very strange sort, utterly different from anything else in the universe’ (Mackie
1977, p. 38). Mackie calls this ‘the argument from queerness’. It is often linked to an epistemological
point. Non-factualists argue that if there were such facts, then ‘if we were aware of them, it would
have to be by some special faculty of moral perception or intuition, utterly different from our ordinary
ways of knowing everything else’ (Mackie 1977, p. 38).

As we shall see, the non-factualist’s metaphysical case does not rest here. And just as well, for there is
nothing here to trouble a committed factualist. With a touch of irony, the factualist can object that
queerness is in the eye of the beholder. Moral facts seem queer to a non-factualist because he already
has a view of the world that excludes them. It is true that they differ from say physical facts; but this
is no more than a truism. Facts about different sorts of things are bound to differ – for how else could
they be about different sorts of things?

This leaves the epistemological point. But here the factualist will point out that her opponent agrees
that we do not make ethical judgements (or whatever sort of judgement is at issue) at random. On any
account, there is some process that leads us to make some such judgements and not others. Why not
regard this process, whatever it is, as the so-called moral intuition? Once again (the factualist will say)
the non-factualist’s mistake is to start with some preconceived notion – this time of what a belief-
forming process should be like – and then to dismiss the moral case on the grounds that it does not
conform to this model.

The non-factualist may feel that these replies do not do justice to the argument from queerness and its
epistemological shadow. But I think the feeling stems from the fact that the non-factualist has a
stronger metaphysical argument in mind. The trouble with moral facts, the non-factualist suspects, is
not that they are odd but that they are unnecessary. Indeed, we are entitled to regard them as odd
because they are unnecessary. The non-factualist thinks that we can explain our talk of values on a
much sparser metaphysical basis.

Similarly in the probabilistic case. Here the argument from queerness consists in an appeal to a
conception of the world in which there seems no place for objective probabilities. The non-factualist
tells us that the world is the realm of the actual, the domain of physical things. Everything is
determinate, every point either occupied or unoccupied, every particle in its place, and every
proposition either true or false. Where, then, is the room for probability?

As it stands, this is no more convincing than its ethical counterpart. True, it presents non-factualism as
a corollary of an attractive Humean metaphysical minimalism. But the blessed poor notwithstanding,
economy is a not a virtue in itself; it needs to keep company with sufficiency. So a proper appeal to
metaphysical minimalism is at best one side of a more complex argument. The non-factualist has to
show that probabilistic facts are unnecessary – that we can account for our talk of probability without
construing it as talk about a distinctive feature of the world. Only thus will non-factualism be shown
to have the conjunctive virtue of economical sufficiency.

4.3 PROJECTIVISM

Like the metaphysical position it serves, the argument for sufficiency is attributable to Hume. It turns
on the claim that in addition to belief itself, there are other kinds of attitude one may hold to things in
the world. There may be partial beliefs, and perhaps valuations and expectations of various kinds.

39
However, we are inclined to treat these attitudes as if they were genuine beliefs. We thus mistake for a
difference in the world what is really a difference in the head. Such a mistake may have pragmatic
advantages – for example, it may enable us to improve our attitudes by reasoning and argument
‘about the facts’. But metaphysically speaking, it is a mistake all the same: the ‘facts’ we seem to find
in the world about probabilities (or whatever) are actually the images or ‘projections’ of our own
attitudes. We project these attitudes onto the world, and hence come to see it as containing
corresponding realms of facts.

Perhaps the leading contemporary proponent of this Humean approach is Simon Blackburn, who calls
the view ‘projectivism’. He emphasizes its economical advantages:

The projective theory intends to ask no more from the world than what we know is there – the
ordinary features of things on the basis of which we make decisions about them, like or
dislike them, fear them and avoid them, desire them and seek them out. It asks no more than
this: a natural world, and patterns of reaction to it. By contrast a theory assimilating moral
understanding to perception demands more of the world. Perception is a causal process: we
perceive those features of things which are responsible for our experiences. It is
uneconomical to postulate both a feature of things (the values they have) and a mechanism
(intuition) by which we are happily aware of it. (1984, p. 182)

Projectivism thus claims much the same advantage as buying furniture in kit form. By doing some of
the work ourselves, we reduce our initial commitment – financial or ontological, as the case may be.
But in both cases it is possible to overdo things. If a tree is a furniture kit then kit builders become
cabinet-makers. And unless the projectivist can distinguish the real beliefs from the ‘patterns of
reaction’, then projectivism becomes a species of idealism – the view that all ‘facts’ are projections of
our reactions to the world. This outcome obliterates the distinction at the very core of the non-
factualist’s thesis. In the process, moreover, it overturns the projectivist’s claim to metaphysical
economy. The idealist’s world can be as cluttered as anybody else’s: idealists differ from realists not
so much as to what exists, as to what its existence amounts to.

To avoid the slide to idealism, the projectivist thus needs a core of genuine facts: a secure foundation,
on which the explanation of higher-level appearances can be constructed. Blackburn talks of asking
‘no more from the world than we know is there’. But how do we know what is really ‘there’, and what
we put there? This issue has an important corollary: it means that projectivism itself is no guide to the
facts. To arrest the slide to idealism, the projectivist needs to distinguish the genuine facts from the
realm of ‘apparent’ facts for which the projective account is appropriate. So the possibility of a
projective account of an area of discourse cannot in itself provide the motive for a non-factualist
construal of that topic. To be sure, the projective account pays off a debt that the non-factualist incurs
in urging such an account: it meets the obligation to explain the appearances. But the net effect is to
balance the books. There is no profit in projective explanation.

The projectivist might object that this ignores the cost of the raw materials. Since Hume, projectivists
have been motivated by metaphysical economy – by the suggestion that if we take ontological costs
into account, then projectivism comes out ahead. But we can now see that the projectivist’s deal is too
good to be true. If we can choose our raw materials, and less is better, then the best choice is none at
all: that is, idealism. But idealism defeats non-factualism, so the projectivist cannot really accept that
raw materials are a matter of choice. In other words, the economy projectivism offers us is ours
already – or so the honest projectivist must believe. Like a good accountant, projectivism may show
that we are better off than we thought, but it cannot take the credit for making us so.

What then would it take to arrest the slide to idealism? Have we perhaps underestimated the role of
other metaphysical considerations, such as the argument from queerness? We shall see below that
Blackburn himself offers another metaphysical argument for a non-factualist interpretation of moral

40
judgement. These arguments bear on the wrong side of the issue, however. They lever the projectivist
cart into motion, by rejecting factualist accounts of moral usage. But the non-factualist’s problem is to
stop the cart, before it bears him to idealism. He needs an argument to the effect that certain topics are
factual – and the better projectivism is made out to be at explaining factual appearances, the less
likely it is that there can be any such argument.

The one live possibility seems to be to appeal to psychology. If the non-factualist could distinguish
the genuine beliefs – in the head, as it were – from the various attitudes from which we project, then
the barrier to idealism would be readily apparent. Projectivism would be safely contained, its task
being to explain the fact that some non-beliefs are treated like beliefs. We shall return to the question
of the availability of such a psychological distinction in chapter 5. But even if the non-factualist is
entitled to such a distinction, the fact remains that projectivism provides no grounds for non-
factualism. Projectivism is powerful medicine. Unless the need that it serves is grounded elsewhere,
however, its effect is to make us idealists – which for the non-factualist counts as an overdose.

Let us turn then to the second major strand in the contemporary case for ethical and probabilistic non-
factualism. This line of argument has a more overt link with psychology than the supposedly
metaphysical arguments we have just discussed. Indeed, it might be thought to provide the
psychological basis the projectivist seems to require. It focuses on the distinctive functional roles of
probabilistic and moral judgements in our mental and behavioural lives – their significance in
thought, speech and action. The notion of functional role might seem to provide the grounds for a
distinction between beliefs and other propositional attitudes. I want to show that in appealing to
aspects of the use of probabilistic and ethical judgements, non-factualists have taken for granted that
the required psychological distinction can be characterized in these terms.

4.4 FUNCTIONAL ROLE (I): PROBABILITY

Probability provides an important mode of qualification of ordinary empirical judgements. Empirical

evidence is rarely conclusive. Ordinary judgements are rarely felt to be certainly true, to be justified
beyond a shadow of a doubt. An interpretation of probabilistic sentential operators needs to be
sensitive to the fact that such operators are used to register this feature of judgement, to indicate the
degree to which a judgement is felt to be justified. True, we often make judgements with no such
qualification, even if we are less than certain that they are true. But we are quick to forgo this
convenience if the circumstances call for precision – if there is a lot at stake, say. Often then we resort
to SPAs.

The fact that probability plays this role raises two sorts of problem for the factualist. For one thing it
threatens a vicious regress: if probabilistic qualification is a more or less universal feature of accurate
factual judgements, and probabilistic judgements are themselves factual, then accuracy seems to
demand that in place of ‘It will rain’ we say ‘It will probably rain’; that in place of that we should say
‘It is probable that it will probably rain’; and so on. The regress is vicious because an interest in
accuracy – an interest in asserting only what we are justified in asserting – will prevent us from
resting content with any particular stage.

The factualist has a couple of options here. For a start, she could say that probabilistic judgements are
a special case, and can be justified with certainty. The main problem with this is that, notoriously, the
statistical rules of inference that are held to justify probabilistic conclusions are themselves
probabilistic. A given range of frequency data will make it likely, but certainly not certain, that a
particular probability has a specified value. (This is the source of concern about the circularity of
statistical inference; see for example Braithwaite 1953, chapters 5 and 6.)

A much more promising suggestion relies on what we may call the limiting case principle: the
principle that as the probability concerned becomes sufficiently high, a probabilistic judgement
becomes ‘equivalent’ in some useful sense to the corresponding unqualified judgement. Given this

41
principle, and the fact that in practice our first level probability judgements are usually themselves
justified to a sufficiently high probability to allow us to apply it, it will follow that in practice we
rarely introduce second (or higher) level probabilities.

The limiting case principle is so natural that it is easy to overlook its importance, and to fail to notice
that for the factualist its justification is far from obvious – indeed, its justification is the second and
more fundamental difficulty that arises, for the factualist, from the role of probability as a qualifier of
ordinary empirical judgements. Clearly there is some important connection between the meanings of
‘It will rain’ and ‘It will probably rain’. Yet for the factualist these statements have different contents.
The problem is to explain the connection between these contents, in such a way as to account for the
limiting case principle, and hence make sense of the use of probabilistic qualifiers of ordinary
judgements.

The non-factualist might explain these things as follows: probabilistic judgements express degrees of
confidence, or degrees of partial belief. So ‘It will rain’ and ‘It will probably rain’ simply express
different degrees of belief in the same proposition. And a series of judgements of increasingly strong
probability that P, with limit 1, is thus associated with a series of increasingly strong partial beliefs
that P, whose limit is the full belief that P. A judgement of high probability is practically ‘equivalent’
to the corresponding unqualified judgement in the sense that, and in so far as, a high degree of
confidence in a given matter is practically equivalent to certainty. This explanation turns on the notion
of the degree of a belief. Hence it is useless to the factualist, who is committed to explaining the use
of probabilistic operators in terms not of differences of degree but of differences in content. The
factualist seems to require an (increasing) similarity of content between the belief(s) that it is
(increasingly) probable that P and the belief that P.

This problem stems from the role of probability in judgement. A related problem arises from its role
in decision making. Factualists and non-factualists agree that probabilistic judgements have certain
typical behavioural manifestations, at least in certain straightforward cases, such as the choice of
betting odds. For present purposes we can assume that both sides accept that Bayesian decision theory
gives an approximate even if very idealized description of this process. The non-factualist again has a
ready explanation: this behaviour is simply the characteristic (non-linguistic) behavioural display of
partial belief. What Bayesian decision theory describes, in effect, is the functional role of a partial
belief, in terms of its content and degree. (The more familiar story about the functional role of a full
belief falls out as a special case.)

For the factualist, however, there is a problem: Why should the full belief that it is probable that P
have the functional role that the non-factualist accords to a partial belief? Why should beliefs about
probability be reflected in betting behaviour, for example? After all, the factualist will expect another
sort of manifestation of the belief that it is probable that P: the sort of behaviour standardly associated
with a full belief, given in this case a content of the form ‘It is probable that P’.

On the face of it, the factualist’s problem is thus to explain why probabilistic judgements should be
accompanied by ‘corresponding’ degrees of confidence: for example, to explain why the belief that it
is probably going to rain should be associated with a high degree of confidence, or strong partial
belief, that it is going to rain. I shall call this the confidence problem. Note that a solution to the
confidence problem would not only allow the factualist to explain the practical manifestations of
probabilistic beliefs; it would also justify the limiting case principle, and hence explain probability’s
role as a qualifier of empirical judgements.

Why should someone who believes that it is probable that P be confident that P? Factualists who
notice the difficulty are inclined to appeal to rationality at this point. (See. for example, Braithwaite
1966, a paper entitled ‘Why is it reasonable to base a betting rate upon an estimate of chance?’) There
are two difficulties with this move. First, the notion of rationality it depends on is problematic, to say
the least. Its natural interpretations either themselves invoke probability, or cognate notions –

42
‘rational’ means ‘most likely to be successful’, for example – or they invoke some notion such as
hypothetical limiting frequency, already invoked by certain of the more reductive factualist accounts
of probability. Either way, the confidence problem remains: why should a belief about probability, or
about hypothetical long run frequency, have a bearing on degrees of confidence, and hence on action
in the single case. It is doubtful whether an account of the relevant notion of rationality can escape
circularity of one of these two kinds. Notoriously, deductive inferences from probabilistic premisses
yield probabilistic conclusions; while the non-deductive inference to degrees of confidence is what
needs justification.

Secondly, it is far from clear that the appeal to rationality makes the connection strong enough to do
justice to our intuitions. Faced with subjects who regularly failed to match their degrees of
confidence, and hence their actions, to their claimed estimates of ‘probability’, I think that we would
be inclined to say not that they were irrational, but that they did not mean what we do by ‘probability’
and related terms.

These factors have led some writers (e.g., Mellor 1971, p. 67) to the view that the inference from a
full belief about a probability to the corresponding partial belief – let us call it the downward
inference – is in some sense analytic. The simplest such account, taking to heart the intuition on
which the earlier suggestion was based, is what may be called the rationalist proposal: probabilities
simply are degrees of rational belief, so that ‘It is probable that P’ simply means ‘It is reasonable to be
confident that P.’ There is then no question of believing that it is probable that P, and yet doubting
that it is reasonable to be confident that P.

However, it is important to realize that there is still a question as to why people who believe that it is
reasonable to be confident that P are (on the whole) actually confident that P. What is it about the
content of this belief (about what it is rational to believe) that leads people to adopt the corresponding
degree of confidence? Of course, this is simply the original confidence problem, dressed up in new
clothes – clothes it acquires in the rationalist’s paraphrase of a SPA.

At this point I think the rationalist has two possible moves. One is to attempt to explicate the notion of
rationality, so as to display a decision procedure that justifies the (newly construed) downward
inference, and in which ordinary self-interested speakers can plausibly be held to indulge. But this
faces the very problems that, as we saw, plague attempts to show that the belief that it is probable that
P makes it reasonable to be confident that P.

As in that case, the objection is perhaps not conclusive. But it does encourage another approach: to
say that the use of the downward rule is constitutive of a grasp of the meaning of the relevant
rationality ascriptions. A person will thus be said not to know what ‘It is reasonable to be confident
that P’ means, unless in general when they profess to believe it, they are confident that P. Let us call
this contextual rationalism: it takes the meaning of the term ‘rational’ to be bestowed by its inferential
context.

The rationalist’s appeal to rationality in accounting for the meaning of SPAs may now seem
redundant. Why not simply be a contextualist about probability itself – saying that the habit of making
the downward inference is (partially, at any rate) constitutive of a grasp of the meaning of SPAs
themselves – without making a detour via ascriptions of rationality? Either way, however, it can be
seen that the confidence problem, concerning the typical use of the factualist’s full beliefs about
probabilities, provides a strong motive for a particular account of their content: it favours the
contextual theory. In effect, the factualist has been forced to retreat from ambitious reductive accounts
of the meaning of SPAs, to the more modest and hence less exposed contextual position.

At this point, it may seem that the factualist has really conceded everything that was originally
distinctive about her position. The factualist claims to take SPAs to express full beliefs about
probabilities; and in response to the confidence problem, has taken pains to establish that in virtue of

43
the content of beliefs about probabilities, such beliefs are typically accompanied by the corresponding
partial beliefs. In effect, then, the factualist has acknowledged that the use of SPAs – indeed, any
manifestation of this sort of single case belief about probabilities – is also an indication of a speaker’s
possession of a corresponding partial belief. In consequence, the factualist seems vulnerable to a
charge of theoretical redundancy. Beliefs about probability now seem idle cogs. Except in the
factualist’s account of probabilistic assertion, all the functional work is done by the corresponding
partial beliefs. So why not abandon that account of assertion? Why not acknowledge that SPAs are
products of partial beliefs?

The factualist thus appears to face a dilemma. If there are full beliefs about probabilities, then the
question arises as to their connection with degrees of confidence. If the connection is too loose, then
beliefs about probabilities will not have the behavioural significance that everyone is agreed that they
should have, if there are such things. If it is too strong, then full beliefs about probabilities come to
seem redundant – dispensable in favour of their constant companions, the corresponding degrees of
confidence.

However, the factualist has one more card to play. The non-factualist’s attack has rested on the
distinction between the full belief that it is probable that P, on the one hand, and the attitude of strong
partial belief that P, on the other. But why shouldn’t the factualist simply reject this distinction,
claiming that these two mental attitudes are one and the same – that a belief that it is probable that P
just is a strong degree of confidence that P? Factualism would thereby take on the explanatory
advantages of non-factualism, without an embarrassing duplication of psychological attitudes.

In other words, this ‘radical’ factualism challenges the psychological distinction on which the non-
factualist has been relying. The non-factualist charges that factualist interpretations of probability are
unable to make adequate sense of the functional role of probabilistic judgement – the role of what the
non-factualist calls partial belief. The radical factualist responds by helping herself to the non-
factualist’s functional account; but insisting that it be regarded as a characterization of the distinctive
content of probabilistic beliefs. The prospects for non-factualism again turn on its ability to
distinguish genuine beliefs from other psychological attitudes.

4.5 FUNCTIONAL ROLE (II): VALUES

The confidence problem has an analogue in the ethical case. Indeed, it is perhaps the most venerable
argument for a non-factualist treatment of moral judgements, tracing its origins at least to Hume. It
relies on the principle that, as Hume puts it, ‘reason alone can never be a motive to any action of the
will; ... it can never oppose passion in the direction of the will’ (Treatise, Book II, Part III, Section III;
p. 413 in Hume 1978). In more modern terms, Hume’s view is thus that beliefs by themselves are
incapable of motivating action. Beliefs need to be supplemented by some desire or ‘pro-attitude’.
(Since precision won’t matter, and I have a con-attitude to ‘pro-attitude’, I’ll continue to talk of
‘desires’.)

The non-factualist wants to show that SEAs do not express beliefs. Given Hume’s principle, it is
enough to show that the attitudes expressed in moral judgements are capable of motivating action; for
this puts them on the desire or non-belief side of Hume’s distinction. And the motivating character of
moral judgement (or indeed of evaluative judgement in general) has usually seemed self-evident.

The non-factualist thus offers the following argument: moral commitments – the attitudes expressed
in moral judgements – have a distinctive conceptual role. Briefly, they motivate actions. Hume’s
principle tells us that no belief, of itself, could fulfil this role. And hence the factualist, who takes
SEAs to express beliefs, has to resort to the claim that such beliefs are ‘accompanied by’ attitudes that
can play this role. The factualist faces the moral equivalent of the confidence problem – we might call
it the approval problem – that of explaining, for example, why the belief that it is a good thing that it
is raining should be associated with an attitude of approval towards the fact that it is raining. On the

44
face of it, the non-factualist will suggest, it would seem that ‘if moral commitments express beliefs
that certain truth conditions are met, then they could apparently co-exist with any kind of attitude to
things meeting the truth conditions’ (Blackburn 1984, p. 188).

Hume’s principle is clearly crucial. Someone who rejects it can agree that moral judgements express
mental states on the active or motivating side, and yet insist that moral commitments simply are
beliefs about moral facts. This will be the ethical equivalent of the radical factualism we have just
encountered in the probabilistic case. In both cases the non-factualist has attempted to burden the
factualist with a cumbersome and implausible duplication in the mental states associated with
utterances whose status is in question (SPAs and SEAs, respectively). The radical factualist responds
by accepting the model but denying that it involves any duplication: rather, she suggests, the non-
factualist has confused the existence of two ways of describing a single mental state for the existence
of two distinct mental states.

We shall return to Hume’s principle in chapter 5. Before that, however, I want to show the relevance
of these issues of functional role to another prominent metaphysical argument for ethical non-
factualism. I want to show that the argument cuts no ice with the radical factualist; and also,
incidentally, that it has an analogue in the probabilistic case.

4.6 THE SUPERVENIENCE ARGUMENT

Moral properties appear to supervene on natural or non-moral properties: it seems uncontroversial that
if two things are taken to be the same in all non-moral respects, then they should not be distinguished
in moral terms. The problem is to explain this evident constraint on moral judgement. Simon
Blackburn has argued that only a non-factualist has an adequate explanation. I want to show that the
radical factualist does just as well.

Blackburn first presents the argument in a paper entitled ‘Moral realism’ (1971). He returns to it,
changing the emphasis to some extent, in (1984) and (1985). Blackburn says, for example, that the
explanation of supervenience is

especially hard for the realist [our factualist]. For he has the conception of an actual [moral]
state of affairs, which might or might not distribute in a particular way across the [non-moral]
states. Supervenience then becomes a mysterious fact, and one which he will have no
explanation of (or no right to rely upon). (1984, p. 185)

As far as realism can show us, it could be true that the moral floats quite free of the natural.
(1984, p. 221)

Blackburn wants to show that the non-factualist (or anti-realist) can explain the supervenience of the
moral on the non-moral. In his first version of the argument he suggests that a

natural way of explaining the supervenience of moral properties would, for an anti-realist, be
something like this. There can be no question that we often choose, admire, commend, desire,
objects because of their naturalistic properties. Now it is not possible to hold an attitude to a
thing because of its possessing certain properties and, at the same time, not hold that attitude
to a thing which is believed to have the same properties. The non-existence of the attitude in
the second case shows that it is not because of the shared properties that I hold it in the first
case. (1971, p. 114)

Blackburn acknowledges that the factualist ‘need not refrain from talking about moral attitudes’: on
the contrary, ‘it is just that [the factualist] thinks this is a less clear alternative to talking about moral

45
beliefs’ (1971, p. 115). But he thinks that ‘if we rephrase the preceding paragraph in terms of moral
belief, it is obviously insufficient to explain supervenience’ (p. 115). For the claim would be that

it is not possible to hold a moral belief about a thing, believe a second to be exactly alike in
all naturalistic respects, yet at the same time not hold the belief about the other thing. But this
doesn’t explain supervenience at all: it merely shows the realist putting conditions upon what
can be believed to be the truth, not upon what is the truth. Our belief, he is saying, has to be
consistent across naturalistic similarities – but this is no explanation of why, on his theory, the
truth has to be. (p. 115)

Blackburn’s claim thus seems to be that the factualist has trouble justifying moral consistency –
justifying the policy of applying the same moral predicates in the same non-moral circumstances. The
non-factualist is better placed not because he can justify such a policy, but because he doesn’t need to.
The need for justification stems from the factualist’s conception of truth – in particular, from the
principle that the truth is independent of what it is possible for us to believe. Both sides have a more
or less causa l explanation of the fact that we follow the consistency principle: if moral attitudes are
the effects of non-moral attitudes, consistency in practice is a consequence of the relevant causal laws.
For the factualist, however, de facto fidelity is not enough. Consistency needs to be earned.

So the non-factualist escapes the supervenience problem – as construed in terms of consistency, at

least – by rejecting the framework within which it arises. The problem arises because we talk of
independent moral truths; and the non-factualist is free to reject that sort of talk.

Not all non-factualists will want to exercise this freedom, however. In particular the projectivist, far
from rejecting factual talk, wants to account for it in non-factual terms. Blackburn repeatedly stresses
this aspect of projectivism, regarding its explanatory potential as a major virtue. In particular, he
thinks that the projectivist can explain our entitlement to doubt our own opinions, and to distinguish
between our believing that something is the case and its actually being the case (Blackburn 1984, p.
219, for example). If so, then the projectivist is equally vulnerable to the charge that the causal
explanation of consistency simply puts ‘conditions on what can be believed to be the truth, not upon
what is the truth’.

Whether the projectivist has an answer to this charge will no doubt depend on the details of the
constructive account of truth, in terms of which he proposes to explain the factual appearances. But if
the projectivist can secure supervenience by this means then what is to stop a factualist from doing the
same? In the (1971) version of the argument, Blackburn takes it for granted that factualism arrives
with its own account of truth. His target is the traditional realist, whose notion of truth is anchored to
a conception of a ‘real’ external world. For present purposes, however – as to a large extent in
Blackburn (1984) – the projectivist’s more threatening opponent is the idealist. As noted earlier,
idealism should here be counted a version of factualism; for it rejects the non-factualist’s attempt to
divide indicative discourse into fact-stating and non-fact-stating parts. And a factualist of this sort will
have no qualms about helping herself to the projectivist’s account of moral supervenience. Thus we
find once again that the non-factualist’s main problem is to stop the projectivist bandwagon – and that
the better projectivism starts, the harder it is to stop.

Perhaps for this reason, the consistency issue gets less attention in Blackburn’s more recent
presentations of the supervenience argument. Of course, no version of the argument can hope to
exclude idealism. As an argument for non-factualism, it simply bears on the wrong side of the issue.
But at least the damage might be controlled: we might hope to present the argument in such a way
that it favours projectivism about ethics without equally favouring projectivism about everything.
Alternatively (though this would be less to Blackburn’s own tastes) we might hope for a version of
the argument that would favour a non-projectivist non-factualism about ethical discourse. It turns out,
however, that the prospects of either course depend on our ability to refute the radical factualist.

46
Blackburn now offers the projectivist the following explanation of moral supervenience:

When we announce the [moral] commitments we are projecting, we are neither reacting to a
given distribution of [moral] properties, nor speculating about one. So the supervenience can
be explained in terms of the constraints upon proper projection. Our purpose in projecting
value predicates may demand that we respect supervenience. If we allowed ourselves a
system (shmoralizing) which was like ordinary evaluative practice, but subject to no such
constraint, then it would allow us to treat naturally identical cases in morally different ways.
This could be good shmoralizing. But that would unfit shmoralizing from being any kind of
guide to practical decision-making (a thing could be properly deemed shbetter than another
although it shared with it all the features relevant to choice or desirability). (1984, p. 186)

This might be taken as a pragmatic justification of the consistency principle – as if Blackburn were
suggesting that in view of the role of moral judgements in influencing our practical decisions, it
would be unwise, though not impossible, not to make one’s moral judgements in accordance with that
principle. But there is a more interesting and I think more accurate interpretation. We have noted that
moral beliefs, indeed evaluative beliefs in general, appear to play a special role in guiding choice. A
(positive) value judgement about a state of affairs is a prima facie reason for choosing that state of
affairs over others not so valued. As Blackburn says, ‘It seems to be a conceptual truth that to regard
something as good is to feel a pull towards promoting or choosing it, or towards wanting other people
to feel the pull towards promoting or choosing it’ (1984, p. 188). To value A higher than B is to be
disposed, other things being equal, to choose A in preference to B.

In order to explain supervenience, we need to add to this the plausible claim that in so far as they
affect us, the outcomes of actions are completely determined by their natural or non-evaluative
consequences. The value judgement guides our choice of destination, but the success or failure of the
expedition depends on what we actually find when we get there. (I am simplifying here. In practice,
the route from valuation to natural consequences of choice may be multi-staged: as, for example,
when we judge that it is better to take the high road than the low road, because the scenery is more
attractive on the high road.) This means that to imagine a case of moral difference without natural
difference would be to imagine on the one hand that choice A would be preferable to choice B; but on
the other that since the two choices have the same natural consequences, it would make no difference
which we choose. This conceptual conflict seems a plausible basis for our belief in moral
supervenience – indeed, for the supervenience of the evaluative on the non-evaluative in general.

If this explanation is on the right lines, then the supervenience of the evaluative on the non-evaluative
depends on the principle that outcomes are judged, in the end, by their natural consequences. This
principle might conceivably be challenged. On theological grounds, for example, someone might say
that the outcome of an action can have an intrinsic value that is independent of its natural
consequences (at least in this world). But this sort of challenge in fact works in favour of the above
explanation. To admit intrinsic values of this kind would surely be to deny supervenience – to allow
values to ‘float free’ of the natural, in Blackburn’s phrase. However, the above explanation of
supervenience turns on nothing more than the action-guiding role of evaluative attitudes. We already
know that the factualist has a problem explaining this – it is the approval problem, the evaluative
analogue of the confidence problem. But we saw that if the factualist is prepared to claim that
evaluative attitudes simply are beliefs, the problem seems to be dealt with. And in the present context,
the same identification will entitle the factualist to this explanation of evaluative supervenience. So it
seems that the supervenience problem is not an objection to the radical factualism we have
encountered in this chapter: a factualism which accepts the non-factualist’s positive account of the
mental states characteristically expressed in evaluative judgements, but insists that the mental states
so described are themselves the beliefs whose existence the projectivist denies.

In his most recent discussion of the supervenience argument, Blackburn says that the explanation of
supervenience depends

47
 crucially upon the role of moralising being to guide desires and choices amongst the natural
features of the world. [But] if, as a realist ought to say, its role is to describe further, moral
aspects of [reality], there is no explanation at all of why it is constitutive of competence as a
moralist to obey the [supervenience] constraint. (1985, pp. 56- 7)

Our radical factualist agrees that the explanation of supervenience lies in the fact that moralizing has
the former role; agrees also that it has the latter role; but denies that the two roles are incompatible.

*4.7 SUPERVENIENT PROBABILITY

Blackburn presents the supervenience argument as a problem for the moral realist. In view however of
the parallels we have already noted between the ethical and probabilistic cases, it is natural to wonder
whether there an analogous argument in the probabilistic case. Do probabilities supervene on non-
probabilistic characteristics of things? I want to show that there is a parallel here, that again the best
explanation turns on the action-guiding role of the relevant sort of attitude, but that here also the
radical factualist has an answer.

Suppose then that we accept that there is a certain probability of a specified outcome to a given event
– let us say an n% chance that a given experiment will result in the emission of an electron. Do (or
should) we allow that, with all non-probabilistic facts held fixed, there could have been some different
chance of this result? The answer depends in part on our view of chance. If we think of the chance in
terms of the actual frequency of such outcomes in similar experiments, or some such, the appropriate
answer seems to be ‘yes’. If we are talking about actual class ratios in finite classes, we have no
difficulty accepting that those ratios could have been different, even if other aspects of the world had
been the same. The possibility is illustrated by the man who broke the bank at wherever it was:
because his fortunes depended on indeterminate events, he could equally well have lost his shirt.

Partly for this reason, however, actual frequencies have never provided a plausible approach to single
case probabilities. It is more natural to suppose that the chance that an experiment will have a certain
result is a further fact, over and above the facts about frequencies in actual similar cases. Could then
this further fact vary independently of the other properties of the experiment – including, now, the
actual relative frequencies themselves?

I think that we are inclined to deny that chances could vary in this way. The reason is nicely
analogous to the above explanation of the supervenience of the moral on the non-moral. Suppose for
example that two experimental set-ups are identical in (what we take to be) relevant non-probabilistic
respects; and, most importantly, that they yield the same relative frequencies of corresponding
outcomes. Let us try to imagine that the two set-ups in fact have different chances of some outcome –
electron emission, say. In set-up A there is a high chance of emission; in set-up B, a low chance. The
problem with such a supposition stems from the role of probabilistic beliefs in decision making.
Roughly speaking, the higher the probability of an outcome, the more reasonable it is to bet on it, at
any given odds. This means that on the given assumptions, we shall be inclined to prefer a bet on
electron emission in set-up A to the same bet in set-up B. However, since there is no difference in the
relative frequency of such an outcome in the two cases, a gambler who consistently bets on electron
emission will do no better in case A than in case B. So there is no reason to prefer one bet to the other.
This contradiction seems to show that chances do supervene – in the same sense as in the ethical case
– on non-chance factors.

As it stands, however, the argument is vulnerable to a serious objection: namely that the fact that the
two cases show the same relative frequency of electron emission does not entail that the same betting
policy will have the same outcome – let alone that a single bet would have the same outcome in both
cases. For individual outcomes may differ, even if the overall frequencies are the same. Of course, if
we could take for granted that identical relative frequencies entail identical probabilities, then we
could say that the probability of success is the same for both bets (either in the single case or in the

48
long run). But to take this for granted is to assume what we are trying to establish: the supervenience
of probabilities on the non-probabilistic.

To escape this objection we have to strengthen our premisses. We need to suppose that set-ups A and
B are non-probabilistically identical in such a way as to justify the following claim: Any trial on
either set-up would have had the same outcome if performed on the other. It then follows that in any
given case (and at any given odds), a bet on electron emission would have the same result on either
set-up. And this conflicts with our preference for a bet on A, stemming from the assumption that there
is a higher chance of electron emission in A than in B.

Thus, on sufficiently strong assumptions, we find that single case probabilities do supervene on non-
probabilistic features of the world. As in the moral case, the argument makes essential use of a special
association between a certain sort of mental attitude – here (what the factualist will regard as) a belief
with probabilistic content – and choice behaviour. We saw that this association may itself seem
problematic for the factualist: it is the basis of the confidence problem. However, the radical
factualist’s reply was that the so-called ‘association’ is actually an identity. We now find that in the
probabilistic as in the ethical case, such a strategy is not to be dislodged by the observation that the
disputed ‘facts’ appear to supervene on facts of other kinds. The radical factualist explains this
observation as well as anybody.

49
5
Fact and Psychological Function
We found that in the ethical and probabilistic cases, non-factualism can be driven to seek its
foundations in the head – to try to analyse in psychological terms the semantic distinction between
fact-stating and non-fact-stating discourse. Statements thus come to be characterized as (what are
normally) expressions of belief; non-statements – or those that the non-factualist is concerned with, at
least – as the (normal) expressions of other sorts of propositional attitude. And this appeal to
psychology is no mere expository gloss. In order to defeat the radical factualist, the non-factualist
must claim to be able to distinguish beliefs from other attitudes, in advance of settling the semantic
issue.

We have not shown that the non-factualist’s dependence on psychology goes beyond the ethical and
probabilistic cases. All the same, it is clear that non-factualists about various other topics do in fact
appeal to psychology. One important case is the non-factualist approach to indicative conditionals.
This relies on the observation (originally due to Ernest Adams 1965) that the subjective assertibility
of an indicative conditional goes by the subjective conditional probability of its consequent given its
antecedent; rather than say by the absolute probability of the material conditional. The view thus
involves that indicative conditionals are typically expressions of high subjective conditional
probabilities; and – crucially, if the account is to be non-factualist – that these cannot be construed as
a kind of belief. Another case is the so-called ‘epistemic’ treatment of possibility and necessity, and
the related treatment of laws of nature as generalizations towards which we hold a certain sort of
attitude (a view advocated by Ramsey 1978, chapter 6B, for example).

I think we are thus entitled to conclude both that non-factualism has often in fact put its faith in
psychology; and that at least in certain central cases it is bound to do so. Our final sceptical task is to
show that this faith is misplaced. Beliefs cannot be adequately distinguished from other propositional
attitudes; not at any rate unless we are prepared to rely, as the non-factualist here cannot, on a
characterization of this psychological distinction in semantic or metaphysical terms.

The chapter is in three main parts. I first consider the Humean approach to the belief/desire
distinction, based on Hume’s account of the origins of action. This account continues to be widely
taken for granted, even in these anti-Humean times. As a result, the task of explaining action in
psychological terms may seem a promising source of the sort of distinction we are looking for. I shall
argue that the promise is empty, however. In search of the source of the belief/desire distinction the
modern philosopher ends up, as Hume himself did, appealing once more to semantics: claiming that
beliefs, unlike desires, have truth conditions.

The second part of the chapter reaches a similar conclusion along a more general, more contemporary
and perhaps initially more promising route. It considers the suggestion that we look to recent
cognitive theory for an account of the content of a propositional attitude. A species of propositional
attitude is a class of beliefs if and only if there is a class of contents such that the species concerned is
the set of beliefs with contents of that class. (This is the psychological analogue of the condition for a
species of utterance to be represented as a class of assertions.) Hence a basis for the belief/non-belief
distinction might be expected to be a corollary of an adequate account of content. I shall argue,
however, that contemporary accounts of the nature of content do nothing to settle this issue – or rather
they settle it, if at all, in the semantic or metaphysical terms that the non-factualist has already
renounced.

Finally, I shall turn to some arguments claimed to show that indicative conditionals are not
expressions of genuine beliefs. In view of the logical role of conditionals, this is potentially the most
significant non-factualist theory of all. My approach will again be deflationary: I shall try to show that

50
these arguments parallel some striking but clearly inconclusive objections to a factualist treatment of
ordinary probabilistic judgements.

5.1 DISTINGUISHING DESIRES: THE HUMEAN STORY

The Humean argues that reason is incapable of action – in modern form, that beliefs are
motivationally inert – and hence that an action explanation is incomplete unless it appeals, at least
implicitly, to some sort of desire or motivating attitude. The view is as orthodox as it was once
controversial. Indeed, it is at the heart of so-called ‘folk psychology’ (a fact that should not be
uncritically attributed to the folk influence of Hume, one suspects). I want to show that the Humean
argument is inadequate, however, even in its most charitable contemporary reconstruction. It fails to
show that belief cannot motivate action; and hence it fails to show that moral commitments (even if
agreed to motivate actions) are not beliefs.

Our interest in the Humean argument stems from its application to these semantic questions. But it
may be as well to point out that in challenging the principle that belief is non-motivational, we are not
committed to rejecting totally the orthodox view of the explanation of action. For in one crucial
respect Hume’s conclusion is stronger than the orthodox belief/desire model requires. It would be in
keeping with the spirit of this view to say that the explanation of action involves two sorts of belief:
beliefs about the desirability, or value, of certain states of affairs; and beliefs about the bearing of the
action in question on the obtaining of those states of affairs. This modified view could agree, in any
actual case, on what constitutes an explanation of an action. It differs from the Humean only as to
how the attitudes mentioned by such an explanation should be described.

This difference is crucial to the use made of Hume’s principle by non-factualists in ethics. The non-
factualist needs to show that moral statements lie on the right side of a psychological distinction – in
other words, that they do not express beliefs. Given Hume’s principle, it only needs to be shown that
moral commitments do motivate actions; for this puts them on the desire side of Hume’s distinction.
Clearly, however, someone unconvinced by the Humean argument could quite well endorse the
standard model of action explanation; accept that moral judgements express mental states on the
active or motivating side; and yet insist that moral commitments are genuine beliefs about moral facts.
This was the response of the radical factualist we met in chapter 4.

Such a view admits two possible accounts of the place of desire. One account would recognize a
category of desires, distinct from beliefs, but claim that motivation is not the exclusive preserve of
attitudes of this category (moral commitments being one sort of exception). The problem will then be
to say what distinguishes beliefs from desires, given that it is not the power to motivate action. This
problem might incline us to the second sort of account, recognizing no significant foundation for a
distinction between beliefs and desires. This view would further divide into two possibilities: the
substantial claim that all desires are beliefs; and the weaker sceptical claim that there is simply no
adequate justification for not so representing an arbitrary propositional attitude. Our interests again lie
with the sceptic.

Let us turn then to the case for Hume’s principle. Why should not beliefs be motivational? The
Humean’s first line of attack might be something like the following argument. It is intended to show
that the belief that a contemplated action A would have some property F (or would bring about an
outcome having property F) cannot in itself constitute a reason for an agent to perform A. If there
were such a case (the argument claims) we could imagine a similar agent who would dislike F, and
would hence be motivated not to do A by a desire that not F. This being so, the action A cannot in fact
have been caused by or be explained in terms of the agent’s belief that A would be F. It is also
relevant that the agent desired that F, or at least did not desire that not F. In other words, the causal or
explanatory story needs to mention the agent’s desires.

51
I shall call this the imagined dislike argument (IDA). Hume himself appeals to IDA, or something
very like it, in this notorious passage:

’Tis not contrary to reason to prefer the destruction of the whole world to the scratching of
my finger. ’Tis not contrary to reason for me to chuse my total ruin, to prevent the least
uneasiness of an Indian or person wholly unknown to me. (1978, p. 416)

Hume’s immediate point is that reason does not control the passions. But given that the passions do
control action, the argument becomes IDA – as indeed the latter example illustrates, in speaking of
choice rather than preference. Hume’s suggestion seems to be that the belief that an action will
prevent his total ruin will not explain his performing that action, because he might have desired total
ruin (in preference, at any rate, to some other consequence of preventing ruin, such as the unease of
another party).

IDA has two major flaws, however. The first is to rely on the principle that if a causal explanation
would be inadequate in the presence of some extra condition, then it is inadequate unless it mentions
the absence of that further condition. This principle is no more plausible here than it is for causal
explanations in general. In almost every case, causes are at best sufficient in the circumstances for
their effects; and to insist on an explicit description of the relevant circumstances would be to cry for
the moon. (We rely on ceteris paribus clauses because we cannot fill in these gaps.) In the present
case, to admit such a principle would be to abandon any systematic approach to the psychological
explanation of action. After all, there is no end to the things that even the most fastidious agent does
not dislike.

IDA’s second fault is just as damning: given that the issue is whether desires are anything but a
special class of beliefs, the argument simply begs the question. For if some beliefs are desires, to
believe that an outcome would be F might be the very same thing as desiring that outcome. In this
case the possibility that the argument imagines – that of someone holding the belief, but desiring an
outcome such that not-F – would be that of a person who desires what he takes to be undesirable. An
advocate of the view that desires are a species of belief seems entitled to regard this as no possibility
at all.

Neither of these objections is particularly profound. So it may seem doubtful whether IDA does
justice to Hume’s distinction, or explains its contemporary appeal. However, I think the argument
does point us in the right direction. It seems to derive its initial plausibility from a model of the mind
which we find intuitively appealing. Perhaps it thus assumes what it is supposed to establish. But at
least it has the virtue of directing us to the crucial assumption.

We may think of the issue in terms of a simple model of the mental structure of an intentional agent.
The functional role of desire seems to be to connect two other components: on the one hand, a unit
(Cog) responsible for cognitive deliberation; on the other, a unit (Will) whose function is to initiate
bodily action, in response to the outputs from Cog. There are several possible models of Will and its
relationship to Cog. At one extreme we could regard it as a kind of store, or processing house, for
intentions already formed in Cog. At the other extreme we could take it to be the unit in which
judgements arriving from Cog are matched with a store of desires to yield intentions. However, since
we want a model in which the action that interests us takes place at the boundary between Cog and
Will, let us avoid both extremes. Let us view Will as a structure on which intentions ‘condense’, as a
consequence of suitable states of Cog.

In these terms, Hume’s principle is that no intention can condense in the absence of a desire – that
desires are a necessary catalyst to the intention forming process. IDA rests on the claim that whatever
beliefs might be present in Cog, we could imagine a desire whose presence would prevent a given
intention condensing from those beliefs. The argument is a poor one, but catalytic view of desire has a
lingering appeal.

52
However, we know that in the chemical case, altering either the composition of a vapour or the nature
of the condensation chamber may make a catalyst redundant. So the analogy suggests that we
consider the effect of changes in the corresponding aspects of the mental model: to Cog on the one
side, and Will on the other. As yet we have no reason to regard these factors as cognitive constants.
So there is nothing to stop us trying to imagine agents who differ in these respects, from each other,
and more importantly from the way the Humean view takes us to be. The appeal of the orthodox view
may rest on the fact that we have not yet noticed the alternatives.

There are two main possibilities to consider: the causal gap supposedly filled by desires might be
bridged from either side. The first possibility would be to expand Cog, so as to admit beliefs with the
capacity to produce intentions in a Will of the orthodox kind. The second would be to endow Will
with the disposition to form intentions in the presence of certain orthodox beliefs. (I ignore the
obvious hybrid possibilities.)

In the former case, we want a belief whose capacity to rationalize an action does not depend on an
agent’s possession of certain desires. It seems to me that we can get a good idea of what such a belief
would be like, by reflecting on familiar expressions of moral or rational necessity – ‘I am bound to
tell the truth’, for example, or ‘I must go down to the sea again.’ Superficially, at any rate, utterances
such as these can be regarded as expressions of beliefs that possible actions have certain properties.
Moreover, such beliefs (if they really are beliefs) seem to provide reasons for actions which are
insensitive to an agent’s desires about actions with the corresponding properties. The belief that one
must go down to the sea seems capable not merely of making one do so, but of overriding, in the
process, considerable distaste for messing about in boats.

These points are notoriously disputable. There are two lines of attack. One will urge that believing
that I must do something, or am bound to do it, only provides me with a reason for doing it because,
like everybody else, I have an implicit desire to do what I must. The other will take the opposite tack
and argue that because utterances such as these express reasons for actions, they do not express
beliefs. However, both lines of argument seem to depend on the very claim in question: namely, that
beliefs alone cannot motivate actions. To settle the issue, we would need a criterion with which to
distinguish beliefs from desires. Hume’s argument claims to offer such a criterion, saying that desires
are distinguished in virtue of their active or motivating role in the production of action. But it now
appears that the argument itself depends on the very criterion it is intended to supply. As McDowell
puts it,

the idea of the world as motivationally inert is not an independent hard datum. It is simply the
metaphysical counterpart of the thesis that states of will and cognitive states are distinct
existences; which is exactly what is in question. (1978, p. 19)

Thus it seems that we have no non-circular reason to deny the possibility of an intrinsically
rationalizing belief: a belief whose capacity to motivate and rationalize a contemplated action does
not depend on co-present desires (even if its motivational effect could be overridden by certain such
desires, were they to be present). And if we have reason not actually to affirm the possibility, it lies
partly in the fact that the distinction between the belief and desire is itself so obscure.

I said that there are two possible ways to bridge the claimed gap between Cog and Will. The second is
to imagine Will itself equipped to handle a somewhat broader range of inputs. Why shouldn’t Will, or
the process by which an agent forms intentions to act, simply be built to respond to certain beliefs
about the properties of actions and states of affairs?

Consider, for example, a classical economic rationalist, whose actions are based entirely on the profit
motive. Such an agent forms intentions to perform those actions he believes will maximize his net
monetary return. One might object that the rationalist wants to make a profit. But why should he?
Why should his mind not simply be organized so that believing that an action would maximize his

53
profits normally leads him to form the intention to undertake it (in just the way, and to whatever
extent, belief/desire combinations are ordinarily supposed to lead to intentions)?

One possible answer appeals to Davidsonian themes. We ourselves wouldn’t be motivated to act as
the rationalist does unless we wanted to make a profit, so we can only make sense of his behaviour –
rationalize it – by ascribing to him such a desire. This objection might lead us too far astray. In
making the ascription of attitudes to cognitive systems radically dependent on the ascriber’s own
system of attitudes, it threatens the entire project of speculative psychology. Fortunately, we can
evade it. We want to know, inter alia, whether our own Wills are ever activated by beliefs alone. But
in the face of a purported real case of this sort of thing, the Davidsonian move will simply be
powerless. It cannot insist that our rationalization of the action in question would need a desire, for
that will be the very point at issue.

I conclude that motivating beliefs are a conceptual possibility. If we begin with an orthodox Humean
model of ourselves, we can imagine two kinds of psychological modification that would allow beliefs
to motivate actions. The effect of this is not simply that the practical question of the nature of the
motivation of action remains unsettled, but rather that the question itself is under threat. Hume’s
principle not only entails that action requires desires, as well as beliefs; it also offers a criterion for
distinguishing desires from beliefs. Without it, the non-factualist lacks a foundation for that
distinction.

Like Hume himself, the non-factualist may therefore feel inclined to try another tack – to say that
desires are distinguished from beliefs in virtue of the fact that it is ‘impossible ... they can be
pronounced either true or false’ (1978, p. 458). This would be to distinguish beliefs in virtue of their
semantic properties; and it is a popular move. However, we shall see that it takes the non-factualist
back to positions already abandoned – positions whose untenability forced the non-factualist to seek a
psychological basis for the fact/non-fact distinction.

This may seem an immediate consequence of such an appeal to semantic properties. After all, the
attempt to analyse assertion in terms of truth was the first target of our sceptical challenge. For the
moment, however, a significant possibility remains open to the non-factualist. Psychological truth
might succeed where linguistic truth failed. If semantic properties apply primarily to mental states,
and only derivatively to language, then the arguments of chapter 2 may have missed the mark. Truth
might indeed be insubstantial, but nevertheless play a crucial role in establishing a psychological basis
for the fact-stating/non-fact-stating distinction.

To exclude this case, we shall need to consider some possible foundations for psychological
semantics. This brings us to the concerns of contemporary cognitive theory. And it brings us to
another possible response to the argument of this section. We have seen that the requirements of
action explanation do not yield the psychological distinction the non-factualist is looking for. On
reflection, however, this avenue was always rather limited. At best it offered a distinction between
beliefs and the range of motivating attitudes we have loosely called ‘desires’. Even if this had met the
needs of the ethical non-factualist, it would have been largely irrelevant to other non-factualists. If
beliefs do not motivate actions, then neither, for example, do the partial beliefs invoked by the non-
factualist account of probability.

From this point of view, our mistake may seem to have been to concentrate on too small an aspect of
the functional organization of the mind – to concentrate on the initiation of action, to the exclusion of
other functions of the mind. As it happens, this is exactly the mistake of which contemporary
functionalist accounts of the mind accuse their behaviourist predecessors. In the next section we want
to enquire whether these contemporary theories succeed where Hume failed, in distinguishing beliefs
from non-beliefs.

5.2 FUNCTION AND COGNITIVE CONTENT

54
We may seem to be faced with a daunting diversity of contemporary views on the nature of the mind.
One of Professor Fodor’s relatives ‘sighs for the days when well-brought-up philosophers of mind
kept themselves occupied for hours on end analysing their behavioural dispositions’ (Fodor 1985, p.
76). But things are not as bad as they seem. For one thing, as Fodor puts it, ‘the chaotic appearances
are actually misleading. A rather surprising amount of agreement has emerged, if not about who’s
winning, at least about how the game has to be played’ (1985, p. 76). We can exploit this agreement –
and Fodor’s guide to the options – for our present purposes.

More importantly, however, it turns out that there is very little to say: none of the contemporary
options tries to analyse the distinction we are looking for, or indeed makes a point of claiming to have
the resources for doing so. Most contemporary theories simply take for granted the Humean model of
action explanation – in the guise, as they put it, of ‘folk psychology’. This provides the framework
within which they tackle such questions as the nature of content. And those that do not assume the
belief/desire model of action explanation are those that reject folk psychology, and with it the
apparatus of propositional attitudes. So on neither side is there a developed claim to provide a basis
for the distinction between beliefs and other sorts of attitudes; nor is there any prospect of such a
basis.

This claim may seem implausible. After all, we know that we can think of the question as to whether
desires (say) are a kind of belief, as that of the existence of an appropriate species of content.
Roughly, desires can be represented as beliefs if and only if there is a content modifier Des such that
the desire that P can be represented as the belief that Des(P). And cognitive science has had a great
deal to say about the notion of content. Surely there is something that will tell us whether there could
be such a Des?

It seems not. The point is perhaps most clearly illustrated in what Fodor calls ‘the Representational
Theory of Mind’ (RTM), according to which propositional attitudes are relations to representations in
a ‘language of thought’. Fodor gives the following ‘canonical formulation of this view’:

For any organism O and for any proposition P, there is a relation R and a mental
representation MP such that: MP means that (expresses the proposition that) P; and O
believes that P iff O bears R to MP. (And similarly, R desires that P iff O bears some different
relation R′, to MP. And so forth ...) (1985, p. 88)

Clearly, the distinctions between the propositional attitudes are here built in from the start: the desire
that P is interpreted in terms of the relation R′ to MP, rather than in terms of the relation R to
DES(MP) (DES being the mental representation of the content modifier Des).

From the Humean point of view this looks perfectly natural. If we question the Humean categories,
however, it begins to seem rather arbitrary. For mental representations and the attitude relations R, R′,
etc., are both individuated in the same kind of way: in terms of their causal or functional roles. One of
the great attractions of the RTM is its ability to exploit the computer metaphor, and conceive of
mental processes as ‘transformations of mental representations’ (Fodor 1985, p. 95). Such
transformations are causal processes whose outcomes depend on two factors: the syntax of the mental
representations with which the process begins, and the mode in which those representations are
initially stored – whether they are ‘believed’, ‘desired’ or whatever. But why not reduce these two
factors to one, by regarding the mode of storage as an aspect of syntax? After all, in any actual case
the available syntax will be capable of representing an enormous range of distinctions. It is surely
capable of a handful more.

It is true that if we knew the syntax of the language of thought – in advance, as it were – then the
question of representing mode of storage as a syntactical distinction would not arise. The syntactical
guidebook would tell us that there was no way of representing certain distinctions within the
language; leaving us no choice but to regard their role as that of a mode of storage of a sentence of the

55
language. The non-factualist cannot hope for an answer of this kind, however. For what is the
semantic role of the sentences of the language of thought? Simply that of the fact-stating sentences of
natural language – or so the non-factualist must say, for it is these sentences he takes to express
beliefs. In other words, the task of delineating the syntax of the language of thought is simply the
original task of delineating fact-stating discourse: the very task for which the non-factualist now
wants to distinguish beliefs from non- beliefs.

So the non-factualist is not entitled to appeal to the syntax of a language of thought, in order to
distinguish beliefs. Indeed, no one is entitled to appeal to the syntax of a language of thought in order
to determine which functional roles should be counted content- determining and which attitude-
determining. For (as Mellor 1988 emphasizes) semantics must precede syntax, in this sense: our
grounds for regarding a system of states and causal processes as a syntactic system of representations
and transformations must be that we assign its elements a semantic role. The title ‘linguistic’ has to be
earned.

There is little agreement on how it should be earned. Fodor again: ‘Of the semanticity of mental
representations we have, as things now stand, no adequate account’ (1985, p. 99). It seems to me that
proponents of the RTM have not always been sufficiently aware of the extent to which it depends on
the assumption that this problem is solvable, at least in principle. For how else could an opposing
view – that of Stich (1983) – present itself as a purely syntactic account? Stich says that the

core idea of the STM [Syntactic Theory of Mind] – the idea that makes it syntactic – is that
generalizations detailing causal relations among the hypothesized neurological states are to be
specified indirectly via the formal relations among the syntactic objects to which the
neurological state types are mapped.’ (1983, p. 151)

But how is this different from a formal description of (an aspect of) the causal structure – say an
account of the operation of a telephone exchange that describes it in terms of a formal model of its
switching operations? In the absence of the semantic notion of representation, the label ‘syntactic’ is
quite inappropriate.

Stich evidently regards the significance of the ‘syntactic’ character of his theory of mind as more than
that of a mere formal description. This shows up in the fact that he entertains (though he does not
insist on) versions of the STM that follow folk psychology in postulating ‘two distinct classes of
states underlying behaviour, belief-like states and desire-like states, with state types of both classes
mapped to a single class of formal objects’ (1983, p. 151). Clearly, if the syntactical story was simply
a formal model there would be no reason not to treat the belief/desire distinction as itself syntactical
(in the manner, presumably, of the belief/disbelief distinction). Belief-like states and desire-like states
would then get mapped to different but related classes of formal objects (and we would again have no
reason not to regard desire-like states as a special kind of belief-like state). From a formal point of
view, this is the obvious approach; and only the vestiges of the semantic point of view of folk
psychology could recommend the alternative.

I emphasize that this is a criticism only of Stich’s ‘syntactic’ characterization of his theory, and not of
the theory itself. On the contrary, I think Stich’s articulation of a semantics-free approach to the mind
is of considerable dialectical value to the proposal I want to develop in part II (where the task, in a
sense, will be to explain how the semantics gets in). The effort would be misguided, to say the least, if
it could not begin from a foundation that does not already include what we want to put in. Stich has
made it very much less implausible that there could be such a foundation.

Back to the RTM: it might be objected that in suggesting that this view leads us back to the claim that
desires are distinguished from beliefs in virtue of their lack of truth conditions, I have ignored the
RTM’s account of content. Given this account, it might be suggested, the RTM can characterize the
sentences of the language of thought without recourse to the semantic distinctions of natural language

56
(and hence, for our purposes, without leading us back to our starting point). In other words (the
suggestion would go) the RTM can explain the fact that it does not represent the desire that P in terms
of DES(P): namely by showing that its account of content does not leave room for the necessary
content modifier, Des.

This objection might be more forceful if there were an agreed approach to the problem of the content
of propositional attitudes. As it stands, the closest thing to an agreed position is that content is
individuated, in part at least, by the causal or functional roles of the attitudes concerned. This
suggestion trades on the appealing hypothesis that the inferential role of propositional contents – the
sort of thing that is traditionally supposed to individuate propositions – is modelled in the causal role
of the relevant mental states. This association is at the heart of the computer metaphor: ‘computers ...
just are environments in which the causal role of a symbol token is made to parallel the inferential
role of the proposition it expresses’ (Fodor 1985, pp. 93–4).

Appealing as this suggestion is, however, it does nothing to justify the belief/non-belief distinction.
On the contrary, it again takes for granted a distinction between those aspects of the functional role of
a state token that determine its content, and those that determine that it is a belief, a desire, or
whatever – a distinction that again looks particularly odd, if we bear in mind the ratio of content
differences to attitude type differences in the finished theory (and take off our folk-psychological
blinkers). It is as though the Taxation Department kept separate databases for its male and female
clients: we would want to know why gender could not be represented explicitly, like everything else
that the Tax Department knows about us. Why buy a filing cabinet, if you intend to keep everything in
two or three colossal files? Or to put it another way, why start a separate file for desires, but not start
one for disbeliefs, say, or beliefs about nutrition?

One last suggestion: perhaps we have concentrated on the wrong aspect of content. Many cognitive
theorists think that there are two aspects to content: the internal aspect, that (more or less) matches
internal functional role; and an external aspect, that depends on the actual reference of the attitude
concerned – on what in the world it is in fact an attitude about. (See McGinn 1982, for example.
Fodor himself now favours the view that content is entirely a matter of the external aspect; see Fodor
1987.) With the external aspect taken into account, the range of contents is circumscribed by the range
of facts. But clearly this takes us back to our starting point. We wanted the belief/non-belief
distinction in the first place because the direct metaphysical attacks looked so unpromising.

In summary, then, we find that if content is individuated in terms of functional role, there is no
apparent reason to distinguish more than one type of propositional attitude; and that if it is
individuated in terms of external semantic relations, the limits of belief depend on the limits of fact.
Neither possibility provides what the non-factualist now needs in order to answer the sceptic or the
radical factualist: a psychological criterion with which to distinguish beliefs from other attitudes, and
thereby to distinguish fact-stating from non-fact-stating discourse.

*5.3 CONDITIONAL FACTS

There is one other important case in which what can be taken to be functionalist arguments have been
widely thought to support a non-factualist conclusion. The case is that of indicative conditionals, and
the arguments rest on Ernest Adams’s (1965) hypothesis, now widely accepted, that the ‘assertibility’
of an indicative conditional goes by the conditional probability of the consequent, given the
antecedent (rather than by the probability of the corresponding material conditional, say).

To see what is meant by Adams’s claim, let us look at the ordinary case, with which Adams contrasts
the behaviour of indicative conditionals. In what circumstances, in general, will a speaker be willing
to assent to a statement S? As we have seen, the standard answer is something like this: When he or
she believes that S, or at least has a partial belief (or degree of confidence) that S that in the context in

57
question is sufficiently close to a full belief that S. This answer is of course very rough, but not too
rough. On the one hand, it is clearly a subjective condition, not immediately concerned with objective
justification for assent to S. On the other, it does not imply that speakers have to think about their
beliefs, in order to assent to S. We do as a matter of fact tend to assent to those propositions in which
we have high degree s of confidence, the claim goes, even if we are rarely aware that this is what is
going on.

We saw in chapter 4 that a determined factualist about probabilities will equate degrees of partial
belief, or degrees of confidence, with beliefs about probabilities. The normal rule of assertibility then
becomes the principle that assertibility goes by increasing probability. ‘Assertibility goes by
increasing subjective probability’ is the way it is often put, but this can be a source of confusion. If
initially applied to a degree of confidence, the term ‘subjective probability’ encourages the (factualist)
view that for any partial belief there is an associated full belief about a probability. What probability
is this? Why, the subjective probability! The term ‘subjective probability’ thus comes to refer
ambiguously to the original degree of confidence, now thought of as a belief about a probability, and
to the probability that figures in the content of this belief. In fact, of course, if degrees of confidence
are beliefs about probabilities, then these probabilities need be no more subjective than the subject
matter of any other belief. While the beliefs themselves are subjective only in the same sense: namely,
that they are our beliefs, rather than somebody else’s. For this reason, I shall avoid ‘subjective
probability’, instead referring to the mental states concerned as credences. The term ‘credence’ leaves
it open as to whether, with the radical factualist, we should think of credences as beliefs about
probability.

Adams’s hypothesis about indicative conditionals amounts to the claim that what in general
determines whether we assent to such a conditional is not our credence in, say, the corresponding
material conditional, but rather what we may call our conditional credence in the consequent, given
the antecedent. For a factualist about probability a conditional credence is a belief about a conditional
probability; and Adams’s hypothesis comes to the claim that the assertibility of an indicative
conditional goes by perceived conditional probability, rather than by the perceived probability of a
truth-functional conditional.

It is harder to say what a non-factualist about probability should make of the notion of conditional
credence, and hence of Adams’s hypothesis. For one thing, if absolute (that is, non-conditional)
credences are not beliefs about probabilities then, given Adams’s hypothesis, the orthodox definition
of conditional probability –

(5.1) P(C/A) = P(C&A)/P(A), so long as P(A) is non-zero –

seems to provide no suitable account of conditional credences. The main problem is that we often
assent to an indicative conditional – and therefore, on this story, have a conditional credence – when
we clearly have no corresponding absolute credence in the antecedent, the consequent, or their
conjunction. I am confident that if it is raining in Moscow then the domes of the Kremlin are wet; but
I simply have no idea as to whether it is raining there. (For this and related arguments, and a
discussion of what the non-factualist should make of them, see Price 1986.)

However, our present concern is to show that despite certain arguments to the contrary, a factualist
about probability can be a factualist about conditionals. Given the factualist’s view of absolute
credences, it is easy to account for the above cases, and yet to reconcile Adams’s hypothesis with the
standard definition of conditional probability. For we can believe that the ratio of two probabilities is
high, without holding an opinion as to the value of either. The factualist about probability thus regards
a conditional credence of degree n in C given A, as a belief that P(C&A)/P(A) = n; and interprets
Adams’s hypothesis as the claim that the assertibility of the indicative conditional ‘If A then C’ goes
by the value of this ratio. Roughly, we assent to such a conditional when we believe that this ratio is
sufficiently close to 1.

58
We can think of such principles of assertibility as fragments of a functionalist account of the mind.
They belong to that part of such a theory that relates mental states to linguistic behaviour. As such,
moreover, these principles suggest a ground for a non-factualist treatment of conditionals. For they
suggest that the general functional account of assent behaviour, summed up in the principle that
assertibility goes by degree of (absolute) credence, does not apply to conditionals. Conditionals seem
to need a functional account of their own – in which case we have a theoretical basis for
distinguishing them from utterances to which the general account does apply.

The argument rests on the assumption that the principle of assertibility for conditionals is not a special
case of the general principle. In other words, it requires that conditional credence is not simply a
species of absolute credence. If there were a conditional, ‘®’ such that

(5.2) P(A®C) = P(C/A)

the factualist could endorse Adams’s hypothesis, and yet, simply by claiming that ‘If A then C’ is to
be interpreted as ‘A®C’, be able to say that for indicative conditionals, as for statements of fact in
general, assertibility goes by absolute credence.

This suggestion was an early response to Adams’s (1965) formulation of his hypothesis (for example
by Stalnaker 1970 – 5.2 is sometimes referred to as ‘Stalnaker’s hypothesis’). But there are two
arguments that have been claimed to show that there can be no such conditional ‘®’. The first is due
to Lewis (1976) and the second to Carlstrom and Hill (1978). I want to show, however, that in both
cases the supposed problem for this reading of conditional credence is already a fairly obvious but
non-fatal feature of a factualist view of absolute credence – and a feature that a factualist about
conditionals can accommodate in much the same way as a factualist about probabilities.

Lewis’s argument offers a reductio of the assumption that there is a conditional ‘®’ whose
probability is given by 5.2. The key is a standard principle of decomposition by cases – a
straightforward consequence of the standard axioms of probability:

(5.3) P(X) = P(Y)P(X/Y) + P(~Y)P(X/~Y).

Substituting A®C for X and C for Y, 5.3 gives us

(5.4) P(A®C) = P(C)P(A®C/C) + P(~C)P(A®C/~C).

Assuming for the moment the equality

(5.5) P(X®Y/Z) = P(Y/X&Z),

from 5.4 we derive

(5.6) P(A®C) = P(C)P(C/A&C) + P(~C)P(C/A&~C)

and hence, since P(C/A&C) = 1 and P(C/A&~C) = 0,

(5.7) P(A®C) = P(C).

From 5.2 it now follows that

(5.8) P(C/A) = P(C).

59
 This conclusion is perhaps absurdity enough. It amounts to the result that any pair of
propositions X and Y are probabilistically independent, provided only that P(X&Y) and P(X&~Y) are
non-zero. But Lewis derives two further results: that there cannot be three incompatible propositions
X, Y and Z that all have non-zero probability; and hence that the probability function has at most four
values.

Lewis’s argument uses 5.5, which he derives quite straightforwardly from the standard definition of
conditional probability (5.1) and the assumption, in effect, that rational agents change their credences
by conditionalizing: i.e., that on coming to believe that X, a rational agent assigns to any proposition
Y a probability equal to the conditional probability she previously ascribed to Y given X. But the
details need not concern us, I think, because 5.5 is more general than the argument requires, and we
can get what we need for 5.6 a little more directly.

Given the conditionalization assumption (‘COND’, for short), it follows that P(A®C/C) equals
PC(A®C) – that is, the probability that (as rational agents) we would ascribe to A®C, if we came to
believe that C. But by 5.2,

(5.9) PC(A®C) = PC(C/A)

and since PC(C) = 1,

(5.10) PC(C/A) = 1.

This shows that P(A®C/C) = 1, and similarly we can use the fact that P~C(C) = 0 to show that
P(A®C/~C) = 0, giving 5.7.

In order to show how the factualist might respond to Lewis’s argument, I want to exploit certain
similarities it bears to a classic argument for fatalism. This argument is generally agreed to be
fallacious. But the key to the fallacy is the key to the factualist’s response to Lewis (which is not to
say that Lewis’s argument is fallacious – more on this in a moment).

The traditional function of the argument in question is to encourage an audience to risk its life on the
proponent’s behalf. In the following fine example (for which I am indebted to Hugh Mellor), the
proponent is Henry V, admonishing Westmoreland for wishing for reinforcements for their impending
battle with the French:

... No, my fair cousin:

If we are mark’d to die, we are enow
To do our country loss; and if to live,
The fewer men the greater share of honour. (Henry V, Act IV, Sc. III)

Philosophers are not insensitive to the pragmatic virtues of this style of argument. Leibniz (1951, p.
153) remarks that ‘the sophism which ends in a decision to trouble oneself over nothing will haply be
useful sometimes to induce certain people to face danger fearlessly.’ He comments that ‘it has been
applied in particular to Turkish soldiers’, but has doubts about its efficacy: ‘It seems that hashish is a
more important factor than this sophism, not to mention the fact that this resolute spirit in the Turks
has greatly belied itself in our days.’ Pragmatics aside, however, Leibniz, like most philosophers, is
convinced that the argument is fallacious.

Its similarity to Lewis’s argument stems from two factors: the first that Lewis’s principle of
decomposition (5.3) can be thought of as a generalization of the logical principle of argument by
Dilemma, used by the fatalist; and the second that both arguments involve hypothetical judgements
about conditionals. In the case of the fatalist argument, the latter characteristic may not be

60
immediately apparent. It is essential, however, for the fatalist must convince his victim not only that
bravery is better in either eventuality, but also that the victim’s choice will not otherwise determine
the course of events. In the above case, for example, Henry must convince Westmoreland that
whether (A) they fight without reinforcements is irrelevant to whether (C) they are victorious. For
otherwise Westmoreland will rightly point out that their choice is not between A and not-A (A being
better whether C or not-C). Rather it is between (A and not-C) and (not-A and C) – the latter
presumably being preferable.

The fatalist thus argues as follows:

1 CÚ~C.
2 C®(A®C)&(~A®C) – i.e., if C, then if A then C and if not-A then C, so that A is irrelevant to
whether C.
3 ~C®(A®~C)&(~A®~C) – i.e., if not-C, then if A then not-C and if not-A then not-C, so A is
again irrelevant to whether C.
4 Hence, by Dilemma, A is irrelevant to whether C.

It is premisses 2 and 3 that involve hypothetical assessments of conditionals.

As well as these two similarities, of course, there are some very significant differences between this
argument and Lewis’s. The two arguments are quite different in intent. Lewis’s is a reductio. Indeed,
the absurdity that Lewis derives from 5.2 parallels the fatalist’s final conclusion. In arguing for 5.5,
moreover, Lewis justifies premises whose analogues (premises 2 and 3) the fatalist takes for granted.
Nevertheless, I think that the arguments have enough in common to allow us to use our grounds for
rejecting the fatalist’s as a guide to a possible answer to Lewis’s.

Let us focus on premises 2 and 3, which compare with Lewis’s use of the equalities P(A®C/C) = 1
and P(A®C/~C) = 0. At this point both arguments rely on the procedure of assessing a conditional
under an assumption concerning the truth of its consequent. I think we can get a good idea of what a
factualist might object to about this procedure by constructing a parallel involving, in place of
conditionals, a claim about probabilities themselves.

Let PROP be the proposition that P(C) = 1. The following argument claims to show that P(PROP =
P(C). By 5.3,

(5.11) P(PROP) = P(C)P(PROP/C) + P(~C)P(PROP/~C).

Now by COND, P(PROP/C) is the probability we would ascribe to PROP, if we came to believe that
C. In these circumstances, we would ascribe probability 1 to C itself (this follows from COND and
the fact that P(C/C) = 1). In other words (now assuming a factualist’s view of credence), we would
come to believe that P(C) = 1 – that is, that PROP. This doesn’t yet tell us what probability we would
ascribe to PROP; but applying COND once more, this time using the fact that P(PROP/PROP) = 1,
gives us what we need: in these circumstances, P(PROP) = 1. Similarly, we can show that
P(PROP/~C) = 0. By 5.11, this gives us P(PROP) = P(C).

Somewhat less subtly, this time following the fatalist rather than Lewis, we can ‘show’ that PROP is
materially equivalent to C itself:

1 CÚ~C.
2 Given that C, then there is no possibility that not-C, and so P(C) = 1 – i.e., PROP is true, as is C
itself.
3 Given that not-C, on the other hand, it is certainly not certain that C, so PROP is false – again
like C itself.

61
 4 Hence, by Dilemma, PROP always has the same truth value as C.

It is clear that on any ordinary view of probability, both conclusions are absurd. The first rules out
cases (easily found in physics, for example) in which the probability of an outcome necessarily lies in
some range that excludes 1; while the second rules out all probabilities other than 1 and 0. But it is
easy to see on what grounds a factualist will find these arguments fallacious. They assume, in effect,
that PROP has a fixed meaning, such as to allow us to assess its truth value or probability in
hypothetical contexts. Whereas in fact the sense of a probabilistic judgement depends very much on
its context. The fallacy becomes apparent (the factualist will say) if we make explicit the normally
implicit indexicality of such a judgement. If we think of PROP as

(5.12) Given the evidence to hand, P(C) = 1,

for example, then it becomes clear that in assessing PROP under the assumptions that C and that not
C – in adding each assumption to the available evidence, in effect – we assess a different proposition
in each case. We are interested in whether, given the evidence actually to hand, P(C) = 1; it is no help
to be told that if further evidence were to hand (namely the fact that C), it would then be the case that
P(C) = 1.

In the long run, we should clearly expect from the factualist a more precise account of the indexicality
of probabilistic judgements. We shall return to this topic in chapter 7. For the moment the important
thing is that from the factualist’s point of view an appeal to implicit indexicality seems to work as
well for conditionals as for probability. If we think of A®C on the lines of

(5.13) Given the evidence to hand, C follows from A,

then both Lewis’s argument and the argument for fatalism commit the same indexical fallacy as the
probabilistic arguments. They show that on the assumption of certain further evidence – the
hypotheses that C and that not-C – A is irrelevant to C; but in general these facts are themselves
irrelevant to the question at issue, which is whether on the actual evidence C follows from (or is
probabilistically dependent on) A.

Against this indexical view of indicatives, it might be objected that it violates the presumption, as
Lewis puts it, that ‘our indicative conditional has a fixed interpretation, the same for speakers with
different beliefs, and for one speaker before and after a change in his beliefs.’ How else, Lewis asks,
‘are disagreements about a conditional possible, or changes of mind?’ (1976, p. 301). The thought is
thus that if conditionals involve an implicit indexical reference to evidence, then as evidence varies
between speakers, and for a given speaker over time, so will the content of conditionals. This seems
contrary to the intuition that different speakers, or the same speaker at different times, can disagree
about a conditional, without, as it were, arguing at cross purposes. For the moment, however, it is
sufficient answer to this objection that speakers apparently cope with the same difficulties in the case
of probabilistic judgements (and certainly do so with many other sorts of indexical expression). The
important thing is that the two cases seem to be in the same boat.

Let us turn then to the second argument claimed to show that Adams’s hypothesis cannot be treated as
a special case of the general principle of assertibility for statements of fact: that of Carlstrom and Hill.
Here we cannot do better than to reproduce Carlstrom and Hill’s own presentation of the argument
(with minor notational changes):

We will derive a contradiction from the assumption that truth condition s can be assigned to the
indicative conditional A®C, for arbitrarily chosen A and C, in some possibly complex way such
that the probability that A®C is true is identical with the conditional probability P(A&C)/P(A).
Let A and C be contingent and logically independent sentences, and let A, A&C, and A®C be true

62
 at some possible world W1. The truth value of A®C at a world is not fixed by the truth values of
A and C, for none of the sixteen equivalence classes of Boolean sentences constructible from A
and C has a member with the property that its probability of truth is always identical with
P(A&C)/P(A). Thus there are worlds W2 and W3 with the following properties: A®C is false at
W2 but true at W3, and each of A and A&C has the same truth value at both worlds. We can
imagine two possible ways in which probabilities are assigned to the class of worlds. Under the
first assignment, one of the two worlds W1 and W2 is probably the case, but each is about as likely
to be the case as the other. In such a situation a sentence true at both worlds has a probability of
nearly one of being true, a sentence true at just one of the worlds has a probability of about one
half, and a sentence true at neither world has a probability of about zero. The second assignment is
just like the first except that the worlds W1 and W3 are considered in place of W1 and W2. Now
since A®C is true at W1 and false at W2, its probability of truth under the first assignment is
about one half. Under the second assignment its probability is nearly one. But the probabilities of
both A and A&C are about the same for both assignments, and in both cases, the probability of A
is considerably larger than zero. Hence, the conditional probabilities are about the same. Although
the probabilities that the conditional is true are quite different, the conditional probabilities are
very close. Conditional probability is not probability of truth. (1978, pp. 156–7)

I want to show that as with Lewis’s argument, the problem that this argument identifies is already
a feature of a factualist interpretation of non-conditional credences. In this case the relevant point is so
obvious that it may seem trivial. For example, let B be the proposition that the probability of a given
proposition A is 0.9. Since P(A) = 0.9 does not imply that A, it is possible that B is true and A false.
Let W be a world in which this is the case. Now imagine an assignment of probabilities to worlds that
gives W a probability very close to one. In terms of this assignment it is true that P(A) is close to zero,
and yet also very probable that P(A) = 0.9. In other words, this assignment justifies the assertions both
that B is false, and that B is very probably true.

The fallacy is fairly obvious: if we want to assign probabilities to possible worlds, and hence to think
of the probability of a proposition as the sum of the probabilities of the worlds in which that
proposition is true, then we must take it that judgements about probabilities are not true or false at
individual worlds, but rather in virtue of facts about the space of worlds as a whole. Probabilistic
judgements must thus be viewed as modal judgements.

Again, our concern is not with the adequacy of this form of factualism about probabilities, but rather
with the point that the same kind of answer defeats Carlstrom and Hill’s analogous objection to a
factualist treatment of indicative conditionals based on Adams ‘s hypothesis. I think the most striking
way to illustrate this is simply to define ‘A®C’ as ‘P(C/A) = 1’. It is true that in one sense this
violates Adams’s hypothesis – more on this in a moment – but it preserves something similar. We
have already noted that it is a consequence of the factualist view of credence that for any proposition
X, to believe (or be disposed to assent to) X is to believe (or be disposed to assent to) P(X) = 1. Let us
express this consequence as

(5.14) Bel(X) « Bel(P(X) = 1).

Given Adams’s hypothesis, this means that if we accept 5.2, we are in any case committed to

(5.15) Bel(A®C) « Bel(P(C/A) = 1).

The present suggestion is in effect that we should regard 5.15 as a consequence not of Adams’s
hypothesis and 5.14, but simply of the fact that ‘A®C’ and ‘P(C/A) = 1’ are synonymous. The
consequence of 5.14 will rather be that

63
 (5.16) Bel(A®C) « Bel(P(P(C/A) = 1) = 1).

To the extent that it tells us why someone who believes that the probability of C given A is
(effectively) 1 also believes that if A then C, this suggestion agrees with – indeed, explains – Adams’s
hypothesis. But it will see it as a mistake to regard Adams’s hypothesis as an assertibility condition, in
the above sense: the same kind of mistake as we would make in claiming that ‘X is a bachelor’ is
governed by the rule that one should be prepared to assent to it just in case one believes that X is an
unmarried male.

In the present case, the difference shows up in cases in which P(C/A) is assigned a value other than 1.
Suppose, for example, that we believe that P(C/A) = 0.75. On Adams’s hypothesis, this implies that
we take ‘If A then C’ to be assertible to degree 0.75; whereas on the present suggestion, since P(C/A
= 0.75 implies that P(C/A) is not one, ‘A®C’ will simply be believed to be false. As it happens, this
suggestion thus makes better sense than Adams’s hypothesis of at least some uses of indicative
conditionals. Consider this exchange, for example:

S: If his heart has stopped, then he is dead.

R: That’s not true – some people have survived cardiac arrests, so even if his heart has stopped, he
may still be alive.

In such cases conditionals seem to be rejected outright, even if the corresponding conditional
probability is quite high.

However, our present concern is not to defend this view of indicative conditionals, but simply to make
the point that for a factualist about absolute credences, the Carlstrom and Hill argument, like Lewis’s,
raises no new problem for a factualist view of conditionals. Carlstrom and Hill’s reductio depends on
imposing on conditional truths a constraint the factualist is already bound to reject for probabilistic
truths. This is really all we need here, though the above suggestion goes further. It proposes a
factualist interpretation of indicative conditionals, automatically available to factualists about non-
conditional credences, that makes sense of the grounds, if not the letter, of Adams’s hypothesis: the
intuition that assent to an indicative conditional requires a high conditional credence.

I conclude that the arguments of Lewis and of Carlstrom and Hill do not prevent a factualist from
endorsing Adams’s hypothesis (or something like it), and yet treating the resulting assertibility rule
for indicative conditionals as a special case of the rule for statements in general. Earlier we construed
these assertibility rules as fragments of a functionalist account of mind and linguistic behaviour. It
seemed that indicative conditionals might be a case in which functionalism could provide the
belief/non-belief distinction required by a psychologically grounded non-factualism. But it now
appears that for conditionals, as in general, the functionalist is unable to justify a division between
those functional differences that reflect differences of content, and those that reflect a difference in the
type of attitude concerned.

It should be mentioned here that some factualists about conditionals – notably Lewis himself and
Frank Jackson – have responded to Adam’s hypothesis in a very different way. Broadly, they accept
that indicative conditionals employ a special rule of assertibility, but consider this to be an addition to
the appropriate case of the rule for truth-conditional utterances in general. On this account indicative
conditionals have truth conditions – those of the material conditional, in fact – and their rule of
assertibility requires, inter alia, a belief in their (probable) truth. But for reasons to do with the
conversational role of such utterances, it also requires that the corresponding conditional credence be
sufficiently high. (Lewis 1976, Jackson 1987.)

The major difficulty for such an approach is to explain the significance of the claim that conditionals
have truth conditions. It cannot be to account for ordinary practices of acceptance and rejection of

64
such conditionals; for to recognize a need for Adams’s hypothesis is to recognize that the truth
conditions of the material conditional do not explain ordinary usage. Moreover, we concluded in
chapter 3 that if we are prepared to explain away usage, any meaningful utterance can be construed as
truth-conditional. In a parallel fashion, any utterance can be construed as expressing a belief. But the
construction does not guarantee that such a belief will play any non-eliminable role in a functional
account of the use of such utterances. In a similar way, it seems to be the functional story associated
with Adams’s hypothesis – the special rule of assertibility – that does the work in a Lewis or Jackson
account of the use of indicative conditionals. Lewis and Jackson accept that in virtue of Lewis’s
reductio argument (and perhaps also Carlstrom and Hill’s) this part of the story cannot be interpreted
in terms of the general account of statement of fact. The above argument suggests that this is a
mistake. Provided that we are factualists about probability, and in particular interpret non-conditional
credences as genuine beliefs, we can also be factualists about indicative conditionals.

What, incidentally, if we are not factualists about non-conditional credences? Are we then prevented
by Adams’s hypothesis and the Lewis and Carlstrom and Hill arguments from being factualists about
conditional credences? It seems to me not. I mentioned earlier that in this case there are strong
grounds for rejecting the usual definition of conditional probability in terms of non-conditional
probability. A conditional credence cannot be thought of as the possession of a pair of absolute
credences of appropriate content and degree, for in most cases in which we need a conditional
credence (according to Adams’s hypothesis), we simply do not have the relevant absolute credences.

In the (1986) paper in which I offered this argument, I suggested that we should therefore think of
conditional credences as what I called ‘inferential dispositions’: dispositions to adopt a consequent
credence of a certain degree, on acquiring an antecedent belief. This is not the only possibility. For
example, Ramsey (1978, chapter 3) thought of both conditional and absolute credences as theoretical
psychological entities, to be characterized in the first instance in terms of their effects on behaviour.
Both can be measured in terms of choice of betting odds. The difference is that conditional credences
call for conditional bets.

The various possibilities agree on one thing, however. The equality 5.1, normally used to define
conditional probability, is no longer sacrosanct. At best, it is likely to be what it is for Ramsey: a
derived result about the relationship between the absolute and conditional credences of a rational
agent. The present significance of this new attitude to 5.1 lies in the fact that it suggests another
answer to the arguments of Lewis and Carlstrom and Hill. In so far as these arguments rely on the
usual definition of conditional probability, it is open to someone who rejects this definition – instead
regarding conditional credences as an independent class of psychological entities – to object that
Lewis and Carlstrom and Hill misuse what is at best a contingent connection between conditional and
absolute credences.

The objection is actually more appropriate for the Carlstrom and Hill argument than for Lewis’s. We
saw that Lewis’s argument can be expressed without reference to 5.1. In the Carlstrom and Hill case,
however, 5.1 is crucial. In this case the claimed absurdity was just that it should be possible for
different systems of belief to assign the same probability to A and A&C, and yet different values to
P(A®C) and hence to P(C/A). But if 5.1 is at best dependent on an assumption of rationality, then
there is nothing contradictory in the possible existence of such systems of belief. (Should irrationality
be impossible?) There might be a contradiction if the Carlstrom and Hill argument showed us that
rational systems of belief could differ in this way, but the argument offers nothing of this kind.

One might think – I speak here from experience, as my (1986) paper shows – that in taking a non-
factualist view of absolute credences, and hence being led to view conditional credence as a primitive
psychological category, one thereby adopts a non-factualist view of conditionals themselves. But on
the contrary, it now appears that a primitive view of conditional credence in fact makes things easier
for a factualist account of conditionals. Such a view can already help itself to our earlier responses to

65
the Lewis and Carlstrom and Hill arguments; and now, in virtue of its new attitude to 5.1, has another
very promising objection to the latter argument.

If we characterize conditional credences as inferential dispositions, or something similar, then at first

sight it seems that we assign them a different role from belief in a functionalist theory of the mind.
Conditional credences become dispositions to add to our beliefs – to move from one belief to another.
But at second sight we should appreciate that there is nothing in the functionalist theory to show that
this functional role is not itself performed by a class of beliefs. It is the familiar problem: as
psychologically inclined non-factualists, we must do more than show that a class of utterances are the
expressions of mental states with a special functional role; we must exclude the possibility that the
functional distinction in question marks a difference in attitude content, rather than a difference in
attitude type.

66
PART II

Truth and Its Limits: An Explanatory View

67
6
The Function of Truth

For the benefit of readers who may have just joined us, let me summarize events so far. We have
looked at a range of attempts to analyse the notion of statementhood – to say what it is about an
utterance that qualifies it to be ranked in the fact-stating part of discourse. However, none of these
attempts seemed able to exclude the possibility that all discourse is fact-stating discourse. None
therefore meets the needs of either of the programmes we identified in chapter 1 – of non-factualists
or marker theorists, who have a common interest in a non-trivial fact-stating/non-fact-stating
distinction, even if conflicting views as to where it should be drawn.

One option at this point would be to walk away from the problem, concluding that non-factualists and
marker theorists are simply misguided in expecting language to divide in this way. I think this would
be premature, however. Before we blame the problem for the failure of the analytic solution, we
should consider the possibility of other solutions. In the remainder of the book I want to argue for
another approach. I want to show that so long as we concentrate on explanation rather than analysis,
we can find considerable illumination in the concerns of non-factualism.

As I indicated in chapter 1, this explanatory approach will be founded on a novel theory of truth. The
attractions of this theory are not limited to its ability to make sense of non-factualism, in my view,
significant as I think this is. So, as a guide particularly to the reader whose interest is in the more
general issues, let me indicate the structure of the remaining four chapters.

In this chapter I first characterize the explanatory approach, abstractly and by example; and then try to
head off some objections to its application to a linguistic distinction of the kind we are interested in. I
shall talk at this stage of the fact-stating/non-fact-stating distinction, but parallel remarks would apply
to a distinction between the truth-bearing and non-truth-bearing parts of language. In any case, I then
concentrate on truth, proposing an account of its function in discourse. The chapter concludes with
some remarks on the relation between this account and other recent ‘explanatory’ approaches to truth.

Chapter 7 begins with a defence of the proposed account against a number of objections. From section
7.6 I turn to the account’s application to non-factualism. Chapter 8 continues this topic, mentioning
also the non-indicative moods. Chapter 9, finally, returns to general issues. I discuss the connection
between the proposed account of truth and Fregean general semantics, and relate it to other recent
positions on truth, realism and objectivity.

6.1 EXPLANATION WITHOUT ANALYSIS

As theorists we often find ourselves with the task of explaining distinctions marked in the ordinary
lives of ourselves and others. We notice that in cases of a certain kind, our subjects treat some
instances in one way, others in another. Sometimes the behavioural difference will amount to
application or non-application of a descriptive term. Sometimes it might consist in some other aspect
of our linguistic behaviour. Or sometimes it might be entirely non-linguistic. Whichever it is,
however, these behavioural distinctions call for explanation. Such patterns do not seem arbitrary – we
want to know how they arise.

One sort of explanation tries to identify some feature of the two sorts of case, to which our subjects’
behaviour can be seen as responsive. This is the analytic or reductive approach. On this view, to put it
crudely, their choice between behaviour X and behaviour Y is governed by their ability to recognize
the presence of certain properties PX and PY, and their conformity to the rule that they do X when

68
they detect PX and Y when they detect PY. In the particular case in which the behavioural difference
amounts to the application or non-application of a descriptive term, this amounts to the view that they
use the term to refer to one of the properties in question; its use again being dependent on their ability
to detect such a property in their environment.

An analytic account should thus be explanatory; the appeal to underlying properties should be capable
of accounting for the behavioural difference with which we began. An adequate explanatory account
need not be analytic, however. There are ways of accounting for behavioural distinctions that do not
depend on an appeal to recognition (even implicit recognition) of underlying properties. This is
perhaps most obvious in the case of non-linguistic behaviour – that of other species, as well as some
of our own. Why do we tend to fall asleep every evening, for example? Recognition of the time of day
obviously has less to do with it than has the physiological basis of our circadian rhythm.

In linguistic cases it is perhaps less obvious that the appropriate explanation may be non-analytic. All
the same, some such accounts are quite familiar. They arise most naturally from a detached
perspective: that of the anthropologist, say, or the historian of science. For example, we would not
expect a fully recognitional account of the origins of a system of religious or defunct scientific
distinctions; for we want to say that there may be nothing in the world that matches the distinctions
concerned.

Some terminology: from now on I shall use ‘explanatory account’ as an abbreviation for ‘non-analytic
explanatory account’. This is not only shorter, but gets the emphasis right. Explanation comes second
in an analytic account. Explanatory approaches put explanation first.

I want to suggest an explanatory approach to two linguistic distinctions that are very far from defunct;
two distinctions marked in our ordinary use of the notions of truth and falsity. One of these is the
distinction whose analytic treatment part I was concerned to discredit – that between statements and
non-statements, or truth-bearing and non-truth-bearing uses of language. The other is the distinction
we mark within the truth-bearing part of language, in applying the descriptions ‘true’ and ‘false’
selectively. I want to explain the former distinction in terms of an account of the latter. ln other words,
I want to show that we can explain the limits of truth in language in terms of an account of its
function in central cases.

ln both cases the explanation is intended to be non-analytic. I want to show that our use of truth and
its limits can be explained without being taken to reflect or be responsive to any such real properties
of linguistic items. Because these distinctions are so much a part of our own practice, however, the
detached explanatory perspective does not come easily. To soften the target a little, I want to begin
with an example in which there seems to have been an analogous shift from an analytic to an
explanatory approach to a perceived distinction in human life. It is the case of social class.

Social stratification is a striking feature of most societies. It is not simply a theoretical notion,
important only to anthropologists and sociologists, but part of the lives of ordinary people. Many
aspects of a person’s life depend on his or her caste, or class, as it is perceived by the rest of the
community in which he or she lives. Ordinary folk apply such notions to themselves, and the theorist
begins with the notions of folk sociology.

Folk sociology takes folk usage at face value. After all, what is folk sociology except folk usage about
society? Folk usage applies class properties to individuals. Hence it gives the impression that the way
to a theory of class is to analyse these properties – to discover, for example, the distinguishing
characteristics of the aristocrat and the commoner, the prince and the untouchable.

The simplest such theory perhaps takes the analysis no further than the folk class properties
themselves. Class is thus itself regarded as part of the natural order of the world. An individual is seen

69
to belong to a class in much the way that an animal or plant belongs to a species. We now find such a
view absurd and repugnant. We should not underestimate its plausibility, however, particularly in
societies without significant class mobility, and in combination with a suitable account of the origins
of such an order.

A more sophisticated approach, less clearly repudiated in contemporary thought, is to analyse class
distinctions in terms of more basic properties of individuals. So long as these properties are taken to
be changeable, this approach allows for class mobility. Unlike the first, it can also distinguish between
real and apparent class, and hence allow for the class exiles of fiction and fact – lumberjack lords,
Downing Street grocers, and the like. Moreover, it promises an explanation of some of the social
manifestations of class. Indeed, it may seem to offer a justification ofsocial stratification – the idea
that because the rich are better at something, they not only gets but deserves the gravy.

Attractive as they have often seemed, such accounts of class founder on one of three main fallacies.
For a start they may appeal to underlying properties that although genuinely individual, are not
unevenly distributed between the social classes. Appeals to biological properties of individuals are
likely to fail for this reason, for example. Unlike carthorses, the working classes are not innately
adapted to the labouring life.

The second kind of fallacy is to call on properties that although both individual and appropriately
distributed, bear the wrong explanatory relationship to the relevant social distinctions. This can
happen in several ways. The properties concerned may be effects rather than causes of the class
distinctions they are supposed to account for – think of the correlation between access to education
and social class, for example. Alternatively, their correlation might stem from a common cause, or
simply be accidental.

The third fallacy is perhaps the most pervasive, however. It is to appeal to properties that are
themselves constituted in social terms. It would be involved, for example, in an attempt to explain
class distinctions in terms of differential ownership of capital or possession of political power.
Property and power are themselves in some sense social constructs. We may well clarify the problem
of class by describing it in these terms – come to understand that some social distinctions are of more
significance than others. But to clarify a distinction in this way is not to explain how it comes to exist
in the first place.

In rejecting such accounts of class, we recognize, I think, that an adequate theory of class would not
be simply an analysis of the class properties of folk sociology. We distinguish the task of describing
the class structure of a given society from that of explaining the existence of such distinctions. As
noted, the descriptive task may have an analytic component. But we would not expect the explanation
of social stratification to appeal to non-accidental differences between individua ls.

What the explanation of class should appeal to is a matter for some debate. By way of illustration we
might mention two approaches, both in some sense evolutionary. On the first, class would be viewed
as a product of notions of kinship – of their tendency over the generations to accentuate, rigidify and
defend what were originally minor and accidental variations in social role. Classes would condense
like galaxies from clouds of dust; or perhaps more accurately, like organisms from the primordial
soup. On the second view, in contrast, a social class would be compared more to an adaptive variation
than to an organism or species in its own right. Class distinctions would thus be explained in terms of
their functional roles, and hence their contributions to the reproductive success of a community as a
whole.

Whatever the appropriate story, however, the class case makes two important points. It shows that the
use of a distinction that seems initially to reflect intrinsic differences between individuals may turn
out to require a non-analytic style of explanation. And it illustrates that in the grip of the analytic
perspective, we are liable to take as explanations what are at best redescriptions of the distinctions we

70
want to account for. It seems to me that the sorts of views we considered in part I make the same kind
of mistake about the fact-stating/non-fact-stating distinction. Naïve theories of class attempt to
explain class distinctions in terms of intrinsic differences between individuals. Analogously, these
linguistic views try to account for the distinctive features of fact-stating discourse in terms of some
intrinsic property, supposedly distinctive of fact-stating uses of language. As in the case of class, the
approach fails for two main reasons. Either the claimed underlying property is not appropriately
distributed (or cannot be shown to be so distributed, independently of the distinction it is supposed to
explain). Or the supposed intrinsic property is itself a manifestation of the difference of usage that
calls for explanation.

However, I think it must be conceded that although the class analogy might suggest this possibility, it
does little to support it. It is clear, I take it, that the proper approach to class is explanatory; but far
from obvious that the linguistic distinction might even admit the same kind of treatment. This
difference seems to stem from an important disanalogy between the two cases. We incline to the
lottery view of class selection: whatever the origins or function of social stratification, an individual’s
class is decided, in the main, by the accident of birth. The lottery view is much less plausible in the
linguistic case, however. The ‘individuals’ differentiated by the linguistic distinction are speech acts,
linguistic tasks, or something of the kind. They are not biological individuals, whose interests we
know to be often subordinate, from a biological point of view, to those of a species or community.
They compare instead to enduring characteristics of a species. We want to explain some striking
differentiation of the tasks for which we employ language – the distinctions marked in English by the
use of the indicative mood, or the application of the notions of truth and falsity, for example. As a
result, the linguistic distinction seems more tied to an explanatory base than is true of the class case.

It is rather as if we had observed that a community uses two quite different methods of performing the
task of moving themselves from place to place. Sometimes they ‘go on foot’, at other times they ‘take
the Tube’. The community may have little sense that these activities have something in common.
They may not realize that in principle anywhere that can be reached by taking the Tube could also be
reached on foot. As theorists, however, we would expect an explanation of this behaviour not simply
to account for the fact that the movement task is differentiated in this way, but to explain why
particular movement tasks tend to be performed in one way rather than the other. We note for
example that from Highgate to High Barnet, almost all our subjects ‘take the Tube’. We would be
surprised if this were simply an accident – if the habit were not explicable in terms of certain features
of that particular movement task (such as the distance from Highgate to High Barnet, and the
availability of the Northern Line).

In the linguistic case, similarly, we expect that as an enduring feature of human behaviour the use of
the statement form will turn out to be well adapted to the needs of particular linguistic tasks – to the
sociolinguistic function it has presumably developed to serve. I think this explains why the analytic or
reductive approach is less clearly inappropriate here than in the case of class. If an account of fact-
stating discourse needs to appeal to some pre-existing difference between linguistic tasks, in order to
explain the selective application of the features of usage characteristic of statements, then this
difference seems itself to form the basis of a reductive account. For can we not say that statements are
those utterances that are used to perform tasks having the underlying property that explains the
assertoric form of usage?

I think there are three things that may seem to recommend this suggestion. The first is the lingering
appeal of the analytic approaches discussed in part I. In the face of the evidence, our intuitions
doggedly maintain that we can explain the assertoric form in terms of its use in expressing beliefs,
carrying truth values or representing states of affairs. Putting this aside, two factors remain: first that it
is difficult to see how we could explain the development of the assertoric form, except as a
manifestation of the kind of base property that would enable the suggestion to go through; and
secondly that an appealing account of the use of language seems to require that speakers be aware of
some such base property, in order to explain their choice of particular forms of words.

71
In the next section I shall try to cure any lingering attachment to intuitions of this kind. Readers not so
afflicted may move directly to section 6.3. But as an indication of the kind of resolution these factors
allow, consider again our anthropological subjects who ‘go on foot’ or ‘take the Tube’. I said that
although these people might have little conception that these were different ways of doing the same
thing, we would expect to find explanatory differences between the ‘movement tasks’ for which they
typically choose one mode rather than the other. But we would recognize, I think, that the correlation
between these underlying properties and our subject’s choice of transport might be far from perfect.
Being creatures of convention and habit, they might sometimes take the Tube when (as we can see) it
would be better for them to walk, and vice versa. These correlation failures might be either type-type,
as when it would generally be better to make the journey in question by the other means; or type-
token, as when the usual advantage of the chosen mode is outweighed by special circumstances
(frozen points at Camden Town, say). From the explanatory point of view these failures are of little
significance. An evolutionary account relies on long run advantages, and tolerates an element of
historical accident. But clearly they defeat any simple reduction of the acts of taking the Tube and
going on foot to the underlying properties that explain their differentiation in nature. And the same
may be true in the linguistic case. Whatever the usual function of fact-stating or truth-bearing
discourse, we should not be surprised to find cases (tokens and types) in which this function goes
unserved. This possibility prevents any simple analysis of assertion in terms of the underlying
properties being held to account for its form and existence in language.

*6.2 ADAPTION AND LINGUISTIC CHOICE

We want to explore the idea that the statement form has developed to serve some identifiable
sociolinguistic function. We want to resist, however, the resulting pressure to characterize statements
as speech acts that serve this function. I mentioned two intuitions that might seem to favour this
move: first, that it would allow us to explain the development of the assertoric form as a manifestation
of the relevant functional property; and secondly that an adequate account of the use of language
might seem to require that speakers be aware of some such property, in order to explain their choice
of particular forms of words.

Concerning the differentiation of language into assertoric and non-assertoric forms, there seem to be
two possible accounts. One would allow that the development of the assertoric form has been at least
partly guided by speakers’ own intuitions about the appropriate form of their language, in the light of
its functions. The other would insist that such development is essentially ‘blind’, not significantly
influenced by speakers’ understanding.

The latter account seems bound to resort to an evolutionary story, explaining the distinctive
characteristics of assertoric and non-assertoric discourse as the result of a gradual process of adaption
and selection. The same will be true of the former account, to whatever extent it relies on ‘blind’
development. In neither case, of course, need ‘evolutionary’ imply a genetic basis for the variation
and retention of linguistic characteristics. Nor is it clear that linguistic selection need depend on
reproductive success. In humans as in certain other animals, new patterns of behaviour can spread
through a population in very much less than a generation; and hence very much faster than could be
explained by their contribution to fertility.

All the same, the likely evolutionary character of the development of language does a lot to
undermine the intuitions that seemed to favour the analytic approach. For one thing, an evolutionary
theory is capable of explaining the development of a behavioural distinction, without treating it as the
manifestation of some pre-existing distinction. This is one of the great achievements of Darwinian
theory: the fact that it can account for functional differentiation in nature, without taking the
development of species and other biological kinds to consist, as it were, in the teleological
colonization of previously vacant roles. On the contrary, the process of development creates the roles
as it goes along. Hence the vacuity of the doctrine, sometimes advanced, that evolution ‘fills every

72
available niche’. Within certain physical constraints, evolution creates the niches as it creates their
occupants.

Take the case closest to hand. Darwinism frees us of the idea that the human hand is a directed
solution to a pre-existing functional problem – an idea embodied not only in ‘designer’ theories of
creation, but also in the Lamarkian view that adaption occurs as organisms strive to reach certain
goals. After Darwin we can say in all seriousness that the functional role of the hand – its biological
niche – is the performance of manual tasks. This truth is as trivial as it seems. Only a teleological
view requires that the functions of the hand have anything more in common than the fact that they are
the functions of the hand. For a Darwinian the interesting truths lie elsewhere: in the story concerning
the development of the hand from its precursors, and the contributions of its developing functions to
survival and success. And so much, I think, for the feeling that an analytic account is required to
explain the development of the assertoric form of usage. It rests on a failure to appreciate the potential
of non-teleological accounts of the development of biological characteristics.

The second intuition also rests on a teleological viewpoint, but at a different level. Whereas the first
intuition concerned the development of forms of usage themselves, the second has to do with their
application on individual occasions of use. It stems from the feeling that any significant feature of an
utterance should in principle be explicable in terms, in part, of the speaker’s understanding of the
relevant part of the language. We think of the choice of a particular form of words as an intentional
action, and therefore as something that ought to be explicable, in part, in terms of the speaker’s beliefs
about the significance of those words. To accept otherwise would seem to be to consign the relevant
feature of linguistic competence to the ranks of the instinctive, the habitual and the ‘wooden’
(Blackburn 1984, p. 96).

I think we are thus inclined to think of the choice of the assertoric form as something like the choice
between synonyms – what goes on when a competent speaker chooses to say ‘Tube’ in preference to
‘Underground’, ‘Metro’, or ‘Subway’, for example. Such a choice is presumably based on a prior
understanding of the nature of the speech act in which the chosen word will play a part. It is difficult
to shake off the idea that the ‘choice’ of the indicative mood and other features of the assertoric form
is similarly based on an implicit grasp of an underlying property of an intended communicative act.
Since such an underlying property would seem to provide the basis for an analysis of the notion of
statement of fact, this ‘choice’ model of utterance lends credence to the analytic approach.

Appealing as the choice model may seem, it cannot be applied without restriction. Unlike Humpty
Dumpty, we find our choice of words constrained by our intended meaning. Of course, these
constraints usually allow us a good deal of freedom. Thanks to the existence of synonyms, there are
usually many ways of ‘saying the same thing’. But the fact that there are such constraints means that
we may in fact have no choice about the assertoric form, and hence no need to be aware of an
underlying property in virtue of which that form is appropriate.

It might be objected that even where language gives us no choice, we have in some sense to know our
intended meaning, in order to be in a position to determine the correct words for the job. If so, then
linguistic intentions are once again prior to the selection of the assertoric form. But there are big
problems for such an account. In particular, it seems to depend on the existence of judgements that are
not linguistic in form. For what does a speaker conclude, when he judges that the sentence ‘P’ is the
appropriate way to express what he means to say? That ‘P’ expresses the proposition that P? This
judgement is not tautologous, as of course we discover if we translate it into a language other than
English. But it is not the sort of thing that a competent English speaker could be in any doubt about.
So to the extent that reasoning is conducted in English, this judgement cannot be a significant
prerequisite for the decision to say ‘P’.

What if we move to another language – perhaps the language of thought? Then we do have a non-
trivial judgement, of the kind of thing that a would-be English speaker would have to know. Let us

73
suppose that in the language of thought, this judgement is expressed by ‘Q’. In other words, we are
supposing

R: In the language of thought, ‘Q’ expresses the proposition that in English, ‘P’ expresses the
proposition that P.

The suggestion is that in order to account for our speaker’s utterance of ‘P’, we have to admit that he
judged that ‘P’ is an appropriate way of saying that P. Since this judgement is trivial in English, we
are supposing that he expresses it in another language – the language of thought, for example.
(Interposing other languages at this point would only delay the inevitable.) Presumably he makes this
judgement by tokening ‘Q’ in an appropriate ‘judgement register’. In effect, he says ‘Q’ to himself.
But why ‘Q’? Is this choice guided by the judgement that R? If so, then the problem recurs again. If
not, then why is there a problem in the first place?

Thus at some point this view seems committed to the existence of non-linguistic judgements. This
takes the view perilously close to the instinctive, habitual and wooden. The mark of ‘non-
woodenness’ seems to be our ability to reflect on our own beliefs and practices. But how can we
reflect on what we cannot express? In other words, it looks as if the preference for the choice model
of the use of the assertoric form (and hence for an analytic account of statement of fact) rests on an
unattainable ideal. In the use of language, as elsewhere, our freedom of choice cannot be boundless.
Our choices are always constrained by the terms in which the options present themselves. We cannot
choose what we cannot conceptualize. The above argument suggests that in the case of ‘choosing
one’s words’, the options present themselves in linguistic form. Choosing one’s words is not a matter
of first settling on the particular (fully specified) speech act one wants to perform, and secondly
selecting the appropriate words for the job. There is only one kind of choice; though it may go in
several steps, as one refines one’s intentions, in the process of settling on a particular form of words.

There is a nice parallel here with accounts of the origin of life. Like teleological misinterpretations of
Darwin, creation myths tend to embody an unwarranted conceptual distinction between a biological
kind and its ‘place’ or niche in nature. For example, the biblical account may leave us with the
impression that when God created a species, He made two decisions: to include a particular niche in
the Great Plan, and to create the creature that fills that niche. These decisions might be taken in either
order, and we find suggestions of both orderings in the biblical account. The more obviously absurd is
that of creation before assignation to a biological role. This is suggested, for example, by God’s
famous injunction – to creatures already equipped, presumably, with their reproductive instincts and
apparatus – to ‘Go forth and multiply.’ The reverse ordering is suggested, for example, by the account
of the creation of Eve. We are given the impression that God first decides to provide Adam with a
mate, and then decides to create Woman for that purpose. But there is room for two decisions here
only to the extent that the first decision is incomplete. God might have initially decided that Adam
should have a mate without making up His mind on some of the details. (Should Eve eat Adam after
mating, for example?) This leaves room for a series of decisions. However, to the extent that God has
filled in such details on the Great Plan, the actual creation of Woman is not a separate decision but
merely the fulfilment of a prior intention. A place in the Great Plan is delineated by the properties of
its occupant. To decide to fill the place is thus to decide to create the occupant.

Analogously, the above view of linguistic competence involves what amounts to a two choice account
of the ‘creation’ of an utterance: the idea that in deciding to speak, we first decide what kind of
linguistic act we wish to produce – what niche we want to fill – and secondly choose the appropriate
sentence for the job. It seems to me that this model involves the same sort of misconception as the
two-choice view of the creation of life. Just as the common feature of the hand’s activities is simply
that they are manual, so what unifies the speech acts for which we use the statement form is just that
they are utterances in the statement form. We cannot choose the statement form on the basis that we
recognize that an intended speech act belongs to a certain kind, because to get that far is already to
have made a decision about the use of the statement form.

74
Given two sentences of whatever form, we can of course choose to utter one, both, or neither. In
doing so we rely, presumably, on some conception of the effect of an utterance of either kind. What
we must resist, I think, is the idea that this understanding of the role of a potential utterance is
somehow independent of our awareness of the linguistic expressions whose use is at issue. In the end,
we must live with the fact that linguistic acts present themselves to us as linguistic acts, and not in
terms of some pre-linguistic characterization.

All the same, in this and the next two chapters it will sometimes be convenient to speak as if we took
it that language had been developed to serve a range of pre-defined functions; in particular, to express
a range of different sorts of mental states – beliefs and suchlike. As is often the case in biology, a
teleological perspective will thus be a useful expository device. This psychological scaffolding will be
thoroughly dispensable, however. The work will be done by the notion of behavioural dispositions,
and we won’t need to assume that such dispositions serve to delineate the familiar psychological
categories – in advance, as it were, of the linguistic distinctions we shall use them to explain.

6.3 TRUTH: THE EXPLANATORY PERSPECTIVE

In chapter 2 I mentioned Strawson’s (1949) ‘amen’ account of truth, which makes a theory out of the
observation that one function of the term ‘true’ is to provide a uniform means of endorsing a
statement made by another speaker. Strawson argues that the description ‘true’ is primarily a device to
confirm or corroborate a previous utterance. He doesn’t say what ‘false’ is, but there are at least two
possibilities: that it ‘confirms’ the negation of the original utterance; or, without assuming an account
of negation, that it indicates some sort of rejection or ‘disconfirmation’.

Strawson’s theory is not an analysis of truth. It is rather an account of the function of the term ‘true’
in discourse – an account that does not presuppose that the function of such a term must be primarily
referential. Strawson does not tell us what truth is, but offers a theory as to why we should talk as if
there were such a property. In our terms it is thus an explanatory theory of truth.

It is not a very satisfactory theory, however. For we noted that if corroboration were the main function
of ‘true’ it would be unclear why the term should not be applied more widely – to questions and
commands, for example. It would often be useful to have a simple device for putting one’s weight
behind a question asked by someone else: for indicating, in effect, that one is asking the same
question oneself. Similarly, there are cases in which it would be useful to be able to say in a simple
way that a question did not have one’s endorsement. The same is true of commands, requests, indeed
of practically any sort of utterance. This amen theory of truth does not explain why we should be
allowed the conversational convenience in some cases but not in others. So if a functional theory of
truth is to account for the non-indicative moods, the function of ‘That’s true’ had better be more than
that of ‘Amen’.

A more promising approach lies in the observation that to call an utterance ‘true? is to say that it is
correct, or right; to call it ‘false’ is to say that it is incorrect, or wrong. As Dummett puts it, ‘the roots
of the notions of truth and falsity lie in the distinction between a speaker’s being, objectively, right or
wrong in what he says when he makes an assertion’ (1978, p. xvii). We might be wary of Dummett’ s
use of the notions of assertion and objectivity – these are the sorts of notions we want to explain – but
the emphasis on the normativity of judgements of truth and falsity is a step in the right direction.

Whatever their origins, normative terms seem to function as verbal carrots and sticks, with which we
can reward and punish certain sorts of behaviour. As with more tangible incentives, such rewards may
be actual (for past behaviour) or merely conditional (on future behaviour). But unlike that of carrots
and sticks, the reward or punishment bestowed in a normative judgement is in a sense self-inflicted.
This not because we ourselves wield these verbal implements – that is true of carrots and sticks – but
because their effect is conditioned and conventional. A child who has not learnt the meaning of
‘good’ and ‘bad’ cannot be encouraged to modify his behaviour by being told that it is good to eat his

75
breakfast and bad to jump on the cat’s tail (which is why parents are often tempted to resort to less
convention-governed incentives).

Non-verbal incentives can also depend on convention, of course. A child who has not learnt the
relevant conventions will not be moved by the offer of a fiver if he eats his breakfast. And just as we
might be taught to be ‘bad’, or to regard ‘Everything you say is false’ as a pat on the back, so we
might learn to attach monetary value to almost any kind of physical object.

In general, at any rate, the use of such conventional rewards and punishments presumably serves
some point. Why else should it have developed? In the case that interests us, we might hope and
expect that the point of use of the normative terms ‘true’ and ‘false’ is such as to explain the ordinary
limits of their application. We might hope to show, in other words, that the application of these terms
to assertions serves a purpose that would not be served by applying them to, for example, questions
and commands.

Again, we would thus be explaining ‘true’ and ‘false’ not analytically, but simply in terms of their
functional role. We would be asking not what these terms refer to, but rather what they do for us (and
hence why we might have developed them). This approach would be uncontroversial for a notion such
as monetary value (or the valuations inherent in class distinctions, to return to our earlier example).
However, it is apt to seem misguided in the case of normative notions such as truth, moral value and
rationality. In the monetary case, we recognize that monetary value is conventional not simply as
regards the objects we choose to attach it to, but also in the sense that the very notion of monetary
value is clearly a social product (and hence conventional). But we are reluctant to allow that
normative terms involve both these kinds of conventionality. The first kind is obvious enough: it is
simply the point that we could have meant by ‘bad’ what we presently mean by ‘good’. But we are
inclined to assimilate this with the point that, for example, we could have meant by ‘kipper’ what we
actually mean by ‘sardine’. That is, we resist the suggestion that what we would be talking about in
using these different words is also a social construct, rather than a natural feature of the world.

Accordingly, I think, we consign ourselves to some of the great treadmills of Western philosophy –
the attempts to analyse the concepts of Truth, Reason, and the rest, and hence to understand their
normativity and the role they play in our lives. These well-worn steps continue to seem more
attractive than the conventionalist alternatives. A striking indication of the depth of this preference
lies in the tendency on the part of philosophers who abandon the attempt to analyse truth in terms of
correspondence to reality, to seek to analyse it in some other terms. We thus have the pragmatist
tradition of coherence accounts of truth, of attempts to identify ‘true’ with ‘useful’, ‘accepted at the
end of inquiry’, and the like. True, in the pragmatist tradition we also find the view that philosophy
should not attempt to analyse truth. But even here there seems to be reluctance to take on the task of
explaining our use of the notion of truth. Instead such pragmatists ‘would simply like to change the
subject’ – to ‘suggest that we do not ask questions about the nature of Truth’ (Rorty 1982, p. xiv).
This is the quietist tendency I referred to earlier, to blame the problem for the failure of the analytic
solution. We shall return to it in chapter 9; but it is further evidence of the near invisibility of the
explanatory alternative in contemporary thought.

Perhaps we suspect that in taking the explanatory approach, the conventionalist buys the easy life at
the cost of what gives point to life, namely the very values in question. In other words, we fear that if
we allow ourselves the explanatory perspective, allow ourselves to ask why we talk in these terms, we
lose our entitlement to use them, to continue to behave as if they matter. We can cheerfully regard
money as a social convention, because we could countenance getting by without it. But how could we
be philosophers, or indeed persons, without a sense of the reality of Truth, Reason and the Good?

At best, this could amount to an argument for not attempting to develop explanatory account of
normative terms – a version of the argument that there are some questions it is safer not to ask. Even
so, I think it exaggerates the risk. In other areas of life, we seem to be able to combine an awareness

76
of the functional role of appetites and values with an undiminished ability to feel their force.
Abstinence does not seem to be assisted by the knowledge that one’s craving has a biological basis. It
might be objected that there is a difference between these cases and those involving the use of
normative terms such as ‘true’ and ‘good’: namely, that since the latter depend on linguistic
convention, it is easier for us to give them up, or to cease to feel their force. But if the use of the word
‘convention’ is thought to carry this implication, we can do without it. The explanatory approach need
not claim that with the benefit of an explanation of our use of these normative terms, we could (let
alone should) do anything to change the practice. Some linguistic practices may perhaps be accessible
to revision. But many, perhaps most, may simply be wired into us, in circuitry over which we have no
control.

6.4 THE CASE FOR SEMANTIC INCENTIVES

We have digressed, however. The suggestion was that we regard the normative notions of truth and
falsity as the core of a conventional system of punishment and reward, and hence try to explain the
limits of their application. In other words, we should try to show that the application of these terms to
assertions serves a purpose that would not be served by applying them for example to questions and
commands.

What purpose could this be? I suggest that the normative character of the notions of truth and falsity
provides speakers with an incentive to justify their utterances to others, and hence to participate in an
important form of argument or dispute. At its crudest level, it ensures that speakers are rewarded
when their utterances are endorsed by others, and punished when their utterances are rejected. Thus it
gives everyone an interest in participating in and attempting to ‘win’ the relevant disputes. It gives us
all an interest in persuading others to endorse our own utterances (and hence, of course, in making
sure that our own utterances are such as to be endorsed by others).

The normative nature of the terms in which such disputes are conducted gives them short term value
to individuals, but cannot constitute their long term value to the linguistic community as a whole. On
the contrary, the development of such a system of linguistic incentives suggests an underlying
behavioural advantage – something that would explain the linguistic development in evolutionary
terms. This advantage presumably turns on the behavioural consequences of the mental states exposed
to criticism in such linguistic disputes. In particular it might lie in the survival advantages of basing
one’s mental attitudes on as wide a body of experience as possible.

Rough as it is, the approach already suggests an explanation for the fact which embarrassed
Strawson’s theory – namely that ‘true’ and ‘false’ can be used to endorse and reject only certain sorts
of utterances. Perhaps not all utterances serve a function such that there is reason to seek or expect
agreement between speakers. Consider requests, for example. Imagine that the two of us are talking to
a third person, Fred. I ask Fred to buy us each a cup of coffee. Fred looks at you, but you say nothing.
Our utterances on the subject conflict, in the sense that no one person could have both asked and not
asked Fred to buy a cup of coffee. But clearly this ‘conflict’ may be entirely benign. There may be no
reason to expect us to make the same requests, of Fred, or in general – and hence no advantage (quite
the contrary) in a linguistic device that would encourage us to do so.

The proposal is thus that some utterances normally express states of mind with respect to which there
is survival value in encouraging speakers to dispute, in the above sense. For these states of mind, two
heads are better than one, and hence there is a general advantage in exhibiting differences, and in
providing incentives to encourage the spread of well-justified beliefs. The use of the normative terms
‘true’ and ‘false’ is an important part of the apparatus that language has developed to exploit this
advantage. However, the apparatus is not deployed where it is generally not appropriate. The roles of
utterances such as questions, commands and requests are not such as to provide an advantage in
seeking unanimity. On the contrary, these utterances are such that different speakers may reasonably
conflict (in the above sense), even if fully acquainted with each other’s point of view. So ordinary

77
usage does not provide the incentives for disputes in such cases – perhaps marking their immunity
with the non-indicative moods.

We thus have the beginnings of a functional account of truth, grounded on the claim that its primary
role is to encourage and facilitate a useful form of linguistic dispute. The present sketch leaves a lot to
be desired, of course. How, for example, is the claimed difference between indicatives and non-
indicatives to be reconciled with the results of chapter 3 – with the apparent possibility of treating
non-indicatives as disguised assertions? I shall return to the non-indicatives, and to this query in
particular, in chapter 8.

It may seem also that the proposed account leaves no room for non-factualism. An interesting non-
factualist thesis is bound to concern a case in which utterances are ordinarily treated as true and false.
The non-factualist claims that although such utterances appear to be statements of fact, they are really
something else. But the non-analytic approach leaves no room for a distinction between appearance
and reality in this case – no room for categories underlying use in terms of which the non-factualist’s
claim could be expressed. We shall see, however, that the account leaves room for a distinction of a
different kind. In particular, the complexities of the relevant form of linguistic dispute give rise to a
range of cases in which the optimum application of the normative notions of truth and falsity is far
from straightforward. The ordinary linguistic apparatus is stretched to its limits, and the strain shows
up as doubts as to whether the utterances in question ‘have truth values’ – whether, in other words,
these notions are applicable at all. Not surprisingly, these are just the cases in which non-factualism
has seemed attractive. Far from excluding non-factualism, this approach thus does a fair job of
explaining its continued appeal.

It is true that the approach entails that traditional non-factualism is misconceived, in taking the
analytic stance – in claiming to discern a categorical distinction underlying use. But in place of this
misconception, the account offers two distinct sorts of question: the first as to whether the use of a
given class of utterances is fully assertoric, or rather involves the identified departures from the
canonical use of truth and falsity; and the second as to whether the actual use is the most appropriate,
in view of the account’s understanding of the function of the notions of truth and falsity. Traditional
non-factualist arguments can be expected to bear on both questions. Indeed, in the next two chapters,
as we examine the structure of linguistic disputes, we’ll repeatedly appeal to the sorts of
considerations that have been taken to support non-factualist treatments of particular topics.

In the next chapter our first job will be to show that the relevant notions of dispute and disagreement
do not themselves depend on the notions of truth and falsity. If we want to explain the notion of truth
in terms of its role in fostering certain activities, our characterization of these activities cannot depend
on truth. Before that, however, I want to distinguish this approach to truth from another approach, also
described as ‘explanatory’, which has received attention in recent years.

*6.5 TRUTH: ANOTHER ‘EXPLANATORY’ PROGRAMME

We have come to the task of explaining truth through that of accounting for the distinction between
fact-stating and non-fact-stating discourse. We are interested in explaining rather than analysing truth
because we have concluded that unless an analysis of truth draws on a prior understanding of the
nature of statement, then, like Strawson’s amen theory and the disquotational accounts of truth, it will
be unable to explain the fact that not all utterances are treated as truth-valued. In explaining the
existence of a more than disquotational (and hence non-universal) notion of truth, we hope to explain,
in effect, the existence of the non-universal linguistic category of statement of fact.

In recent years several writers have noted that although the ordinary notion of truth (or at least the
philosopher’s ordinary notion of truth) seems stronger than a disquotational notion, it is hard to see
why we need a stronger notion. The disquotational notion (or something of similar ‘strength’, such as
a Tarskian truth theory) seems to be all that we need for a theory of linguistic understanding. And in

78
psychological theory, it seems to be the ‘internal’ cognitive role of an attitude, rather than any
‘external’ relation of correspondence with the world, that does the explanatory work. The recent
appreciation of these points, and of their significance for an account of truth, is particularly due to
Hartry Field (1972, 1978).

Field asks ‘why we need to be able to apply semantic notions like truth and reference to the sentences
that people believe and desire’ (1978, p. 48). This question seems to me to involve very much the
same explanatory perspective as the approach suggested above. It sees the task as that of explaining
the ordinary use of the notion of truth. (I think Field’s answer is inadequate, however – more on this
in a moment.) But in the discussion to which Field’s work has given rise, the phrase ‘explanatory
theory of truth’ has come to mean something different. The emphasis has shifted from explanation of
(the use of) the notion of truth, to explanation by means of the notion of truth.

This shift is associated with a slightly different starting point: a philosophical commitment to a
‘realist’ world view, and hence an interest in justifying, rather than merely explaining, the use of the
kind of notion of truth that such a view seems to require. This approach to truth is particularly due to
Putnam (1978), who suggests that a realist notion of truth may be required to account for the fact that
linguistic behaviour contributes to our overall behavioural success: ‘The success of the “language
using program” may well depend on the existence of a suitable correspondence between the words of
language and things, and between the sentences of language and states of affairs’ (1978, p. 100).

Putnam’s approach makes truth a theoretical notion, an element in the kind of story that a theorist
would have to tell about language and its place and nature. Hence there is at least a danger that this
theoretical notion will have nothing to do with the ordinary notion of truth. In what sense is it the
same notion? And even if it is the same notion, why do ordinary speakers need it? These questions
emphasize the difference between an explanatory theory of truth in Putnam’s sense, and a theory that
seeks to explain the ordinary use of ‘true’ and ‘false’. In principle a theory could be explanatory in
both senses. Truth might be a necessary part of an essential folk theory of language and the world.
(Devitt 1984, chapter 6, seems to take this view, though not on Putnam’s lines.) But the two
approaches are independent. Putnam’s does not guarantee ours, and ours does not depend on
Putnam’s – folk notions may be explicable without being explanatory.

All the same, we can distill several proposals for an explanatory theory of truth in our sense from the
work of Putnam, Field, and others. If any of these succeeds in explaining the fact that the ordinary
notion of truth is not merely disquotational, then presumably it will explain (where the disquotational
theory could not) the fact that truth is restricted to a particular sort of discourse. Our own suggestion
would thus be unnecessary, or perhaps repetitious.

Putnam’s own suggestion has been criticized (McGinn 1982, p. 222; Devitt 1984, p. 90) on the
grounds that the explanation of the success of linguistic behaviour does not require a realist or
correspondence notion of truth; a notion of warranted assertibility will do the same job. As Devitt puts
it:

The role claimed for truth in [Putnam’s] explanation must be ... as follows: the expectations
that a belief gives rise to tend not to be disappointed if the belief is true but tend to be
disappointed if it is false. True beliefs tend not to face recalcitrant experience. The problem is
that warranted or justified beliefs tend not to either. (1984, p. 90)

It seems to me that there is a more serious objection, however. Not only does the explanation of
success not need a correspondence notion of truth; all it needs is a disquotational notion. For the
central claim, in effect, is that beliefs tend to lead to success if and only if they are true. And this is
just the sort of generalization that has long been recognized to provide gainful employment for a
disquotational notion of truth. The classic example is Ramsey’s: ‘He is always right’, or ‘Everything
he says is true.’ Ramsey points out (1978, p. 45) that if we allow ourselves to quantify over

79
propositions (or sentences, or some such), we can read this as ‘For all P, if he says that P then P.’
(This and more complicated cases are discussed at length by Grover, Camp and Belnap 1975.)

In the present case the claim at the heart of the appeal to success is something like

(6.1) True beliefs tend to lead to successful behaviour.

Applying Ramsey’s technique gives us

(6.2) For all P, if P then the belief that P tends to lead to successful behaviour.

If 6.2 does indeed mean the same as 6.1, then the notion of truth cannot really be doing any
explanatory work. This is not to deny that we can explain our success by means of the hypothesis that
most of our beliefs are true. It is simply to point out that in the formulation of this hypothesis, ‘true’
may be nothing more than a useful syntactical device, as the disquotational theories claim. This does
not make the hypothesis empty. There is plenty of room for doubt as to whether

(6.3) For most P, if we believe that P then P –

or, as it might be better put, as to whether

(6.4) For all P, if we believe that P then it is probable that P.

If it is any good, the argument from success provides reasons to allay this doubt. But what we thus
have reason to accept does not concern a more than disquotational notion of truth.

Moreover, it seems to me that this objection applies equally to the accounts that McGinn and Devitt
suggest in place of Putnam’s success-based explanation of a realist notion of truth – as it does to the
following suggestion from Field, from which McGinn and Devitt develop their accounts:

I think that the reason why we need to be able to apply semantic notions like truth and
reference to the sentences that people believe and desire is that we hold the theory that
people’s beliefs are, in many circumstances, reliable indicators about the world; and the only
way to state this theory is to use the notion of truth .... Moreover, this theory is not a piece of
gratuitous metaphysics that could easily be dispensed with: it is central to our getting
information about the world, for we are constantly using our opinions about other people’s
beliefs in forming opinions about the world .... These inferences evidently proceed by means
of certain reliability principles, principles that say under what conditions a person’s beliefs
about certain things are likely to be true. (Field 1978, p. 48)

A typical reliability principle is presumably something of the form

(6.5) Everything mother believes is true.

However, we have just noted that there is no need to regard such a use of ‘true’ as anything more than
disquotational. What matters is that

(6.6) For all P, if mother believes that P, then P.

It is true that eliminating apparent references to semantic properties is more difficult when we believe
only that mother is reliable on a limited range of topics. As Field says, ‘we want to be able to say in
the theory that some people have very reliable beliefs about physics but very unreliable beliefs about

80
the state of the economy, and so forth’ (1978, pp. 48–9). If mother knows her onions, we may want to
say that

(6.7) Everything that mother believes about onions is true.

This becomes

(6.8) For all P, if mother believes that P and P is about onions then P.

But we are still left with the notion of aboutness, that Field takes to require a theory of reference.
However, there is scope for treating ‘about’, like ‘true’, as an eliminable syntactical convenience. If
we allow ourselves to quantify over predicates, for example, we can get close to 6.7 with

(6.9) For all F, if mother believes that onions are F then onions are F.

So it looks as though Field’s reliability principles require neither truth nor reference, in the realist’s
sense.

McGinn and Devitt both modify Field’s proposal by placing more stress on a broader notion of
communication. Devitt says that we need to explain the fact that speakers both learn from and teach
each other, and that ‘it is these explanations that require talk of truth’ (1984, p. 92). In so far as his
account concerns learning, Devitt follows Field, emphasizing our need for judgements about the truth
of the beliefs of others. As I have argued, these judgements need no more than a disquotational notion
of truth (which is to say that they don’t need truth at all, so long as they are prepared to put up with
some syntactical inconvenience).

Concerning teaching, Devitt notes also that a speaker S must believe that an audience A will take an
utterance or other sign as an indication that S has a certain belief (that it is raining, in Devitt’s
example). What matters is that A

takes the behaviour to ‘mean that’ it is raining; or, better, he takes it to be true if and only if it
is raining. He takes S to be sincere, and so infers ... that S has a belief that is true if and only if
it is raining. He takes S to be reliable and so comes to believe that it is raining. (1984, p. 93)

Putting aside reliability, the question is whether A’s belief about the significance of S’s behaviour (or
S’s belief about that belief) needs to mention non-disquotational truth. If we are allowing non-
linguistic behaviour, it is surely misleading to say that A must believe that S ‘means that’ it is raining.
If S means that it is raining then he is saying so, and his behaviour amounts to a speech act. In this
case what A must believe is first that S has said that it is raining; and secondly (since A believes S
sincere) that S believes it is raining. The non-linguistic case requires even less: simply the belief that
the behaviour in question indicates that S believes that it is raining. Thus, there doesn’t seem to be a
role even for a disquotational use of truth.

It may be that the point Devitt is aiming for here is the one suggested by McGinn. McGinn proposes

that we locate reference in the point of communication – in the intentions with which
assertions (and other kinds of speech act) are made. A hearer understands a speech act as an
assertion just if he interprets it as performed with a certain point or intention – viz. to convey
information about the world. ... To fulfil the intention to communicate knowledge (or belief)
about the world in language the speaker must exploit signs standing in representational
relations to things in the world. (1982, p. 226)

81
However, this suggestion seems to involve the view we rejected in chapter 2: namely, that assertions
are distinguished from other kinds of speech act in virtue of being ‘aimed at truth’. McGinn seems to
be proposing that in order to understand the point of an assertion, we must see it as made with such an
aim; and hence that we require a notion of truth, thought of in terms of ‘representational relations to
things in the world’. The lesson of chapter 2 is that as an explication of the notion of assertion, this
gets the relationship between truth and assertion the wrong way round. If we try to characterize
assertion in terms of the notion of being aimed at truth, we cannot say why not all utterances should
be regarded as assertions.

The conclusion of chapter 2 was thus that in order to understand truth, we must first understand the
nature of assertion. That is, we need to be in a position to understand why it is assertions, particularly,
that admit the standards of correctness and incorrectness embodied in the notions of truth and falsity.
At the time, the possibility remained open that this called for an independent analysis of the notion of
assertion, on the basis of which we could hope to explain the function of ‘true’ and ‘false’. We have
now rejected the analytic approach, however. As a result, we have to think not so much of the role of
truth in the individual case, as of the function of the assertoric form of discourse as a whole. We have
to think of truth not as a property uniquely possessed by linguistic entities of a certain kind, but as the
most striking component in a pattern of linguistic behaviour, whose significance we want to explain.
Instead of explaining truth in terms of assertion, we want to explain the assertoric pattern itself, of
which truth seems simply the most striking facet.

Another part of the same pattern is perhaps the notion of representation itself: the very idea of ‘signs
standing in representational relations to things in the world’, as McGinn puts it. Traditionally the
notion of representation has seemed to be prior to truth. Truth has been viewed as correct
representation. In order of development I think that these notions might be better taken the other way.
Truth might arise first, in virtue of its role in argument; and representation follow, in virtue of our
attempts to understand the nature of the ‘correctness’ of true beliefs and utterances. However, the
important point is that explaining truth and explaining our idea that signs represent the world are part
of the same enterprise. So although it is true of truth, as McGinn says of reference, ‘that finding a
place for [it] forces us to take a world-directed view of communication’ (1982, p. 226), this is not the
basis of an explanation of truth, but simply a comment on the nature of the explanatory task.

Thus it seems to me that the recent search for an ‘explanatory theory of truth’, stemming from the
work of Field, does nothing to provide an explanation of truth in our sense. Nor, incidentally, does it
seem to make any progress towards a theory that would be explanatory in Putnam’s sense – that is, in
which truth is itself an explanatory notion. As we saw, the purported applications of such a notion
seem to need nothing more than the disquotational property of truth.

82
7
The Power of Negative Thinking
I have suggested that we try to explain the existence and application of notions of truth and falsity in
terms of their functional role. In particular, I suggested that their main function might be to encourage
a useful form of linguistic behaviour, namely reasoned argument. It seems that in virtue of their
normativity, these notions enable and encourage us to favour opinions with which we agree and to
disfavour opinions with which we disagree. Disagreements thus become socially unstable, and we are
led to defend our opinions, to argue and to seek agreement. In the long run, I suggested, such disputes
can be expected to have a beneficial effect on the behavioural dispositions of individual speakers.
They help to ensure that as individuals we hold and act on attitudes that reflect, to some extent, the
combined wisdom of our linguistic community. Our behavioural dispositions can thus be tested
against those of other speakers, before they are put to use in the world. The guiding principle is that it
is better to be criticized for claiming that tigers are harmless than to discover one’s mistake in the
flesh.

Obviously, the notions of disagreement and dispute do a lot of work in an account of this kind. A
natural suspicion is that these notions themselves need to be explained in terms of truth and falsity.
This would put paid to the proposal to base a functional account of truth on its role in promoting such
disputes. In this chapter our first task will be to explore this objection. We shall see that there are
several forms it might take, but that none is inescapable. In the process we shall be led to a clearer
account of the origins and conduct of such disputes and a better understanding of their behavioural
advantages. That in turn will prepare the ground for the chapter’s second main concern: to show how
this functional approach to truth leaves room for non-factualism.

7.1 TRUTH AND THE TRIGGERS OF DISSENT

In its simplest form, the circularity objection would be the claim that a disagreement (of the relevant
sort, at any rate) about an assertion P is a disagreement about the truth of P. However, in this form the
objection is easily dismissed: for a dispute about the truth of P would have to be a dispute about the
truth of ‘P is true’, and so on, ad infinitum.

A more interesting form of the circularity objection: How is it that we recognize the existence of a
disagreement (and hence the grounds for a dispute) with another speaker, if not in virtue of the
perception that he or she has said something false? The proposed account claims to explain our use of
‘true’ and ‘false’, in terms, primarily, of their role in these disputes. What it claims to explain, in other
words, is our habit of describing the grounds for argument as being that our fellow disputee has said
something false. The claim is that our use of this normative term serves the function of providing an
incentive to the other speaker to enter into a dispute.

The account thus aims to explain the ordinary speaker’s perception of the basis of a disagreement: the
perception that someone else has spoken falsely. In order to explain it, however, the account needs to
appeal to some more basic feature of the cases in which it takes this perception to be appropriate – to
say that (perhaps unconsciously) we recognize the grounds for a dispute in terms more basic than the
perception that someone else has spoken falsely. Similarly, the cue that a dispute is resolved cannot
simply be the perception that the other disputee now speaks the truth, for the account wants to explain
our inclination to describe it in these terms. So what could be the cues of argument, if not our
judgements concerning the truth and falsity of the utterances of others?

A note here on terminology: I take dissent to be analogous to judgement rather than to assertion – that
is, it happens in the head. A dispute begins when someone expresses dissent. Presumably some of the

83
utility of truth and falsity lies in their encouraging dissenters to speak up. But even silent dissenters
may profit from the experience, by reconsidering their own beliefs.

A speaker makes an assertion, another expresses dissent, and a dispute is under way. We want to
know how the latter speaker came to believe that dissent was appropriate. And we want to avoid the
suggestion that it is a matter of coming to believe that the former speaker spoke falsely.

The mistake thus avoided would be that of pitching the cues of dissent too high – of making them
depend on perceptions of truth and falsity. We should also be careful not to pitch them too low. We
want to say that the biological utility of linguistic disputes stems from the behavioural consequences
of the states of mind that come under challenge (albeit indirectly, in the form of their linguistic
proxies) in such a dispute. So it is important that dissent be cued to something of behavioural
significance. However, these behavioural consequences are long run affairs, and no doubt hideously
complicated. It would be absurd to require that in individual disputes, speakers are cued by an
understanding of the behavioural significance of the opposing points of view. Moreover, this would
actually discourage disputes in cases in which the behavioural significance of differing beliefs is
balanced by differing desires: I want to see Fred, and believe he is in the kitchen; you want to avoid
him, and believe he is not in the kitchen; we both head for the kitchen, but one of us will be
disappointed. So it is important that dissent be cued not by an awareness of the behavioural
consequences themselves, but by some manifestation of the mental states that, in the long run, have
those consequences. We need to separate the practice of dissent from the underlying facts that explain
its development.

We want to be able to say that the triggers of dissent lie some- where between these two extremes –
between perception of truth values themselves, and perceptions of behavioural consequences. The
answer might seem obvious: shouldn’t a speaker simply look for cases in which he asserts ‘P’ and
another speaker asserts ‘Not-P’ (or vice versa)? This is too quick, however. For one thing, speakers
may express conflicting points of view in completely different words. There need be no easy
syntactical clues. More importantly, your ‘Not-P’ may not conflict with anything I have said; but if it
conflicts with something I believe (i.e., that P), then I ought to dissent. So it appears that we need to
credit speakers with the competence to discern that an utterance expresses a mental state incompatible
with their own beliefs. There seem to be two components to such a competence. The first is the ability
to interpret an utterance – to see what a speaker means. The second is the ability to detect
incompatibilities between one’s own beliefs on the one side, and a claim embodied in a received
utterance on the other.

On some views of meaning and interpretation, the former ability depends on the notion of truth – for
to interpret an utterance is to know under what conditions it would be true. However, we have seen
that such views seem to require only a thin, disquotational notion of truth. This was the basis of our
objection in chapter 2 to the proposal to derive a substantial theory of truth from a Tarskian truth
theory, constrained by the needs of radical interpretation. So the ability to interpret an utterance does
not seem a source of circularity in the proposal to base a thick notion of truth on its role in
encouraging argument. (I do not mean that it is clear what the notion of meaning and interpretation
will come to under such a proposal; I discuss this in chapter 9.)

This leaves the ability to detect incompatibilities between beliefs and hypotheses. The present view is
not alone in crediting speakers with an ability of this kind. The question is not whether we have such
an ability, but whether it depends on a grasp of the kind of notion of truth we want dispute behaviour
to explain. As far as I can see, such a view could only arise from the idea that in judging that beliefs X
and Y are incompatible, we are really judging that these beliefs could not both be true. This use of
‘true’ is less obviously thin and disquotational than some we have encountered, but only because it is
applied to beliefs rather than to sentences. But why suppose that incompatibility judgements need to
be about beliefs and hypotheses? What speakers need is not the ability to reason about their own
mental states, but rather the ability to employ these mental states in reasoning about the world in

84
general. Prompted by the suggestion that Q, they need not to decide that the hypothesis that Q is
incompatible with their belief that P, but simply to employ their belief that P in deciding that not-Q. It
is a nice question just how we notice such incompatibilities. But it is hard to see why anyone should
think that perceiving that P and Q could not both be true is any more easy, or primitive, than
perceiving that P implies not-Q, or than utilizing one’s belief that P in ‘perceiving’ that not-Q.

Thus for the moment I take it that what triggers dissent from the assertion that Q is simply the belief
that not-Q. (Here ‘belief’ need mean nothing more than ‘disposition to assert’.) This invites the
objection that a grasp of negation itself depends on a grasp of the notions of truth and falsity, along
the standard truth-functional lines. I have some sympathy with this objection, to the extent that I doubt
whether we are entitled to help ourselves to the notion of negation in this way. I think it would be
more accurate to say that negation has its origins in the expression of dissent. However, I doubt that
this makes dissent dependent on falsity. Instead the view might be something like this: in acquiring a
language, we learn to assent to sentences in some circumstances and dissent from them in other
circumstances; dissent, like assent, is triggered by the activation of the sorts of habits thus acquired.
Such a view of linguistic competence has independent advantages. (In Price 1983a I argue that it
enables us to block Dummett’s move from manifestation constraints on linguistic knowledge to a non-
classical logic.) But I cannot defend it here.

7.2 TRUTH AND THE WORTH OF BELIEF

Let us turn to another form of the circularity objection. According to the proposed account, the utility
of argument (and hence our use of ‘true’ and ‘false’) lies in its tendency to ‘improve’ the
commitments with which we meet the world. Thus it rests on the principle that some commitment s
are ‘better’ or more valuable than others. But what is the measure of the worth of a belief, except its
truth or falsity? Isn’t a belief poor, or incorrect, just in case it is false? The objection is thus that in
explaining the function of ‘true’ and ‘false’, the proposed account will have to appeal to the idea that
commitments can be true or false in some prior sense.

The way to meet this objection is to note that on any plausible account, truth cannot be the only
measure of the worth of belief. For imagine for a moment that there were a substantial property of
truth, possessed by certain of our mental states. Why should we aim for beliefs that possess this
property? Not for their own sake, presumably, but because such beliefs are valuable in some other
sense – because they dispose to behavioural success, for example. Vague as this notion is, there is
clearly no simple correlation, on anyone’s view, between truth and behavioural appropriateness. False
beliefs can lead to good outcomes, true beliefs can lead to bad ones, and many beliefs in fact have no
behavioural consequences. Unlike truth, behavioural ‘rightness’ is bound to be a long run matter.

So from the point of view of behavioural utility, true belief is not an end in itself. Truth is at best an
end in virtue of its usual consequences – something we should aim for in virtue of the long run
behavioural advantages of doing so. If one already takes for granted a view of truth as a substantial
property of mental states, this will be the natural way to construe its role in discourse; though it will
need to be explained why this property should be a good guide to behavioural success. This is not the
only possible view, however. Our proposal seeks to explain the ordinary notion of truth in terms of its
role in discourse. Instead of beginning with the conviction that truth is a substantial property, we
begin with the observation that ordinary speakers display such a conviction (and then seek to explain
this fact). We thus regard truth as a ‘mythical’ goal of enquiry, whose popularity is to be accounted
for in terms of the benefits of subscribing to such a myth. Of course, it remains to be shown that the
main characteristics of the notion of truth can be explained on such a basis. However, the one
characteristic that is immediately explained is the fact that truth is treated as the ordinary goal of
enquiry. Here both sides agree on the appearances; and only a prior commitment to a more substantial
view of truth could support the objection that our view fails to make sense of the fact that truth is the
real end that we aim for in revising our beliefs.

85
The above objection calls also for the following reply. In the end, far from construing the benefits of
argument as the production of true beliefs, our account should not depend on the notion of belief at
all. As I said in chapter 6, our use of belief at this stage is no more than an expository convenience. It
is dispensable, in principle, in favour of talk of behavioural dispositions. If this were not so then the
present objection would be rather beside the point. A well-founded notion of belief, or indeed of
representational attitudes in general, would bring with it not only an account of truth but also a basis
for the fact-stating/non-fact-stating distinction. So relying on a substantial notion of belief would be
damning in itself. The worst that could be said of relying on a substantial notion of true belief would
be that it opens the stable door after the horse has bolted.

Thus far in this chapter we have been mainly concerned to show that the notions of disagreement and
dissent do not themselves depend on those of truth and falsity. It is time to pay more attention to
disputes themselves. We are interested in the nature and origins of their advantages, and hence in
showing that in order to exploit these advantages, the goal of argument must have more or less the
characteristics of the ordinary notion of truth. And we want to be on the lookout for the origins of
non-factualism.

7.3 THE VALUE OF ARGUMENT

Arguments arise from disagreement and are resolved by agreement. In some sense, they thus serve to
diminish disagreement among the members of a linguistic community. What could be the point of
such a function? Why should disagreement be collectively or individually disadvantageous?

Let us for a start approach these questions in terms of an orthodox notion of belief. On such a view,
beliefs have truth values, and a disagreement entails that at least one of the speakers involved has a
false belief. This suggests that the disadvantageousness of disagreement rests on that of false belief.
Resolving disagreements helps to correct false beliefs – not infallibly, of course, for a dispute may
lead both parties to false beliefs – but perhaps more often than not.

But what is wrong with false beliefs? Here, as we saw, the answer will presumably be that false
beliefs tend to lead to inappropriate behaviour. ‘Tend to’ is as strong as the claim can go – as we have
seen, it is possible for mistakes at this level to cancel out, in virtue of the holistic nature of decisions
to act. So on the standard view the justification for argument is something like this: verbal
disagreements – for simplicity, pairs of utterances of the forms P and not-P – tend to indicate that one
participant holds a false belief. (This ‘tend to’ stems from several sources, but perhaps most
significantly from the fact that even the most well-intentioned conversationalists may not ‘mean the
same thing’ – express the same belief – by saying P.) False beliefs tend to be a behavioural handicap.
And argument tends to replace false beliefs with true beliefs.

I want to suggest that the orthodox notion of belief – of a truth-valued mental state – plays no
significant role in this account. What matters is a certain correlation between utterance and action in
general: roughly, that within a speech community, utterances of a given sentence tend to be correlated
to mental states with similar causal or functional roles in the determination of behaviour. An utterance
type is thus correlated with a class of mental states that share what I shall call the Same Boat Property
(SBP). A class of mental states have the SBP if their typical behavioural consequences are such that
their behavioural appropriateness, or utility, is predominately similar across a speech community. If a
mental state has the SBP, then if it is appropriate for any one of us, it is appropriate for all – we are all
in the same boat.

The notion of the SBP is far from precise; and not, I suspect, capable of being made significantly
more precise. But provided that we do not expect it to bear too much weight, I think it will be useful
at several points. Later it will help us to explain some of the variations from the statement form that
are associated with non-factualism. More immediately, it clarifies the role of truth in the above
account of the benefits of argument. For given the SBP, we can offer a justification for argument very

86
much like the one outlined above: verbal disagreements tend to indicate that one party is
behaviourally disadvantaged (or potentially so); arguments tend to improve matters. This does not yet
establish that in the earlier version, the standard conception of belief is redundant. After all, it might
be the standard conception that entitles us to the SBP (for the class of mental states typically
correlated with a given utterance). Indeed, it may seem that the required case of the SBP is a trivial
consequence of the truth-conditional view of belief; that beliefs with the same truth condition are
clearly appropriate in the same circumstances – namely, when and only when their truth conditions
obtain. However, this ignores the fact that there is no simple correlation between truth and appropriate
behaviour. So the orthodox account above of the value of agreement does not derive the SBP (from
sameness of truth conditions); it assumes it, in the form of the assumption that throughout a
community, false beliefs tend to be a behavioural liability.

The principle that some mental states have the SBP can be regarded as a theoretical hypothesis. The
account requires that it hold to a sufficient extent, in some form, to explain the utility of a procedure
whose function is to encourage resolution of certain disagreements between the commitments of
different speakers. In the end, if need be, it can derive its plausibility partly from that of the theory
that requires it; and partly from the demonstration that particular violations of the principle turn out to
be correlated with departures in usage from the procedure that it explains.

The step from the hypothesis that certain of our mental states have the SBP to the claim that argument
is advantageous depend s on two main assumptions. One is that this is a case in which individual
interests recommend co-operation, which of course is far from true in general, even within a single
species. But this is hardly likely to be controversial, given that language itself is a co-operative and
social activity. The co-operative basis of dispute behaviour seems simply a special case. (After all, we
are free to avoid disputes – to decline to co-operate – if it seems to our advantage.) More importantly,
the step depends on the assumption that in general, at any rate, pushing for agreement does more good
than harm. Clearly disputes can be counter-productive, leaving both participants with inappropriate
commitments. However, it seems reasonable to assume that on average argument does improve
matters, both by encouraging reconsideration of individual grounds for judgement, and by the effect
of pooling of evidence.

This account of the function of disagreement and argument thus rests on the notion that in virtue of
their behavioural consequences, direct and indirect, some commitments are more appropriate than
others; and moreover, that the appropriateness of a given commitment is more or less constant
throughout a linguistic community. These factors enable individual speakers to benefit from a social
activity, made possible by language: the drive to seek agreement. This drive is fuelled by the
conventional notion of an external standard for belief and judgement, in virtue of which
disagreements are socially unstable. The concept of truth is the catalyst that drives our systems of
commitment through reactive argument, in search of the equilibrium that comes with agreement.

So why does disagreement matter? The account gives two answers. Internally, from the participant’s
perspective, disagreement matters because it is the cue for a mode of criticism – we criticize
judgements with which we disagree and are criticized ourselves in turn. Externally, from the
explanatory perspective, disagreement matters because in the long run it is an indication of
behavioural liability. How are these perspectives related? The account suggests that the external
explains the existence of the internal – or more specifically, of the special notion of correctness on
which the internal perspective rests.

Note that agreement is not an end in itself, on this view, on either the internal or external levels.
Externally, agreement need have no value of its own, but only the contingent virtue that stems from
being the normal result of an otherwise useful process. And internally, it is simply the state in which
disputants accept that they have achieved their common goal – that is, as they see it, the truth of the
matter.

87
Indeed, it is crucial that disputants should not treat agreement as an end in itself. In the absence of any
felt allegiance to an external goal, agreement is easily engineered – for example, as in many animal
societies, by a system of deference to authority. I don’t mean to suggest that argument is free of
deference to authority, and such like. Clearly it isn’t, in practice. Nor do I deny that such ways of
reaching agreement can be valuable, even in human societies. The point is rather that there is a quite
different benefit, that stems (in practice imperfectly) from the conception of an external goal for
assertion and judgement. This benefit comes not from the social benefits of agreement itself, but from
the improvements in individual behavioural performance that stem from the reactive process of which
agreement is the stable state.

Imagine, for example, a community of creatures who use sounds as signals for a variety of purposes.
Certain sounds might be uttered in response to significant features of their environment – food,
dangers of various kinds, and so on; others, perhaps, in advance of significant kinds of behaviour, not
otherwise predictable by other members of the community. In both cases it is easy to see how
agreement – uttering the same sounds at any given time – might be valuable in itself. To the extent
that the community ‘s success depends on co-ordinated behaviour, it will be important that the
members of the community are disposed to the same behaviour at the same time.

This aside, however, there might be a quite separate point to noting and resolving disagreements. It
might enable individuals to improve their own behavioural commitments, even when these
commitments have no role to play in the co-ordinated behaviour of the community as a whole. Such
resolutions could be achieved in a number of ways. One possibility, obviously appropriate in some
cases, would be to defer to experts. On visual matters, bats would do well to defer to hawks; on
auditory ones, the reverse. (This doesn’t mean that hawks and bats need co-ordinate any resulting
behaviour, of course.) On some matters, however, the community might possess no experts. It would
then seem advantageous for everybody to be encouraged to have their say; in other words, for
disagreement to be resolved by argument among equals. I am suggesting that it is in this context that
the normative notions of truth and falsity are important. They are at least part of what it takes to
encourage the members of a community to argue when disagreements are noticed. The method is not
infallible, of course. Once in place, for one thing, truth itself becomes a matter on which
authoritarians may exercise authority. But an imperfect system of norms may be very much more
useful than no system at all.

7.4 TWO COMPARISONS: ELLIS AND HABERMAS

The above point marks one of the main differences between the account of truth I am proposing here
and one of the very few explicitly explanatory approaches to truth in the recent philosophical
literature: the account suggested by Brian Ellis, in a paper called ‘Truth as a mode of evaluation’
(1980). Ellis distinguishes ‘objective and subjective modes of speech’:

To say that ‘p’ is true, or more conventionally, just p, is for me to express my belief that p
objectively. But for me to say that I believe that p is to express my belief that p subjectively
.... In speaking in the objective mode, the belief or attitude we express is disembodied and put
forward neutrally for discussion as though it were a thing that had some independent
existence .... The participants in the discussion see themselves as arguing about this neutral
detached object, and deciding whether to accept or reject it. (1980, pp. 88–9)

Ellis associates truth with the objective mode of speech. He leaves it unclear, I think, why this mode
should employ both the forms ‘p’ and ‘“p” is true.’ That point aside, he suggests that objectification
is a

manifestation of a process, ... necessary for human communication, which makes it possible
for human beings to develop what is roughly a communal system of beliefs and attitudes,
which in turn is necessary for co-ordinated and co-operative action .... We are social beings,

88
 and our survival depends on cooperative activity. How we act normally depends on our
desires, values, and belief systems .... And we may suppose that intersubjective agreement in
our belief and value systems at least facilitates, even if it is not a necessary condition for,
such activity. (1980, pp. 89–90)

Thus it seems that Ellis sees the advantage of the objective mode almost entirely in terms of
benefits of agreement per se. It is true that for social creatures agreement or commonality of purpose
is often useful, indeed indispensable. As I have said, however, I don’t think that this can explain why
agreement should be sought in argument, rather than say imposed by authority. In so far as what
matters is simply that we all do the same thing, it would be more economical for evolution to arrange
that we all defer to a leader. The fact that we take the trouble to argue suggests a further benefit, not
guaranteed by easier paths to agreement. I have suggested that this benefit lies in the tendency of
argument to improve a participant’s individual adaption to his or her environment. I think there is
something of this idea at only one point in Ellis’s paper:

Basically, we are people with belief systems who are concerned to interact with each other,
and so to widen the bases of experience upon which our belief systems depend, and to reach
intersubjective agreement about things. (1980, p. 98, my italics)

In virtue of its emphasis on agreement, Ellis’s approach to truth seems vulnerable to the usual
objection to coherence theories. It is often felt that such theories fail to make sense of the intuition
that truth has something to do with answerability to some external standard – the intuition, to put it
crudely, that we can agree until we are blue in the face, yet still be wrong. Correspondence theories
try to make sense of this world-imposed ‘rightness’, as the basis of an analysis of truth.

Like Ellis’s, the present approach rejects this attempt at analysis, in favour of an explanation of our
possession of the notion of truth. Unlike Ellis’s, however, my account retains the intuition that world-
imposed rightness has something to do with truth. On my account it is crucial to the explanation of
truth that some commitments are (or tend to be) more appropriate than others. This ‘appropriateness’
is world imposed. Behaviour that is successful in one world would be disastrous in another. This
notion of appropriateness is not an analytic base for truth, however, but part of the explanatory base.
In the terms I distinguished above, appropriateness belongs to the external perspective, whereas truth
belongs to the internal perspective.

I have emphasized the argumentative role of truth and its contrast to authority as a means of achieving
agreement. These two points may call to mind another similarity the present account bears to
contemporary work on truth. Both are prominent in the theory of truth developed in recent years by
Jürgen Habermas. Here I can do little more than to note this connection. Two cautions seem to be in
order, however. For one thing, it is unclear to what extent Habermas’s theory is intended or can be
regarded as an explanatory account of truth. Certainly it has been interpreted as a variant of a Peircean
consensus analysis of truth (see Pettit 1982, for example). Secondly, it is far from clear that Habermas
would countenance the naturalistic and evolutionary flavour of the account here proposed; Hesse
(1978, p. 390) suggests that ‘any such interpretation of human evolution seems to fall foul of
Habermas’ own strictures against Chomsky’s naturalism [cf. Habermas 1970] – it is itself a social
theory and hence presupposes its own normative standpoint.’ (I return to this kind of objection in
chapter 9.)

7.5 ARGUMENT AND THE SOURCE OF BELIEF

We have been discussing the value of disagreement and argument, and hence trying to account for the
nature of the ordinary notion of truth. So far, we have characterized the benefits of argument in terms
of its effect on the very commitments whose linguistic expressions are the overt matters of dispute. It
is time to recognize another benefit. An inappropriate commitment may be not only a behavioural

89
liability in itself, but also a manifestation of a potentially more serious liability – a sign of a flaw in
the system of background beliefs, rules or habits of inference, from which the commitment in
question arose. It seems that disputes have the function not merely of correcting individual mistakes,
but also of encouraging improvements in the general beliefs and dispositions that lead to those
mistakes.

It will be helpful at this point to appeal to a simple model of the origins of belief. Let us say that a
person typically forms a belief in virtue of two things: a prior evidential belief, and a general
inferential disposition to adopt a belief of one kind, given a belief of another kind. Since our use of
this model will be largely heuristic, accuracy need not be one of its virtues. To our folk-psychological
intuitions, however, many cases of belief formation do seem to be of much this kind.

The model allows disagreements to be traced to two possible sources. If we disagree about the truth of
some proposition P, we may have had different evidential beliefs, or we may have employed different
rules of inference. Our disagreement may turn out to be evidential or inferential (or a combination of
both, of course). And our dispute, if we engage in one, has the potential not only to rid one of us of a
behaviourally disadvantageous attitude to P, but also to correct the source of that mistake.

I think it would be an interesting and profitable task to examine the structure of such arguments in
more detail. The model seems to throw light on the connection between truth, inference and validity.
It seems that in the process of argument, speakers attempt to trace the claimed ‘incorrectness’ of their
opponent’s assertions to its roots in any proffered defence. At the same time they try to import
‘correctness’ from the claimed roots of their own position. Central to this is the notion of an inference
that claims to preserve ‘correctness’, and can itself be challenged on the grounds that it fails to do so.
However, our immediate concern is with another application of the model. As yet there is nothing in
the notions of disagreement and argument to indicate that there is a place for non-factualism. This
model will help us to explain the limits of truth, and hence the origins of non-factualism.

In section 7.3 I emphasized the importance of the SBP (‘Same Boat Property’), only in virtue of
which can it be true that if two people hold conflicting commitments, then one of them is likely to be
behaviourally disadvantaged. Without the SBP there would be no reason to encourage agreement, and
hence none to treat disagreement in evaluative terms – as an indication that someone has made a
mistake.

Given that we do think in these terms, argument seems to be an attempt to trace the origins of the
disagreement to some evidential or inferential basis. In effect, argument consists in a search for what
both parties will agree to be the original mistake. In general, in other words, we argue as we should if
the justification for evaluative treatment extends to the precursors of the belief at the centre of an
initial disagreement; as we should if the SBP applies as much to evidential beliefs and rules of
inference as to the judgements that lead us into conflict.

However, it turns out that this extension of the SBP is not always in order. There are cases in which it
seems proper that speakers be considered to be initially in different boats – and therefore initially
justified in disagreeing – even though not justified in preserving that disagreement, when each
becomes aware of the other’s point of view. We shall see that such cases populate the margins of
factual discourse. It is in these ill-defined fields of usage that non-factualism finds its place.

7.6 PROBABILITY AND THE END OF ARGUMENT

The most striking cases, in my view, are those involving probabilistic judgements. Such judgements
reflect evidence that is less than complete, in the following sense. If I am interested in whether P, then
I make a judgement such as ‘It is probable that P’, or ‘There is an n% chance that P’, only because my

90
evidence does not permit me actually to determine whether P. I do the best I can with the limited
evidence available to me.

However, my evidence may differ from yours. It need not be that either of us is mistaken about our
evidence, but simply that we have access to different facts. So my best judgement about the
probability that P may differ from yours. But when I declare that it is probable that P, you may well
reply that on the contrary, it is rather unlikely that P. There appears to be a disagreement, the basis for
an argument. When the facts emerge, however, we needn’t insist that in the first place one of us must
have been mistaken. On the contrary, I may well recognize that in your place I would have made your
assessment of the probability that P, and vice versa.

It is important to appreciate that the SBP does still apply in a limited form. Once we become aware of
the difference between our judgement s of the probability that P, then we are in the same boat. On the
one hand, the behavioural consequences of probabilistic judgements are the same for both of us
(subject to the usual qualifications). And on the other, since in principle we now have access to each
other’s evidence, there is no longer a systematic difference between us, such as to justify different
judgements about the same probability. So there is a reason for us to seek agreement, and hence for
convention to apply the normative incentives that encourage us to do so.

The fact that the SBP does not generalize to the initial disagreement suggests that in a properly
organized linguistic community, speakers would be prepared to withhold criticism of each other’s
initial utterances. Indeed, they may wish to commend, rather than criticize, if it becomes apparent that
the initial difference of opinion rests on their each having access to a different body of evidence. So if
‘true’ and ‘false’ are evaluative terms whose function is to encourage useful revision of judgements
and patterns of inference, then we should expect that speakers who discover that a disagreement is of
this kind would be reluctant to say that the other party’s initial claim was false. And this does seem to
be what happens. When we discover that our disagreement has this sort of basis, we are content to say
that we were both correct, given the evidence we then had; and we go on to consider the bearing of
the combined evidence.

Some terminology: I shall say that in such a case the initial disagreement evaporates or becomes
insubstantial. A disagreement is substantial so long as the parties concerned regard each other as
mistaken, and evaporates when such evaluations are no longer felt to be appropriate. We shall see that
this can happen in several ways.

In the probabilistic case, disputes seem to run as follows. Any disagreement is initially treated as
substantial. If it then transpires that the disagreement rests simply on the fact that the disputees
originally had access to different bodies of evidence – in other words, that each recognizes that the
other has made best use of his or her own evidence – then the original disagreement evaporates. Once
both parties have access to their combined evidence, however, any remaining disagreement is treated
as substantial.

What determines whether a disagreement about a particular sort of judgement should always be
treated substantially? The following answer might suggest itself: a disagreement is or at least should
be treated as substantial if it is a dispute about a matter of fact – if it answers to an objective reality, in
virtue of which one party must be mistaken, as long as they disagree. As will by now be apparent, I
think that this gets things back to front. I suggest that judgements of a certain kind are factual just to
the extent that we do treat all disputes involving such judgements as substantial. Evaporative
disagreement marks the limits of what we might characterize as the factual pattern of usage. In other
words, the crucial feature of the factual pattern is that it requires us to say that of any pair of
conflicting judgements, at least one must be false.

In chapter 6 I distinguished the descriptive and explanatory components of a non-analytic approach to

the notion of statement of fact. This characterization belongs on the descriptive side. The job of the

91
descriptive part of such a theory is to characterize the manifestations of the distinctively fact-stating
form of language; and to do so in such terms as to enable us to explain its existence. I think that by
focusing on dispute behaviour – especially on the extent of substantial disagreement – we can do
much to explain the relevant distinctions we find in usage in terms of the value and limits of
argument. At the same time, I think we can do considerable justice to non-factualism, at least in its
more important applications. For we can show that the intuitions that seem to support non-factualism
are very much those that incline us to leave room for evaporative disagreements. Non-factualism
seems largely a product of these intuitions, viewed through the lens of philosophical analysis. The
present approach rejects the analytic perspective, but it enables us to explain the intuitions that that
perspective distorts.

In the remainder of this chapter I want to defend the suggestion that factual discourse be characterized
in these terms against a variety of objections. In particular, it needs to be shown that the distinction
between substantial and insubstantial disagreements puts the limits of factuality in roughly the right
place. By this I don’t mean that we can tell in advance where the right place is; but simply that it
would be a reductio of the proposal to characterize statements in this way, to show that evaporative
disputes commonly occur about utterances whose factual nature is not in doubt. So to begin with we
need to show that what gives rise to evaporative disputes in the probabilistic case is not a feature of
assertoric utterance in general. We shall also want to show – anticipating a likely objection – that
evaporation cannot adequately be explained by the supposition that probabilistic judgements are
indexical.

7.7 COMMITMENT AND THE CASE FOR TOLERANCE

We have observed that probabilistic disagreements are liable to evaporate when they turn out to stem
from the fact that speakers have relied on different evidence. The same might seem true of almost any
kind of judgement. It often happens that although we disagree with another speaker, we recognize that
in her position, with her evidence, we would have come to the same conclusion ourselves.
Accordingly, we are inclined to excuse her mistake. Doesn’t this show that almost any assertoric
utterance can give rise to an evaporative dispute?

I think that these cases are actually quite different from those that can stem from probabilistic
judgements. The difference shows up in the use of the labels ‘true’ and ‘false’ in the two sorts of case.
In the probability case, we observed that we are reluctant to say that a probabilistic judgement is false
when it turns out to have been properly based on different evidence from our own. On the contrary,
we are likely to say that it was ‘true, given the speaker’s evidence’, or something of the sort. But now
consider a case of the other kind. In fading light, a guest mistakes my rabbit for a rat. ‘There’s a huge
rat in your garden!’, he cries. I reassure him, and politely excuse his mistake: ‘That rabbit does look
ratty at this time of night.’ Excusing the mistake is not the same as acknowledging that no mistake has
been made, however. If it came to the point, then, clearly, I would continue to maintain that the
guest’s assertion was false. From his point of view, it was understandable, perhaps, but certainly not
true. There is no room for points of view on the question as to whether a creature is a rabbit or a rat.

The special character of probabilistic judgements shows up particularly in the following fact. It can be
reckoned correct to make a probabilistic judgement even if one knows of the existence of evidence
which, if one had it to hand, might make one revise that judgement. For example, a doctor might say,
‘You are probably not infectious, but if we had your test results we would have a better idea’ ; or
even, ‘You are probably not infectious, but the test results will tell us one way or the other.’

Contrast these examples with the following: ‘You are not infectious, but if we had the test results, we
would have a better idea’; or ‘You are not infectious, but the test results will tell us one way or the
other.’ (Or: ‘That’s a rat, but in better light it might turn out to be a rabbit.’) Such assertions take
away with one hand what they have given us with the other. Utterances such as ‘You are not
infectious’, or ‘There’s a rat in the garden’, seem to be expressions of ‘final’ commitment, in the

92
sense that to accept such an assertion is to accept that further evidence is unnecessary. If there is a rat
in the garden, then any evidence that seems to suggest the contrary is misleading. As long as we are
committed to claim that there is a rat, we are committed to the view that the case is closed – additional
evidence is mistaken or irrelevant, if not confirmatory. The above examples violate this ‘closure
principle’, explicitly rejecting the commitment to ignore further evidence.

Gilbert Harman has discussed this feature of commitment. In a recent book on reasoning he observes
‘that people find it difficult to accept things only tentatively .... There is a strong tendency quickly to
convert such tentative acceptance into full acceptance’ (1986, p. 49). Harman characterizes ‘full
acceptance’ as follows:

Belief in or full acceptance of P involves two things. First, one allows oneself to use P as part
of one’s starting point in further theoretical and practical thinking. Second, one takes the issue
to be closed in the sense that, when one fully accepts P, one is no longer investigating
whether P is true. ... Full acceptance of P may therefore involve an implicit commitment to
the claim that further investigation of P would not be worthwhile, for example, in leading one
to discover further evidence that would lead to a reassessment of one ‘s conclusion. (1986, pp.
47–8)

It is a nice question why we make such final commitments. Clearly, they are often unjustified. The
very fact that we can appreciate this, that we can change our mind, shows that in a wider sense our
actual commitments are not final. On the whole, we do not close off the possibility of conflicting
evidence. Why then don’t we express ourselves in a form that allows for this possibility? Why doesn’t
my startled guest cry, ‘There is probably a large rat in the garden!’? Or more interestingly, why isn’t
‘There is a large rat in the garden’ simply understood in this way, so as to allow for insubstantial
disagreements? (The interpretation would not have to be probabilistic. ‘It seems that P’ has much the
same effect.)

A possible answer is that in the development of language full acceptance and assertion simply came
first. Perhaps we could do without it, at least in theory. But in practice it developed before the various
forms of qualified assertion, and so we have both. Harman suggests a less contingent constraint: that
in practice there may be no alternative to full acceptance, for creatures of our limited mental abilities.

If one had unlimited powers of record keeping and an unlimited ability to survey ever more
complex structures of argument, replies, rebuttals, and so on, it would be rational always to
accept things only tentatively as working hypotheses, never ending inquiry. But since one
does not have such unlimited powers ..., one is forced to limit the amount of inquiry in which
one is engaged and one must fully accept most of the conclusions one accepts, thereby ending
enquiry. Tentative acceptance ... cannot be the general rule. (1986, p. 50)

These factors suggest that most commitment is full commitment, and hence that most expressions of
commitment are normally expressions of full commitment. But they do not seem to entail that
ordinary expressions of commitment must always be taken as expressions of full commitment. That
is, they seem compatible with the possibility that all or most assertions can in exceptional
circumstances be the basis of an evaporative disagreement. I do not want to deny this possibility,
which in my terms is the suggestion that the (descriptive) distinction between statements and non-
statements is always a matter of degree. But the issue is somewhat premature, until we have a clearer
idea of the range of possible ways in which utterances can give rise to evaporative disagreements. For
the moment I want simply to make the point that probabilistic judgements are subject to evaporative
disagreements, for a reason not characteristic of assertoric judgements in general.

This reason can now be described as follows. Probabilistic judgement is a species of intrinsically
‘non-final’ judgement. Probability has its origins in situations in which we feel that final commitment
is unwarranted. In this context not only is judgement sensitive to evidence, but there is no warrant for

93
ruling out further evidence. Suppose for example that we are interested in whether P, and that the
context is such that we make judgements about the probability that P. These judgements are based on
our evidence about P – and we would not make such probabilistic judgments if we thought that the
evidence was sufficient to warrant a final judgement that P (or that not-P). In other words, we hold
open the possibility that further evidence would make a difference to our views as to whether P. But
this means that whatever judgement we do make must be non-final – we must hold open the
possibility of revising it, if new evidence comes to hand. So the non-finality of judgement about
probability stems from its origins in the expression of non-final judgements about other topics.

It is illuminating to contrast this explanation to a characteristic argument for non-factualism about

probability, that we encountered in chapter 4. That argument focused on the same feature of
probabilistic judgement – its origins in the expression of qualified judgement about other matters –
and hence observed that probabilistic assertions typically express ‘partial beliefs’, or degrees of
confidence. The non-factualist wanted to infer from this that such assertions do not express full
beliefs, and hence that probabilistic utterances are not statements of fact. However, we saw that a
factualist need not accept this inference; for the non-factualist has not shown that the same mental
state could not be both a partial belief that P and a full belief that it is probable that P. On the contrary,
the factualist might suggest, the notion of a degree of confidence, functionally construed, provides a
characterization of the content of a belief about probability.

The present argument does not depend on this questionable step. It is not intended to show that
probabilistic statements are not factual, but rather to explain the fact that their usage differs in a
significant respect from that of typical non-probabilistic statements. The argument does not take this
difference to prove that probabilistic utterances are not ‘genuinely’ assertoric, descriptive, or what-
ever; for it belongs with a point of view that rejects such underlying distinctions. It simply attempts to
explain, in terms of the most evident function of probability in discourse, a subtle but important aspect
of the application of the notions of truth and falsity to probabilistic judgements.

It may seem surprising that a disagreement about the very same sort of judgement can be sometimes
substantial and sometimes insubstantial. There is a simple explanation, however. In the probability
case, for example, I said that disagreements evaporate in cases in which it turns out that agents are
initially relying on different pieces of evidence. Cases like this are more common in some areas than
in others. In scientific discussions, for example, it can usually be taken for granted that people share
the same evidence, and moreover that new evidence will not come to hand, at least in the context of
discussion. Probabilistic judgements can then be treated as final, and evaporation will be rare.

Even in more ordinary contexts, there seem to be good reasons for treating a probabilistic dispute as
substantial, until it proves not to be. For one thing, it helps to ensure that the important disputes – the
cases in which somebody has made a mistake – don’t slip through the net. It is better to risk
prosecuting a few innocent differences than to risk excusing a guilty one. Secondly, we noted that if
the grounds for an insubstantial disagreement are that speakers have initially relied on different
evidence, then those grounds are effectively cancelled by the communication that calls attention to the
disagreement. Communication puts the SBP into effect – if two people can disagree, then they can
compare and combine their evidence. In order that they have an incentive to do so, it seems important
that any current disagreement be treated as substantial. Only if such a disagreement is treated as if
there is a single correct (or ‘true’) answer, will participants be encouraged to reassess their positions,
to re-examine the latest evidence, and hence to strive for agreement.

Thus in the probability case it seems that only non-current disagreements should be allowed to
evaporate. (Not all non-current disagreements, of course, but only those that turn out to have stemmed
from access to different bodies of evidence.) The basis for such disagreements is automatically
cancelled by communication. It is natural to wonder whether this is a peculiarity of the probability
case; or whether, on the other hand, all current disagreements should be regarded as substantial. To
settle the question, we shall need to examine some non-probabilistic sources of evaporative

94
disagreement. That will be the main task of chapter 8. Before turning to non-probabilistic cases,
however, I want to deal with the objection that a factualist can explain the existence of evaporative
disagreements about probabilities by appealing to indexicality – by allowing that probabilistic
judgements make implicit indexical reference to the evidence on which they are based.

7.8 THE INDEXICAL OBJECTION

An evident feature of indexical judgement is that apparent disagreements may turn out to stem from a
misunderstanding as to the intended reference of an indexical term. For example, the judgements ‘It is
cold here’ and ‘It is warm here’ are incompatible, in the sense that no one could correctly assent to
both at the same time; but obviously both may be correct if said by speakers at different places or
times. In chapter 5 we saw that a probabilistic factualist has other reasons to think of ‘It is probable
that P’ along the lines of ‘Given the present evidence, it is probable that P.’ Does not such a reading,
coupled with the above feature of indexical judgement in general, explain what happens when two
speakers base judgements about the probability of the same state of affairs on different bodies of
evidence?

There is an important difference between the two cases, however. Imagine for example that you
phone me and say ‘It is snowing here today.’ I will disagree with you if and only if I think it is not
snowing where you are; and agree if and only if I think that it is snowing there. But in a probabilistic
case I may disagree with you even if I think your evidence does support your assessment of the
probability at issue; and agree with you even if I think that it does not. In other words, a disagreement
may exist in the probability case, even if both parties acknowledge that neither has incorrectly
assessed his or her own evidence. There is no analogous possibility in the usual indexical case.

The objector will reply that in the probability case this possibility depends on a shift in the reference
of the indexical term, in the course of a conversation. It is simply the possibility that, for example, I
might agree that given your evidence it is probable that P; and yet disagree that given the presently
available evidence it is probable that P. Clearly such changes of reference are likely to occur in
conversation. At least in principle, conversation allows us access to each other’s information on the
topic at hand. We may begin with different bodies of ‘available evidence’, but conversation ensures
that what is now available to each of us is our combined evidence.

But can this account for the facts? The impression that it does so seems to me to trade on the
vagueness of the notion of ‘available evidence’. There are various readings of this phrase, ranging
from the completely subjective – ‘The evidence of which I am actually aware’ – to the fairly objective
– ‘The evidence available in principle.’ At the subjective end of the scale, we certainly explain how
apparent disagreement s about probability can evaporate; but at the cost of making it mysterious why
they should be treated as disagreements in the first place. Take a forensic example: you say that your
evidence points to organized criminals with contacts in high places; I think that mine suggests a clever
twelve-year-old with a personal computer. My initial reaction is not to think that we disagree but to
assume that we must be relying on different evidence. Moreover, since there is no evident
disagreement between us there is no apparent motive for us to take matters any further. What needs to
be explained is the fact that in a case like this we should and do go on to consider our combined
evidence. On my account the motive is provided by our treatment of such a case as a substantial
disagreement, answerable to a normative notion of correctness. (This is the internal motive, of course;
the external justification comes separately. More on the latter below.) But on this subjective version
of the indexical interpretation, the sanitized exchange provides no motive for anything more than
polite acknowledgement of different but thoroughly compatible judgements.

So much for a subjective interpretation of the supposed reference to ‘available evidence’. What if this
phrase is read as something less unambiguously subjective – for example, as something like ‘the
evidence reasonably accessible’? The problem is now a dilemma. On the one hand, if this reading is
taken sufficiently strongly to exclude the possibility that two parties to a conversation have different

95
‘available evidence’, then it excludes the very cases the indexical treatment is supposed to explain –
those that seem to allow evaporative disagreements about probability. If on the other hand the reading
allows for differences between speakers in their ‘available evidence’, then the above problem recurs.
The supposed disagreement doesn’t get started, because speakers who understand probabilistic
assertions in these terms will be inclined simply to attribute what would otherwise be a conflict to
differences in the evidence that each refers to.

Thus it seems to me that an indexical reading cannot explain this distinctive feature of probabilistic
judgement. The impression that it does so rests on a systematic equivocation concerning the supposed
indexical reference – in effect, on a vacillation between the horns of the above dilemma.

7.9 EVAPORATION AND THE PROPERTY OF TRUTH

At this point a determined opponent, though conceding that the indexical interpretation does not
provide an orthodox explanation of evaporation, might hold out the possibility of some other such
explanation – some other way of accounting for evaporative disagreements within an orthodox truth-
conditional interpretation of probabilistic judgement. I want to finish with an argument that seems to
count against this possibility.

The analytic view thinks of truth and falsity as properties; the goal of analysis is an understanding of
the nature of such properties. Suppose that P is a sentence (or statement, proposition, or whatever) of
the kind that bears these properties. ‘P is true’ is then a respectable assertoric sentence in its own right
– it can be used to say that a certain linguistic entity has a certain property. Moreover, since it is not
probabilistic it cannot be the subject of the kind of evaporative dispute described above.

Imagine now that P itself is probabilistic. The analytic view faces a dilemma. On the one hand we
have just observed that if two speakers disagree about a simple predication such as ‘P is true’ then one
of them is mistaken, whatever their evidence – one of them has spoken falsely. Yet on the other hand
to disagree about ‘P is true’ is to disagree about P. If the former disagreement entails that one party is
mistaken, then so does the latter. If we allow evaporative disagreements about P, we must allow them
about ‘P is true.’

If truth and falsity are properties, in other words, their literal ascription must be confined to sentences
not subject to the ‘no fault’ evaporative disagreements characteristic of probabilistic judgements. It
follows that our ordinary treatment of probabilistic judgements is not entirely compatible with the
truth-conditional model. From the orthodox point of view, either we misuse such judgements, or they
do not possess genuine truth conditions; for their use is incompatible with their having the property
for possession of which truth conditions are conditions.

This point has been obscured, I think, by the assumption that the features of use on which it rests can
be explained by indexicality – by some sensitivity of probabilistic truth to evidential context.
However, we have seen that the appeal to indexicality itself rests on an equivocation in the relevant
notion of ‘available evidence’. In any case, the result is an embarrassment for the analytic approach.
In one way or another, it detaches the analyst’s desired ascription of truth conditions from its
evidential basis in usage. It implies that we either use truth-bearing sentences as if they were not truth-
bearing, or use non-truth-bearing sentences as if they were truth-bearing.

Note that the present approach encounters no such problem. An explanatory view is not committed to
the idea that truth is a property, seeking merely to explain the fact that ordinary speakers are inclined
to treat it as such. If it turns out that ordinary usage is sometimes incompatible with a treatment of
truth as a property, then so much to the credit of the explanatory view – at least so long as it can
account for this fact.

96
I have argued that in the probabilistic case the appropriate account turns on the role of probabilistic
judgement in contexts of incomplete evidence. In chapter 8 I want to show that by appealing to
different factors in the same general spirit, it is possible to explain the peculiarities of certain other
areas of discourse – areas which, like probabilistic usage, have been popular targets for the non-
factualist approach.

97
8
The Limits of Intolerance
We are trying to account for the ordinary use of the notions of truth and falsity in terms of their
function in encouraging argument. In view of our original concern with non-factualism it is important
that the account should work ‘at the edges’ – in the borderline cases in which talk of truth and falsity
seems to some degree inappropriate, and in which non-factualism has hence seemed particularly
attractive. We looked at one such case in chapter 7. I argued that our occasional reluctance to apply
the notions of truth and falsity to probabilistic judgements can be correlated with a systematic failure
of the normal functional justification for the use of truth and falsity.

In this chapter I want to examine some further cases, and show that again the functional theory of
truth can explain the kinds of peculiarities that have been taken to favour non-factualist treatments of
the topics concerned. I’ll pay most attention to moral and conditional judgements (the topics whose
non-factualist treatment we considered in part I). But I shall also comment on the account’s potential
in other areas in which non-factualism has been appealing. And I’ll finish, as promised, with some
remarks on the non-indicatives.

Perhaps I should say again that in attempting to explain the appeal of non-factualism, I won’t be
trying to make a case for non-factualism as normally construed. Non-factualism normally takes for
granted the analytic approach. It sees itself as concerned to determine the true extent of the property
that distinguishes fact-stating discourse. Part I discredited that endeavour. As we’ll see, however,
there is considerable scope for reconstruction of non-factualist arguments along explanatory lines. A
subsidiary concern of this chapter will be to distinguish several stances that remain open in this
connection, when the analytic project has been closed down.

8.1 RELATIVISM AND EVAPORATIVE ARGUMENT

In chapter 7 we drew a rough distinction between two sorts of linguistic dispute, according to the root
of an initial disagreement: in an evidential dispute the root lies in speakers’ evidential beliefs; in an
inferential dispute it lies in the rules or habits of inference that lead from evidential beliefs to the
conflicting judgements that are the immediate grounds for disagreement. We noted, however, that
disagreements can have multiple roots; and the distinction between evidential beliefs and rules of
inference is in any case tendentious.

At that stage we had not raised the possibility of evaporative disagreements. However, the one sort of
evaporative disagreement we have so far considered is clearly evidential rather than inferential. It
turns on the fact that in making probabilistic judgements, speakers may have access to different
bodies of evidence. So long as the speakers concerned do not disagree with each other’s evidential
beliefs or probabilistic inference rules, we have an evaporative evidential disagreement.

Can there be evaporative inferential disagreements? I think that there can, and indeed that some of the
best-known tenets of the non-factualist movement amount to the endorsement of this possibility. For
consider moral or aesthetic relativism – the view that different individuals or groups may hold to
incompatible but equally ‘valid’ aesthetic or moral principles. Such a possibility would reveal itself in
a case in which the two groups or individuals make conflicting ethical or aesthetic judgements, on the
basis of agreed non-moral or non-aesthetic facts. Since the non-evaluative evidence is agreed, the
disagreement must stem from the general evaluative principles, in virtue of which each group or
individual infers moral or aesthetic conclusions from such a basis. In other words, in our present

98
terminology, the disagreement has an inferential basis. The relativist claims that in such a case,
neither side need have grounds for regarding the other as mistaken. But in our terms, such no fault
disagreement is the mark of insubstantiality; and hence the relativist seems to be urging the possibility
of evaporative inferential disagreements.

This case bears out some points I made earlier. It illustrates both the element of reconstruction of non-
factualism that follows the rejection of the analytic viewpoint, and the fact that the reconstruction
leaves open several distinct kinds of non-factualist conclusion. In appealing to the possibility of
relativism, non-factualists in ethics and aesthetics have on the whole been concerned to argue that
moral and aesthetic truth is relative – that it is relative, as a matter of fact, and therefore should be
treated as relative. It follows from this view that if a community treats moral truths as absolute, its
members are mistaken about the nature of moral judgement. In other words, the non-factualist
traditionally begins with claims about the ‘real’ status of moral judgements, and proceeds to
prescriptions concerning the use of moral utterances. In rejecting the analytic point of view, we seem
bound to reject this form of non-factualism. We have given up the idea that there is an independent
standpoint, with respect to which there is anything more to being a statement of fact than being treated
as such. We may now describe usage and hence explain it, but there is no longer a place for trying to
analyse usage and hence prescribe it.

I think that a good deal of contemporary non-factualism can be reconstrued in this descriptive sense,
admittedly at the cost of systematic disrespect for the intentions of its authors. The trick is simply to
reinterpret what the analytic non-factualist normally appeals to by way of evidence – such things as
our ordinary intuitions about the truth and falsity of the kinds of judgement in question. From the non-
analytic point of view this appears not as evidence for the presence or absence of some underlying
property, but rather as the basis of a descriptive taxonomy of use. Much of orthodox non-factualism is
thus directly accessible to our first kind of reconstructed non-factualist, whose role is to catalogue the
subtle variations of usage on the boundaries of the statement form.

Perhaps this kind of non-factualism need not be merely descriptive. It cannot prescribe ordinary usage
as such, but it might seek to correct any corruption of usage that could be attributed to the influence of
a mistaken philosophical theory. All the same, to a traditional non-factualist this looks like a dull
substitute for prescription grounded on analysis. Is there room for anything more colourful – for a
non-analytic non-factualism that would seek to revise ordinary usage itself? I think there are two
possibilities, stemming from two kinds of reflection on the use of the class of judgements in question.

8.2 SECOND ORDER EXPLANATORY NON-FACTUALISM

The first possibility would involve the recognition that actual practices of disagreement simply fail to
resolve certain sorts of dispute. In the moral case, for example, it would flow from the perception that
moral disagreements may simply reach a stalemate, in which neither party can do more than to
continue to affirm some disputed moral principle. This perception needn’t itself recommend
relativism. But the inherently unstable nature of such a stalemate, in which each party retains the
conviction that the other is at fault, might seem to recommend a change in usage – a degree of
relativistic tolerance.

The second possibility is perhaps only a more theoretical version of the first. It would concentrate not
so much on the actual procedures we employ in argument as on the role of a type of discourse in our
lives, and that of argument in general. From such a standpoint it might be argued that particular types
of utterance would be better treated in one way rather than the other – ‘better’ being cashed in terms
of the overall value to the individual, community or species of such forms of behaviour. Given the
present functional account of truth, for example, it might be argued that we would do better to allow
for evaporative moral and aesthetic disagreements.

99
Thus there seems to be room for a non-descriptive non-factualism – a second order theory whose
prescriptions concerning usage would flow from a view of the significance of factual usage in general,
and an understanding of the role of the particular area of discourse whose status is in doubt. I shall not
directly attempt to decide to what extent existing arguments for or against relativism about morality
and other topics can be interpreted in this light. But I do want to illustrate the range of considerations
that might be relevant to such a second order non-factualism; and some of these, as we shall see, call
to mind standard moves in the discussions of relativism about morality and other matters.

Let us begin at one extreme, with a challenge to the conceptual possibility of the second order stance.
In the moral case, for example, the stance seems to depend on the claim that it is possible to discuss
the merits of moral principles – of different rules of inference from non-moral beliefs to mora l
conclusions. This seems to depend in turn on an understanding of the meaning of moral judgements
which would be prior to the adoption of a particular set of acceptance rules for such judgements. After
all, these acceptance rules are what the discussion is about. Now the objection: such an understanding
is impossible, because what we mean by moral terms is a function, inter alia, of what we take to be
their assertion conditions. In a slogan, meaning is use, and therefore cannot prescribe use.

In effect, this is the objection that relativism collapses into the indisputable but uninteresting
possibility that different speakers may simply use the same word with different meanings. The proper
reply is obviously that some uses are more important than others. In the moral and aesthetic cases, at
any rate, the feature of use that does most to fix meaning seems to be not the adoption rules in virtue
of which speakers acquire commitments, but the ‘application rules’ – the conventions in virtue of
which commitments have behavioural consequences. It is relatively easy to imagine someone who
adopts moral and aesthetic commitments in different circumstances from ourselves. But as we have
already had cause to observe, it is difficult to make sense of someone who means what we mean by
‘good’, ‘beautiful’, and so on, and yet who is not motivated much as we ourselves are by
commitments involving these terms. There are certain exceptional cases, of course – for example
cases in which the effect of such a judgement is overridden by some other motivation – but we would
be inclined to regard the fact that someone lacked that pull in general as evidence that he meant
something different from us by the terms concerned.

Conversely, it seems that these definite behavioural consequences prevent arguments over matters of
moral and aesthetic principle from degenerating into terminological border disputes. If a person’s
moral and aesthetic commitments – those of her commitments that are expressed in what are for us
moral and aesthetic terms – have for her the same kind of behavioural consequences as such
commitments would have for us, we take this as evidence that she means the same by these terms as
we do. We are then in a position to compare her moral and aesthetic sensibilities to our own; and free
to conclude, with the relativist, that although such a person’s sensibilities differ from our own, she
need not be mistaken.

To show that relativism is a conceptual possibility is not to show that we are or should be relativists,
however. The probabilistic case illustrates this point. As we observed in chapter 4, simple
probabilistic judgements also have fairly definite consequences for behaviour. Roughly, judgements
of increasing probability are reflected in an increasing willingness to risk a given loss for a given gain,
in an action that depends for success on the state of affairs whose probability is in question. If
someone consistently failed to act in this way, having sincerely professed to what appeared to be
probabilistic commitments, we would be inclined to say that he did not mean what we mean by
‘probable’ and associated terms. Conversely, if someone does apply what appear to be probabilistic
judgements in this way, then we are inclined to assume that she does mean what we mean. And we
continue to assume it, even if we disagree with the statistical principles on the basis of which she
makes her judgements. Probabilistic relativism is thus a conceptual possibility. However, it seems to
have little of the attraction of its moral and aesthetic counterparts. Once evidential differences have
been sorted out, we don’t say that chance lies in the eye of the beholder.

100
If we were to try to justify this difference between the probabilistic and ethical cases, I think it would
have to be in terms of the claim that barring evidential differences, probabilistic judgements involve
no systematic failure of the SBP. Presumably our ability to make probabilistic judgements has some
survival value, and presumably its value would vary if we varied our probabilistic inference rules. (It
is easy to think of cases in which the wrong rules would regularly lead to disaster.) Access to
evidence aside, there seems to be no difference between speakers that could explain how the optimal
probabilistic inference rules could be different in different cases. And in the absence of such a
difference, there may seem to be no case for probabilistic relativism.

However, this is to overlook the possibility that there might simply be no optimal set of probabilistic
inference rules. Any such set of rules is a strategy for making judgements on the basis of less than
conclusive information. In principle there might be a range of such strategies, none clearly better than
any other. Probabilistic relativism then begins to look appealing, I think, on inferential as well as
evidential grounds. This suggests that the apparent lack of evaporative inferential disagreements in
ordinary probabilistic usage may simply reflect the fact that ordinary probabilistic reasoning is not
complex enough to have encountered the difficult cases. (If so then we might do better to look at the
reasoning of professional statisticians. I am not sure to what extent differences between statistical
schools of thought can be viewed in these terms.)

8.3 RELATIVISTIC PRACTICE: FOR AND AGAINST

The point remains that in the moral and aesthetic cases, as for probability, relativism is neither
entailed nor recommended simply by the fact that it is a conceptual possibility. One further factor is
whether moral and aesthetic judgements have the SBP. The relativist’s traditional appeal to
subjectivity can be seen as an argument that the property does not hold in these cases (and hence, in
our terms, that moral and aesthetic disagreements are properly treated as potentially evaporative). The
subjectivist claims that moral and aesthetic utterances are expressions of desires, motivations, or
something of the kind. Ideally such attitudes should be correlated with dispositions to feel pleasure or
displeasure in certain states of affairs. (Masochism is not a particularly good survival strategy.) Hence
if different speakers can have different such dispositions, the SBP need not apply: if pleasure lies in
the eye of the beholder, then so (we should allow) does beauty.

This sort of argument can be challenged at a number of points, however. In particular, it seems to
depend on the view that appetites or motivations are simply givens, and not themselves contingent
products of cognitive processes. If motivations are allowed to be changeable, then there are at least
two lines of argument in favour of non-relativistic moral and aesthetic practices.

The first argument would rest on the claim that some practices are in some sense ‘better’ than others,
and hence that there is a place for the kind of improvement that is encouraged by the notion that
expressions of motivation are answerable to an external standard. ‘Better’ might be cashed in several
ways. Perhaps most plausibly, it might be argued that some tastes and appetites are potentially more
rewarding than others. Hence there would seem to be an individual advantage in a social process of
moral and aesthetic discussion, argument and refinement. The more educated one’s palate, the greater
one’s enjoyment, on this view – and hence we all gain, on the whole, by the process of comparison,
criticism and improvement that is encouraged by the practice of treating moral and aesthetic
judgements as true and false.

This argument seems more plausible for the aesthetic than the moral case. For the moral case,
however, the recognition that motivations are changeable states suggests another line of argument.
For it suggests that we regard moral judgements not as expressions but as sources of motivational
states. Roughly, we tend towards what we take to be good not because ‘taking to be good’ is a
manifestation of a pre-existing tendency but because ‘good’ functions as a conventional reward for
action. Clearly, this is a view of moral discourse that parallels the present account of truth and falsity.
It is a likely development of the theme, familiar in ethical theory, that moral concepts play a

101
regulative role in society. The view would thus be that morality provides a system of incentives that
functions to encourage socially ‘desirable’ forms of behaviour (‘desirable’ being cashed in non-moral
terms, of course).

However, such a system of incentives would be useless if everyone was held to be entitled to a
reward, in virtue of satisfying his or her own individual standards. Relativism thus threatens the social
function of morality – which perhaps explains why it is such a disturbing thesis, at least in extreme
forms. The practice requires that the system of rewards be seen as objective, as answerable to an
external standard; or in other words, that speakers regard differences in their application of moral
descriptions as indications of mistakes.

Thus we have two possible arguments for disallowing evaporative inferential disagreements on moral
and aesthetic matters. The first is based on the claimed advantages of educated aesthetic and (less
plausibly) moral tastes. The second turns on the regulative role of moral and (less plausibly) aesthetic
judgement. Both, I think, leave scope for limited relativism. The advantages of conformity may
sometimes be outweighed by the disadvantages of intolerance. ‘Live and let live’ may sometimes be
the optimal survival strategy. Perhaps we should behave as if everyone in our own community is in
the same boat, but allow that foreigners do things differently.

I do not propose to examine the above justifications for moral and aesthetic objectivity in any detail.
Our present purpose is simply to illustrate the variety of factors that, from the non-analytic
perspective, may be held to explain or recommend a particular pattern of usage. However, the
arguments do call for a number of comments. The first of these concerns the parallel between the
latter approach to morality and our suggested account of truth.

8.4 TRUTH AS INCENTIVE – A CIRCULARITY?

We described ‘true’ and ‘false’ as incentives, whose purpose is to encourage speakers to seek
agreement, and hence to participate in linguistic disputes. The second argument above suggests that in
the moral case, such a system of incentives requires that speakers treat moral truth as objective – in
other words, that they treat moral disagreements as substantial. This seems to imply that in the case of
truth and falsity themselves, the claimed system of incentives would itself require that speakers treat
disagreements about the truth of utterances as substantial. If so, isn’t the account in trouble? Doesn’t
it commit us to an endless hierarchy of disagreements, corresponding to iterations of the predicates
‘… is true’ and ‘… is false’?

This is a nice point, to which I want to return in the next chapter. The brief answer is that although it
is true that if ‘true’ and ‘false’ are to function as incentives then speakers must treat disputes about the
truth of utterances as substantial, this is guaranteed, as long as it needs to be, by the general
presumption that disagreements are substantial unless proved otherwise. Evaporative disagreements
are the exceptions. There is a kind of circularity in the use of ‘true’ and ‘false’ as incentives, but it is
not vicious. In order that ‘true’ and ‘false’ should function as incentives, their ascription takes the
objective or declarative mode. Truth is presented as an objective property, that utterances either have
or lack. We add truth to our idea of the world, to help us get on in it. But without truth, we would not
have an objective or declarative mode of speech. What is distinctive about this mode of speech is that
it is regarded as answerable to an external standard; that is, to truth.

8.5 PROJECTIVISM NOT VINDICATED

The second point arising from section 8.3 concerns an apparent contrast between the above case for a
substantial treatment of moral disagreements and the kind of justification previously suggested – a
difference that seems to echo the old idea that beliefs and desires have different ‘directions of fit’ to
the world. In the previous non-evaluative cases (including the probabilistic case) we used the SBP to
explain and justify the use of the normative notions of truth and falsity. We argued that because the

102
behavioural consequences of such commitments are broadly similar across a linguistic community,
there tends to be an individual advantage in a conventional system of incentives and disincentives that
encourages attempts to reach agreement. The world puts the members of a community in the same
boat, as it were, and hence we tend to do better if we co-operate. In the moral case, however, the
above argument may seem to reverse things – to suggest that we should behave as if an external moral
reality had put us all in the same boat, in order that morality might serve its community purpose.

This is an important objection. If sustained, it would suggest that the present account is a form of the
projectivist theory we discussed in chapter 4. For it would seem to admit a distinction between those
areas of discourse for which truth can be cashed in terms of a world-imposed notion of correctness,
and those in which there is a useful function for a pretence of world-imposed correctness.

However, I think the objection is mistaken. It is correct in noting a difference, on the proposed
account, between the grounds for the application of the notions of truth and falsity in different areas of
discourse (at least if their application to normative discourse is explained by the latter of the two
arguments above). Indeed, I think the explanatory approach should explicitly affirm this possibility.
Once we detach the explanation of the use of truth from the search for an analytic base, we should see
it as possible, even probable, that across the range of functions served by the assertoric form of
language, the advantages of truth-encouraged argument will not be the same in all cases. Certainly,
we should expect that these cases have something in common, to explain the fact that they all employ
the assertoric form and the notion of truth. But this degree of uniformity is compatible with
considerable diversity in its explanatory base. The advantages of argument may vary from case to
case.

The objection is mistaken, however, in so far as it assumes that in the non-evaluative cases (or some
sub-class of these cases), the account provides a ‘genuine’ notion of truth, cashed in terms of ‘world-
imposed correctness’. I think this mistake stems from a tendency to construe the SBP in terms of the
‘success’ account of truth. From such a point of view it seems obvious what property is shared by the
same belief in different heads: that of leading to behavioural success in the same circumstances – that
is, on this view, the property of being true in the same circumstances. However, in introducing the
SBP I emphasized that it should not be conceived in these terms. In order to account for an evaluative
notion whose function seems to be to encourage (reasoned) agreement, we needed to claim that
disagreement is in general an indication of some sort of behavioural disability. That in turn seemed to
require the assumption of the SBP. Without something of the kind, we could well have disagreements
between speakers in cases in which each holds what from his or her own point of view is the
behaviourally optimal belief. I noted that this can indeed happen in individual cases, such as when the
effects of a mistaken belief are balanced by a mistake elsewhere in an agent’s system of belief. But to
account for truth in terms of its propensity to encourage agreement, we need to assume that it does not
happen globally.

The SBP was thus a theoretical hypothesis, needed at this point in our explanatory account of truth.
As such, it derives support from the demonstration that departures from the standard use of truth can
plausibly be correlated with specific reasons to favour (or at least not to penalize) disagreement, in
special cases – with systematic factors in virtue of which the SBP might be expected to fail.

This route to the SBP is non-committal as to the nature and origins of notion of ‘appropriateness’ on
which it depends. As just noted, there is therefore an understandable temptation to think of it in terms
of truth, ‘correct representation of reality’, or something of the kind. This temptation must be resisted,
however, for these are the very concepts whose use we are trying to explain. Behavioural
appropriateness belongs to an explanatory but not an analytic base for the suggested account of truth.
The key to the explanatory approach is the realization that these things can be kept apart.

Provided we do keep them apart, the above objection can be seen to be mistaken. Certainly there may
be a difference between the senses in which the ‘appropriateness’ of evaluative and non-evaluative

103
judgements depends on more than individual preference or point of view. But this does not entail that
in the latter case ‘appropriate’ can be equated with ‘true’. The difference lies within the scope of the
explanatory approach as a whole. It does not, as for the projectivist, mark a point at which explanation
takes over from analysis.

8.6 PRESCRIBING USAGE

The final point arising from section 8.3 concerns the sense in which the arguments for and against
aesthetic and moral relativism can be considered to be ‘prescriptive’. It needs to be emphasized that
their conclusions are ‘external’ rather than ‘internal’ prescriptions for moral and aesthetic usage. That
is, they arise not from within a particular moral or aesthetic practice, but from an external standpoint –
a point of reflection on the practice as a whole.

This distinction is not always easy to draw, but I think it has a place in many areas of discourse. Many
practices yield prescriptive principles – principles that will seem to the practitioner to follow from the
‘ground rules’ of the practice in question. But these are internal prescriptions. They flow from the
practice, and we cannot feel their force unless we already subscribe to the practice from which they
stem.

External prescriptions, on the other hand, depend on an ability to detach ourselves from the practice in
question, and hence to make recommendations about its conduct from some wider standpoint. This
exercise is problematic in a number of ways. Even if we can achieve the required external perspective
– survey usage from outside, as it were – there are difficulties with the notion of a recommendation
from such a perspective. Here, as in other areas of the human and social sciences, the boundary
between justification and explanation is far from precise. Naturalistic explanation tends to be
presented as conservative defence.

Whether conservative or reformist, however, similar questions arise. In what sense is the
recommended practice better than others? And how is it possible for us to adopt such a
recommendation – to decide on a particular pattern of usage? These are difficult issues, and I have no
definite answers to offer. But it does seem that we are not entirely constrained by current convention.
In philosophy and in ordinary life we do have some ability to reflect on and to modify these kinds of
patterns of linguistic practice; even if the process of change seems to involve not so much an explicit
decision on current and possible usage as a communal shuffle in a direction we cannot quite describe.

I have suggested that some of the concerns of the non-factualist movement can be viewed in this light,
as attempts to prescribe the appropriate pattern of usage for particular classes of utterances – to show,
in particular, that the application of the notions of truth and falsity is sometimes inappropriate. I noted
that to interpret non-factualism in this way is to separate it from its usual analytic presuppositions.
However, we saw that prescriptive arguments of the above kind have a natural place in the analytic
view. So it may be that we should regard the prescriptive side of non-factualism as merely a legacy of
its usual analytic origins, and not try to reconstruct it on an explanatory basis.

My impression is that this is unduly pessimistic, and that there has been an aspect of non-factualism,
not particularly tied to the analytic view, involving a potentially reformist process of reflection on
usage. The point is incidental, however. Our main concern is with the explanatory view itself. In the
remainder of this chapter I want to consider some further applications. The first of these involves the
third of the main examples we discussed in part I – the case of conditionals.

8.7 CONDITIONALS

As we saw in chapter 5, the difficulties for a straightforward factualist interpretation of indicative

conditionals seem to stem from the behaviour of conditionals with false antecedents. The suggestion
that the indicative conditional is simply the truth-functional material conditional comes to grief on the

104
fact that English speakers do not take grounds for denying an antecedent as grounds for asserting a
conditional in which it appears. This fact underpins Adams’s hypothesis: the claim that the
assertibility of an indicative conditional follows the subjective conditional probability of the
consequent, given the antecedent. And Adams’s hypothesis, in turn, underpins the anti-factualist
arguments we discussed in chapter 5. Even if, as I claimed, these arguments are not conclusive, it is
clear that the behaviour of conditionals with false antecedents makes things very much more difficult
for a factualist interpretation.

From the explanatory viewpoint, however, there is a ready account of this behaviour. Suppose that the
main function of the commitments we express with indicative conditionals lies in hypothetical
reasoning. Crudely, the judgement that if P then Q is a kind of insurance, or contingency plan, against
the possibility that P. (This seems plausible enough, but the following explanation could be modified
to cope with other possibilities.) Presumably some contingency plans are better than others. Add the
communal element – the SBP – and we have what it takes to make the argumentative quest for
agreement a useful social practice. Individuals then stand to benefit, improving their contingency
plans by comparing them to those of others. And hence there is a role for truth, the dialectical catalyst
whose function is to provide the (internal) point of argument – to make disagreement seem to matter.

What then of false antecedents? How should we assess an insurance policy against an eventuality that
we know has not arisen, and will not arise? That is, what bearing should it have on our assessment of
a plan to deal with the contingency that P, that we learn that not-P? From the explanatory point of
view, the suggestion that a false antecedent makes a conditional true (or rather provides a reason for
calling it ‘true’) amounts, in effect, to the claim that any plan against an unrealized possibility is
worthy of approbation.

I think that this convention would have several disadvantages. The best that can be said for it is that
any such plan is bound to be harmless, in the immediate sense, because it will never be put into effect.
But this is not to say that we should encourage each other to carry such policies. For one thing, even a
redundant policy bears a cost. To encourage redundant plans would be to encourage unnecessary use
of limited cognitive resources. For another thing, to encourage such redundant conditional
commitments would be to devalue the incentives and disincentives on offer in the important cases –
those in which such a commitment is not known to be redundant.

Most importantly, however, to reward conditional judgements on the grounds of their redundancy
would be to undermine the function of such rewards in influencing habits of judgement. We noted
that argument tends to improve not merely the commitments that form the immediate grounds for a
disagreement, but also the general principles of inference that lead to such commitments. In the
conditional case, the latter form of improvement seems to require that speakers not escape criticism
simply in virtue of the fact that an inappropriately adopted conditional commitment turns out to be
redundant (due to a false antecedent). Otherwise it would be as if we were to excuse (or worse still,
commend) a habitual drunken driver, on the grounds that on the occasion in question he or she had
been lucky enough to avoid an accident.

A similar range of considerations seems to explain why we should not regard conditional judgements
with false antecedents as uniformly false. The first factor is different. The problem would be not that
we would waste cognitive space by encouraging redundant commitments, but that we would
underutilize the available resources by discouraging potentially useful commitments. (Why take on a
life insurance policy that in addition to the ordinary premium imposes a heavy penalty for failing to
die?) But the second and third factors are the same: this practice would devalue the incentives and
disincentives in the more significant cases; and most importantly, would interfere with the optimal use
of truth and falsity in encouraging the best habits of conditional judgement.

This seems to explain why (pace the material conditional interpretation) we do not argue about the
truth of indicative conditionals whose antecedents we take to be false. Conversely, we can also

105
explain why we do disagree about pairs of conditionals that on the material conditional view might
both be true. We are ordinarily inclined to treat ‘If P then Q’ and ‘If P then not-Q’ as incompatible,
rather than as compatible claims that together imply that not-P (as we should if we were dealing with
material conditionals). The contingency plan model of conditional judgement explains why this
should be so. To take on both commitments would be to subscribe to a pair of plans that could not
both be put into effect. Faced with the competing claims we therefore cannot accept both, unless in
the sense that we affirm that neither will be put into effect. The disadvantage of the material
conditional is that it makes this compromise the natural choice. Like the indexical treatment of
probabilistic judgement, it defuses potentially valuable disputes. The appropriate course is therefore to
treat the co-occurrence in a conversation of ‘If P then Q’ and ‘If P then not-Q’ as the basis of a
substantia l disagreement, and hence as prima facie grounds for a dispute. In this way we are provided
with an incentive to call in our evidence, and to attempt to justify our conditional commitments.

In the process we may find that the initial disagreement rested on the fact that the two sides relied on
different evidence, and that the combined evidence shows that not-P. We will thus be led to the
conclusion that the material conditional interpretation would have made immediate. But we are led to
it at the proper time. Properly conducted contingency planning may lead to the conclusion that a
particular contingency will not arise, on the grounds that it would entail inconsistent consequences.
That is, it may lead to the realization that such planning is inappropriate in the case in question. But
this is a far cry from discounting a possibility whenever we find ourselves faced with incompatible
plans for dealing with it.

Thus, it seems possible to explain in functional terms the most striking differences between the
ordinary use of truth and falsity with respect to indicative conditionals and the pattern that would be
appropriate for material conditionals. Of course, there are other factors that a more detailed treatment
of indicative conditionals would need to take into account. One such concerns the connection between
indicative and subjunctive conditionals. The focus of a disagreement about an indicative conditional
will often shift to a connected subjunctive, particularly if doubt is raised about the truth of the
antecedent. Suppose for example that we are arguing about whether if Rudolph is here, he is wearing
a wig. If we learn that Rudolph is not here, we are inclined to go on to argue about whether if he had
been here, he would have been wearing a wig. So a proper treatment along functional lines will need
to consider subjunctives, as well as indicatives.

However, if we allow that in such a case the function of the original argument is partly to improve our
habits of conditional judgement, there is an appealing explanation of the move to the subjunctive. For
even if we find out that the immediate judgement is redundant (because Rudolph is not here), the
more general potential benefits provide a reason for a continued dispute. As long as we take it that
there is still a substantial disagreement between us, we are liable to reassess the principles on which
our original judgments were based; and hence, on the whole, to do better in the future. This suggests
that subjunctive conditionals might themselves be explained as tokens in a process of reasoning and
argument whose significance lies at another level: in the general principles on the basis of which we
form the contingency plans with which we meet the future.

We saw above that indicative conditionals give rise to a species of evaporative disagreement, in cases
in which the effect of combined evidence is to convince the parties to a dispute that the conditional in
question has a false antecedent. Subjunctive conditionals seem to be subject to a different species. As
has often been noted, our assessments of counterfactual conditionals depend on background
assumptions. In order to reason about what we take not to be the case, we must rely on some but not
all of what we do take to be the case. To appeal to everything we currently accept would trivialize the
exercise, since in particular we could appeal to the fact that not-P in reasoning about what would be
the case if P.

How then do we select the background information to which we appeal? However we do it, it is
clearly sensitive to context. As a result, a disagreement about a subjunctive conditional may turn out

106
to stem from the fact that the participants have ‘held fixed’ different parts of their commo n
background evidence. Thus:

(8.1) If the author of Waverley had been alive today, he would have been over two hundred years
old (because Scott, the author of Waverley, was born in 1771);

and yet

(8.2) If the author of Waverley were alive today, he or she would certainly be less than one
hundred and thirty years old (because nobody lives longer than that).

Such cases suggest that a dispute about a subjunctive conditional can evaporate, if it turns out that the
participants are taking for granted different aspects of an agreed body of background information.
This behaviour seems consistent with the suggestion that the significance of subjunctive conditionals
lies in the general beliefs on which they are based. Evaporation here seems to follow recognition that
there is no disagreement about these underlying generalizations.

Subjunctives are a difficult case, however. No doubt there is much more to be said about the
connections between on the one hand their functions in speech and thought, and on the other their
usage in argument and as bearers of truth and falsity. I want to conclude these remarks by mentioning
an advantage of the present explanatory approach that seems particularly prominent in the case of
conditionals.

In recent discussions of indicative conditionals it has been noted that their interpretation is constrained
by their evident affinities, in ordinary usage, with disjunctions on the one side and subjunctive
conditionals on the other. (See particularly Jackson 1984, pp. 68–9; 1987, pp. 67–70.) These affinities
are problematic for any approach that wants to draw a major semantic distinction between
disjunctions and indicatives or between indicatives and subjunctives – for example, the view that
disjunctions have ‘genuine’ truth conditions (given truth-functionally), whereas indicative and
subjunctive conditionals are ‘projections’ of non-truth-conditional mental dispositions. Why should
classes of utterances whose meanings are determined in such different ways behave so similarly? I
want to emphasize that here, as in other areas of discourse, the explanatory approach offers a unitary
approach. Hence it avoids these problems. At the same time, however, it has (or at any rate claims to
have) the flexibility needed to account for the relevant range of variation we find in usage – variations
in patterns of argument and in the application of the notions of truth and falsity.

8.8 PRESUPPOSITION

Conditionals have often been linked in philosophical discussion with assertions dependent on
presuppositions – themselves a class of utterances whose truth-conditionality has been thought to be
problematic. Much of the discussion of the latter class concentrates on the question as to whether a
sentence such as

(8.3) The king of France is the author of Waverley

can be considered to have a truth value, given that it depends on a false presupposition. I want to
comment briefly on this topic, partly as a further illustration of the explanatory approach to such
issues, and partly to exhibit a contrast to the conditional case.

From the explanatory point of view the question as to whether 8.3 ‘really’ has a truth value is
misconceived, of course. We can only ask whether and why such an utterance is or should be treated
as having a truth value. This aside, it may seem that this case ought to be similar to that of an

107
indicative conditional. Isn’t discovering that a presupposition is false similar in effect to discovering
that an antecedent is false?

I think there is actually an important difference between these cases. The difference between 8.3 and

(8.4) If there is a king of France, then he is the author of Waverley

is that a commitment to 8.4, unlike to 8.3, involves no commitment to the claim that there is a king of
France. Other things being equal, acceptance of 8.3 can thus be expected to be accompanied by
behavioural effects not shared by acceptance of 8.4: namely, those effects that would be expected to
flow from the belief that there is a king of France. These effects give a point to criticism on the
grounds of the failure of a presupposition, not shared by the conditional case.

This is not to say that in virtue of having such effects 8.3 is false when there is no king of France. It is
simply to say that in virtue of these effects, the practice of describing such an utterance as ‘false’ in
such circumstances is likely to serve a useful function. Perhaps the same function could be performed
in other ways. For example, we might have a separate mode of criticism, reserved for this kind of
mistake. The intuition that some philosophers have had that failure of a presupposition renders an
utterance neither true nor false is perhaps evidence for the existence in usage of some degree of
separation. Certainly it suggests that it would be within our capabilities to sustain such a distinction.
But this is not a concession to the analytic standpoint – the basis for an account of ‘real’ as opposed to
‘constructed’ truth. For why stop here? Why not discern a separate kind of truth for every subject
matter and every critical nuance? We saw in part I that what one needs to worry about, in leaping
aboard the projectivist bandwagon, is not the engine but the brakes. Getting it started is easy enough.
The problem is to pull the thing up where the projectivist wants to stop.

The question arises as to whether our explanatory view is simply projectivism unconstrained – the
idealism that we saw to be the projectivist’s most threatening danger. I shall return to this in the
concluding chapter. I want to turn now to some preliminary questions. How widespread are
evaporative disputes? Does the explanatory approach allow us to draw a distinction between those
utterances that admit such disputes and those that do not? And if so, is there anything to be said for
the idea that the latter are the genuine statements of fact?

8.9 FACTUALITY: A MATTER OF DEGREE?

Evaporative disputes are a boundary phenomenon, a manifestation of the felt inappropriateness of a

continued pretence that a particular disagreement is answerable to an external standard of correctness
and incorrectness. The cases we have so far considered suggest that the phenomenon is a product of
the local peculiarities of particular types of utterance. Hence we might expect that in a broad range of
central cases, it would be unknown. This expectation might in turn suggest a characterization of the
‘genuine’ statements of fact: namely, that they are those indicative utterances not prone to evaporative
dispute.

Interpreted correctly, this characterization would be unobjectionable. If there is a distinction in usage

between those indicative utterances that do and those that do not give rise to evaporative disputes,
then by all means let us mark it. But this unobjectionable interpretation is not the source of the
suggestion’s appeal. This lies rather in the misconceived intuition that the suggested characterization
gives us the long-sought analysis of statement of fact – a venerable confusion of taxonomy for
analysis.

That aside, it seems to me that the assumption concerning usage that underlies this suggestion is
significantly mistaken. There seems to be at least one potentially global source of evaporative
disputes. Hence it is doubtful whether there is any class of indicative utterances conventionally
immune from the patterns of usage that supposedly distinguish ‘non-genuine’ statements of fact. This

108
global source is a form of indeterminacy of meaning. In section 8.2 I mentioned the possibility that
apparent disagreements may turn out to rest on the fact that the speakers concerned are using the same
term with different meanings. This possibility threatened to trivialize the issue about relativism. To
defend a space for an interesting relativist position, we suggested that the motivational character of
moral judgement provides a fixed component in the meaning of moral terms, around which the
relativist issue can be built. But the problems raised by meaning variance are not exhausted by its
abortive role in this case.

Meaning variance can affect disputes in several ways, some of them relatively uninteresting. Perhaps
the dullest involves a term that is known to be ambiguous in ordinary usage, whose multiple meanings
find their way to the heart of an argument – for example as to whether in the modern computerized
hospital the doctors should be expected to grasp the rudiments of ‘terminal care’. Marginally more
interesting is the case in which a dispute stems from the fact that one participant has simply
misunderstood the meaning of a central term. This raises an issue as to whether (and if so, why) we
should say, when the mistake becomes apparent, that this person’s original assertion was false. This
issue seems to have something in common with the case of false presupposition. Indeed, it could be
argued to be a sub-case of that of presupposition, on the grounds that assumptions about meaning can
be treated as presuppositions. Alternatively – I think it is not the same point – it could be said that we
should not expect, post-Quine, a sharp distinction between mistakes about meanings and mistakes
about facts.

I want to concentrate on a third kind of case, however. It does not depend on ambiguity, in the normal
sense, or on any mistake about meaning. It stems rather from a form of indeterminacy about meaning
that seems endemic to language. In recent years philosophers have paid a lot of attention to the fact
that language is finitely acquired. Each of us has acquired her or his linguistic knowledge in the few
short years since birth – in so far, at any rate, as we did not have it at birth. This observation has been
applied in a number of ways: by Chomsky and his followers, in arguing for the existence of innate
linguistic ability; by Dummett, in his argument for an anti-realist conception of truth; and in the
literature stemming from Wittgenstein’s treatment of the notion of following a rule. And a closely
similar factor, concerning the finiteness of the evidence available to the radical interpreter, is crucial
to Quine’s indeterminacy thesis, developed in Word and Object.

For our purposes, the significant point is that no finite level of experience can determine the
application of a linguistic term to all possible cases, or exclude the possibility that two speakers –
fully competent speakers, by existing community standards – will diverge in their application of the
term concerned to some future case. It is true that this does not happen very often. The explanation of
this fact, particularly in the light of finite acquisition, is one of the central problems in the philosophy
of language. (Wittgenstein’s treatment of rule following in the Philosophical Investigations can be
seen as an attack on one conception of what a solution would look like. Cf. McGinn, 1984, chapter 1.)
Given however that it can happen, and cannot be excluded in advance, it seems that we live with the
possibility that disagreements may turn out to stem from the fact that participants have made different
(though equally consistent) ‘projections’ of a term from a common finite base.

What happens in cases such as this? One common if often abused response is to say that the disputed
point is ‘terminological’. I think the important thing is that we recognize that neither party is
mistaken, in either of two ways: in using the term concerned with what is the wrong meaning, by
previously established standards; or in a sense that would lead us to say that they had spoken falsely,
given their own reading of the equivocal term. The dispute thus evaporates. Neither side concedes,
and yet each recognizes that there is no longer a point in insisting that the case is answerable to an
external standard, that only one side can meet.

These cases reflect the intuitively appealing idea that our judgements have truth values only in so far
as they are determinate in meaning. An ambiguous judgement may have a determinate truth value by
default, as it were, if all its possible interpretations share the same truth value. But there is no

109
guarantee that this will be so, in general – ambiguities tend to show up when it is not so. And in any
case, it is far from clear how to apply this notion when we are not dealing with an ambiguity that is
finitely specifiable in terms of non-ambiguous atoms. To recognize ineliminable indeterminacy of
meaning seems to be to recognize that our utterances are at best approximations to genuine (and
therefore strictly truth-evaluable) statements of fact.

I hope it is no longer necessary to emphasize that the present approach does not claim to show that
such intuitions are correct. We have offered no analysis that would afford such a judgement. Here, as
elsewhere, our point is rather to observe that ordinary intuitions concerning the limits of factuality can
be correlated with a certain pattern of usage – the existence of evaporative disputes – and to argue that
this pattern can be explained in terms of a general account of truth, falsity and the function of
argument.

The indeterminacy of meaning thus seems to be the basis for a form of universal non-factualism. That
is, it offers a reason for departing from the pattern of usage characteristic of the statement form that
may be applicable in any area of discourse. This topic-neutral form of non-factualism should not be
confused with the form that Quine himself extracts from the indeterminacy thesis: the view that there
are no genuine facts about meaning. (Cf. Quine 1960, p. 221.) Ignoring for a moment our scruples
about the analytic viewpoint, we might say that the present view, in contrast, is that in virtue of the
nature of meaning there can be no genuine facts about anything.

More correctly, the result suggests that the factuality of usage is always a matter of degree. Across the
broadly fact-stating part of language – in English, roughly, the indicatives – the appropriateness of the
statement form varies from case to case, both type to type and token to token. From the explanatory
point of view there is nothing surprising about this conclusion. Only the analytic presumption leads us
to expect a sharp distinction.

8.10 THE SOURCES OF SUBJECTIVITY

What then can be said in general about the sources of evaporative disagreement, and about its
significance in other areas in which non-factualism has seemed attractive? A survey and prospectus
seem in order.

Disputes evaporate when disputees decide that in virtue of the peculiarities of their own standpoints it
is not appropriate for each to insist that the other is mistaken. Evaporation is a manifestation of a
degree of felt subjectivity in judgement. The cases discussed above illustrate the range of potential
sources of such subjectivity. On the one hand are cases in which the source lies in the context or
conversational role of the judgements concerned. Probabilistic and conditional judgements seemed to
be of this kind; their tendency to yield insubstantial disagreements being attributable to their
distinctive roles in discourse, and hence their sensitivity to the evidential context in which they are
used. On the other hand are cases in which the source of subjectivity seems to lie more in some
distinctive feature of the speakers themselves – in the idiosyncrasies of their grasp of the meaning of a
relevant term, for example. (A more familiar case of this kind might be the subjectivity of judgements
about secondary qualities; more on this in a moment.)

The distinction between intrinsic and extrinsic sources of subjectivity is far from sharp. The sources
of moral or aesthetic relativity seem to lie somewhere between the two extremes, for example.
However, I think it provides sufficient structure to enable us to locate the likely basis of evaporation
and hence non-factualism itself in some of the other areas in which such an approach has seemed
attractive.

On the contextual side, the modal operators are the obvious candidates. The epistemological approach
to possibility and necessity can be seen as an attempt to clarify the dependence of modal judgement
on evidential context; even if usually under the misguided assumption that what is called for is an

110
analysis of modal judgement – an account of its truth conditions, or some such. Another possible
candidate is causal judgement. Here, particularly, the issue is clouded by the appeal of Humean
metaphysics – a route to non-factualism whose viability we disputed in part I. Metaphysics aside,
however, non-factualism about causality has an appeal that rests on contextual and conversational
factors. This is not the place to explore these factors in detail, but I mention two. In view of the
connections between causality and probability – particularly the intuition that causes increase the
probability of their effects – disagreements about single case causes seem able to evaporate when the
same happens to connected judgements about probability. And the well-recognized association
between causal judgement and counterfactual conditionals suggests that the sensitivity of the latter to
background assumptions – about ‘the normal course of events’, for example – will also give rise to
insubstantial disagreements about causality.

At the other end of the scale, I have already mentioned judgements about secondary qualities. It has
long been felt that these exhibit a degree of subjectivity, attributable to their dependence on
contingent features of our perceptual apparatus. The view is supported by cases of major variation
between human subjects in sensory response to a given stimulus. A substance may taste bitter to some
subjects yet be tasteless to others. It is interesting that our willingness to treat such a disagreement as
insubstantial depends on the relative frequency of the different responses. We are inclined to say that
the aberrant judgements of colour-blind subjects are indeed mistaken – that it is a fact that jacarandas
are mauve rather than blue. But who would insist that chemical X is really bitter, when about 50 per
cent of people find it tasteless? This phenomenon is easily explained from the functional viewpoint, of
course; but again, it is troublesome for an account that expects a sharp underlying distinction.

What else lies at this end of the scale? I mentioned the possibility of idiosyncratic grasp of meaning –
a potentially universal source of evaporative argument. My impression here is that among the
determinants of meaning there might be some we would count as strongly individual – perhaps innate
dispositions to categorize experience in certain ways – and others more dependent on factors external
to the individual speaker, such as communal systems of convention. The latter possibility suggests a
place in the scheme for another topic about which non-factualism has been popular, namely
theoretical discourse. The recent popularity of scientific anti-realism stems largely from an
appreciation of the difficulties of inter-theoretical discourse. Once it is recognized that proponents of
different scientific theories may lack a basis for argument, it has seemed difficult to hold onto the idea
that each is really making claims about the same objective world. The present view suggests that the
source of the problem is an unacknowledged commitment to the notion that there is such a thing as
genuinely fact-stating discourse. If we reject this notion, we are free to say that the inter-theoretical
case is simply a prominent form of a phenomenon endemic to language. Theoretical discourse is not a
special case; it differs only in degree from discourse of any kind. (The anti-realist might agree, of
course, but interpret the result as global anti-realism. More on this in chapter 9.)

How much of non-factualism can be explained in these terms? It is difficult to make any firm
pronouncement – the field is quite large, after all – but the prospects seem to me to be good. If there is
a major area of uncertainty it seems to lie in the cases in which non-factualism is motivated by a
combination of metaphysical and epistemological considerations. I argued in part I that a determined
factualist should be untroubled by these factors. The appropriate conclusion might therefore be that in
so far as it rests on such intuitions, non-factualism is simply mistaken, and should not be expected to
conform to the pattern we have identified. However, there are several less dismissive interpretations
of an appeal to such factors.

The first involves the interpretation of a disputed non-factualist thesis simply as a first order claim
about what there is in the world, and not also as a second order claim about the nature of discourse.
The practical difference here is that the weaker thesis, unlike a typical non-factualism, will
recommend that we abandon the form of discourse in question. The force of the argument of chapters
4 and 5 was that if we are to retain an area of linguistic practice, the non-factualist cannot dislodge a
determined defence of a factualist interpretation of that practice. This leaves it open for someone to

111
argue that we should discard the practice, however interpreted. Some of the appeal to metaphysics and
epistemology in these contexts might be taken to be of this kind.

Secondly, it is easy to imagine that someone might appeal to metaphysical and epistemological
considerations in order to establish the relatively theoretical nature of a particular area of discourse.
To give a crude example, it might be said that since we don’t directly perceive probabilities,
probability must be a theoretical notion. Moves of this kind bring the discourse in question within the
scope of a general claim about the status of theoretical language. We have seen that theoretical non-
factualism can be brought within the pattern suggested by the explanatory approach. This
interpretation of the appeal to metaphysics and epistemology thus brings it within the scope of our
general account of non-factualism.

The last possibility is in some ways similar. There is clearly a role for metaphysical conceptual
analysis in an approach of this general kind. We have already noted an example: an analysis of
causality in terms of probability, if sustained, would be likely to ensure that causal judgement was
similarly liable to evaporative disagreement. (This might cut both ways: the lack of such disagreement
in the causal case might count against such a conceptual analysis.)

These possibilities suggest that there is a significant place for metaphysical and epistemological
concerns in an explanatory approach to non-factualist concerns. But I must leave it to the reader to
judge to what extent existing appeals to these factors can be interpreted in any of these ways – as, of
course, to assess the approach to non-factualism as a whole.

8.11 THE SIGNIFICANCE OF MOOD

What then of the non-indicatives? Here, more than anywhere, usage may seem to call for the analyst’s
sharp dichotomy. Hence, as we saw, the appeal of the marker view. But here too, I think the call
should be resisted. It can be countered by the strategy of indicative paraphrase, which was the basis of
our argument against the marker view in chapter 3. And we can now add to that argument a possible
explanation for the existence of the non-indicative moods, in terms of our account of the function of
truth-evaluability.

If a non-indicative is paraphrased in the way we suggested in chapter 3, it is normally true in virtue of

being uttered. At any rate, we saw that the suggested indicative paraphrase is such that its truth is
normally in the gift of a speaker who utters it. In so far as non-indicatives are informative, the
information they carry is not normally open to dispute. This suggests that if the linguistic functions
served by the non-indicatives were to be assigned to indicatives, the argument-encouraging machinery
associated with the indicative mood would usually be redundant in these cases. The qualification is
important: first because as we’ll see in a moment, such machinery would not always be redundant; but
more importantly because the evidence suggests that limited redundancy is often tolerated in the
interests of the general benefits of the fact-stating form. After all, there are many ordinary indicatives
whose truth is similarly guaranteed by utterance – ‘I have not lost the power of speech’, for example.

To explain why more frequent redundancy should be disadvantageous, I think we need to mention
another factor: namely, that unused machinery may be a positive liability. For one thing, the
argumentative mode of speech is presumably sustained at some conceptual cost. It requires that
speakers hold themselves ready to assess the utterances of others by the standards reflected in
judgements of truth and falsity. Where such assessment is generally inappropriate, or predetermined
by the nature of the utterance, the general conceptual cost may not be worth the very occasional
benefit. For another thing, the possibility of argument might itself be disadvantageous. If non-
indicatives serve linguistic functions that call for specific and often immediate responses from an
audience – answering a question, meeting a request, and so on – then it might well be counter-
productive to use a form of language that encourages a different kind of response: namely, an
argumentative challenge to the ‘correctness ‘ of the utterance in question. This is not to say that one

112
cannot challenge a non-indicative utterance. Clearly one can, and we do – amongst other things on the
kinds of grounds that would constitute a challenge to the truth of the kind of indicative paraphrases
suggested in chapter 3. Ordered to complete a form, for example, we may object that we are not by
law required to do so. The point is simply that there may be a social advantage in not actively
encouraging such challenges.

These factors, taken together with the inevitable element of historical accident, are perhaps as much
as we can expect of an explanation of the existence of the non-indicative moods. The striking feature
of these moods is the fact that they are not taken to be truth-evaluable, in ordinary usage. The above
factors bring this fact within the scope of our explanatory account of the role of truth and falsity. It
would seem that the non-indicative moods occupy a region of language in which, on the whole, the
disadvantages of the notion of truth outweigh its advantages. But the boundaries of this region are
necessarily imprecise, and the sharpness of the mood distinction, if taken at face value, is liable to be
misleading. The linguistic functions it masks embody no corresponding discontinuity, but rather a
complex and multi-dimensional field of factors, varying by utterance token as well as by utterance
type.

An artificial order is imposed on this field by the need for regularity in language. The rules of
language cannot be sensitive to every nuance of the individual case. But we should not underestimate
the considerable flexibility that does exist in natural language. In practice the mood distinction is far
from sharp. As we have seen, English speakers will switch between indicative and non-indicative
utterances for what are recognizably similar linguistic tasks – according, at least in part, to the
perceived role for an idea of an objective standard. And within the indicative mood, we have observed
the striking tendency of ordinary speakers to recognize the limits of truth and falsity, to opt for
tolerance, and to allow inappropriate disputes to evaporate. Usage is very much less sharply polarized
than first appearances might suggest. It is therefore a point in favour of the present explanatory
approach that it does not advocate or assume a sharp distinction between fact-stating and non-fact-
stating uses of language.

113
9
End of Enquiry

We began with the conflict between non-factualism and Fregean linguistic theory – a somewhat ironic
tension, we noted, given that non-factualism was a distinctive and popular product of the mid-century
linguistic movement in philosophy. In this concluding chapter I want to begin by showing how this
conflict seems to be resolved by the present explanatory approach to truth and factuality. This will
lead us to the question as to why, on the present account, the notion of truth should be the kind of
notion that it is. We encountered this question in chapter 7, and there and elsewhere have touched on
the main aspects of the answer. Here I shall try to draw together the threads. As foreshadowed in
chapter 2, David Wiggins’s ‘marks of truth’ provide a useful focus at this point. We want to confirm
that the proposed functional account can explain why truth should have such characteristics. The third
concern of the chapter will then be to clarify some of the connections and advantages of the functional
approach, with respect to projectivism, quietism and other positions in contemporary debates about
realism and objectivity. We’ll finish with some comments on the scope of the functional perspective
in other areas, including particularly its application to the problem of the status of folk psychology.

9.1 TRUTH: A CONFLICT RESOLVED

In chapter 2 we looked at Wiggins’s attempt to ground an account of assertion on a substantial notion

of truth, itself derived from a Davidsonian programme for radical interpretation. Our main criticism
was that the programme of radical interpretation turns out to be insensitive to the distinction between
fact-stating and non-fact-stating parts of language. The theorems of the radical interpreter’s so-called
truth theory do not have to mention truth at all; and if they do mention it, need only the disquotational
property – the equivalence principle – in terms of which the truth predicate can be extended trivially
to utterances of any kind.

The impression that Fregean general semantics requires a substantial notion of truth thus rests on an
uncritical acceptance of the doctrine of the primacy of assertion. If we equate saying with asserting,
then of course we equate the disquotational predicate needed in a Fregean theory of meaning with the
ordinary predicate ‘... is true’. However, if we keep in mind that there may be sayings that are not
assertings then such an equation looks highly implausible – the ordinary predicate does not extend to
non-assertoric sayings. Hence the needs of radical interpretation, or of a systematic theory of meaning
of the kind it requires, seem to have little bearing on the question as to why the ordinary notion of
truth should be restricted in this way.

I think that the resolution of the conflict between non-factualism and the Fregean programme for
general semantics thus turns on a distinction between the disquotational and the normative properties
of truth. Contemporary semantic theories exploit the disquotational property. The non-factualist,
however, needs to be concerned with the normative property; for only the normative property seems
able to explain the limited application of the ordinary notion of truth, and hence to ensure that truth
has something to do with a distinction of language into ‘fact-stating’ and ‘non-fact-stating’ parts.

A Fregean might object that one thing in favour of a truth-conditional approach to a theory of
meaning is the intuition that to know the meaning of a sentence is to know its correct use – to know
under what conditions it would be correct to utter that sentence (with serious intent ). This intuition
surely depends on the normative notion of correctness. So to the extent that it calls for a truth-
conditional theory of meaning, it seems to require a normative notion of truth, rather than merely a
disquotational notion.

114
However, I think this objection trades on an ambiguity in the notion of correct use. In the case of
assertoric discourse this ambiguity is well recognized, if not always well heeded. The term
‘assertibility condition’ (or ‘condition for correct assertion’) can mean one of two things. Taken
subjectively, it refers to something like the state of mind that would normally be considered to be
evidenced by a competent speaker’s utterance of the sentence in question. According to the
conventional model, a belief that P is thus the (subjective) assertibility condition for the sentence P;
and there seems to be something to be said for the claim that to know the meaning of P is to know that
this is the case. Taken objectively, on the other hand, ‘assertion condition’ refers to some external
state of the world, whose obtaining would normally ‘justify’ the assertoric use of the sentence in
question. This reading itself admits of various interpretations; for present purposes the important thing
is that at least one interpretation has the effect of equating ‘correctness’ with ‘truth’. And again, there
seems to be something to be said for the claim that to know the meaning of an assertoric sentence P is
to know when its use would be correct in this sense – in other words, to know its truth conditions.

As long as we restrict ourselves to assertoric discourse, there is little incentive to keep in mind that
there are these two readings of the principle that to know meaning is to know correct use. To the
extent that they are distinguished, the objective notion has seemed of greater interest. But if we
broaden our field of view to include the non-assertoric part of language, I think we should regard this
as another case of taking for granted what we ought to explain. The availability of the subjective
reading shows that to decline to rely on the objective reading is not to reject the principle itself. And
in extending to non-assertoric discourse, the subjective reading has the generality that the objective
reading lacks. In practice, indeed, the subjective notion of an assertibility condition has tended to
receive attention in cases in which the notion of truth seems problematic, such as that of indicative
conditionals.

We can thus aim initially at a theory of meaning that does not depend on a radical distinction between
the assertoric and non-assertoric part s of language (regarding this rather as something that calls for
explanation). Apart from its elegance, generality and explanatory potential, this approach has the
advantage we discovered in chapter 2: it turns out to be a product of its opponents’ best efforts at
following the alternative approach, given that truth theories do not in fact rely on the normative aspect
of truth.

9.2 WHY IS TRUTH TRUTH-LIKE ?

This separation of the disquotational and normative aspects of truth may itself seem problematic. Why
do these aspects go together? How is it that the ordinary normative truth predicate functions
disquotationally? These are questions that an explanatory theory of truth should be expected to
answer.

More generally, an account of truth on the proposed lines should be able to relate the features of truth
as we find it to the functional role that is being claimed to explain its development in language. True,
there is some flexibility in an evolutionary account of this kind. Some features of the ordinary notion
may be more or less accidental. But the plausibility of any such claim seems to rest on the relative
insignificance of the feature in question. The more a property seems essential to truth, the more we
should expect an account of why truth should have such a property. In particular, we should expect an
account of the marks of the concept of truth, in Wiggins’s sense (1980, p. 194).

In the case of the connection between the disquotational and normative aspects of truth I think the
answer is fairly straightforward. To explain the disquotational property we have to explain why P and
‘P is true’ can in some sense be regarded as equivalent assertions – for example, that they can be
taken to be expressions of the same commitment. On the present theory, to say that P is true is to
allow to judgements that P a conventionally-generated property that we are all conditioned to want
our judgements to have. Roughly, it is to ascribe worth to judgements that P. The point of this
conventional system of value is to encourage argument. In making our case for these conventional

115
rewards, we indulge in a process whose effect tends to be to improve our behavioural commitments.
But this improvement depends on the fact that (in the main, at any rate) argument is prompted by
some mismatch between individual behavioural commitments, and resolved by a resolution of such a
mismatch. So the process requires that we ascribe value to those judgements we are prepared to make
ourselves. Only in this way can the ascription of these conventional rewards make socially unstable
the appropriate cases of mismatched commitment. In other words, the role of truth in argument
requires the disquotational property – it requires that we be prepared to acknowledge that P is true in
the same circumstances as we acknowledge that P.

This answer bears on the problems that might be thought to flow from the fact that ascriptions of truth
and falsity themselves take the grammatical form of indicative sentences, and are thus the sorts of
expressions to which ‘true’ and ‘false’ can be ascribed. We touched on this point in chapter 8,
mentioning in particular the objection that it might commit us to an endless hierarchy of
disagreements, corresponding to iterations of the predicates ‘... is true’ and ‘... is false’. In explaining
how we avoid embarking on such an escalation of ordinary disagreements, the disquotational property
is part of the answer. It means, I think, that from the point of view of our behavioural commitments
there is no reason to escalate. Behaviourally speaking, semantic ascent doesn’t increase the stakes.

The other part of the answer, I think, is that in the case of disagreements about truth and falsity
themselves there isn’t the normal reason to introduce these normative terms. Truth comes into the
picture to provide the incentive for argument. But once it is already in the picture – when we are
disagreeing about the truth of some sentence – there is nothing to be gained by bringing it in again, by
moving to second order truth ascriptions. There’s a prize for winning the argument, but no further
prize for winning the prize. This is not to say that ascriptions of truth and falsity are not themselves
regarded as answerable to the kind of objective standard the concept of which their use embodies. On
the contrary, as we stressed in chapter 8, it is crucial to the motivating role of such evaluations (as of
others) that they be regarded as objective – that there be no easy subjectivist escape hatch. It is simply
to say that in the case of truth and falsity themselves, the manifestations of this felt answerability to an
external standard do not have their usual importance in argument.

This brings us to Wiggins’s marks of truth. We are now in a position to explain why truth should have
the characteristics that Wiggins identifies – or rather, as we’ll see, why it should have those four of
the five such characteristics that are properly regarded as properties of truth.

Wiggins’s first mark is the normativity of truth. It consists in truth’s being ‘the primary dimension of
assessment for sentences’ – ‘that property which sentences have normally to be construed as aiming
to enjoy’ (1980, p. 205). This property is central, of course, to our proposed account of the function of
truth in discourse. Unless we take truth as something to be aimed at, we will be given no incentive to
argue by the claim that one of our utterances lacks it.

The second mark of truth concerns its attainability. Wiggins puts it as follows.

A second mark of the [truth] of a sentence s would appear to consist in s’s having a content
such that with respect to this content there should be a tendency, under favourable conditions
of investigation, for disagreement to diminish and for opinion to converge in agreement.
(1980, p. 206)

Here the relevant point would seem to be that unless speakers subscribe to some such view of truth
then even its normativity will not encourage them to argue. Heroics aside, why should we argue with
someone unless initially there seems some prospect that he can be brought to share our own opinion –
in other words, some prospect of agreement? A degree of attainability thus seems a necessary part of
our idea of truth, if truth is to serve the role we have claimed for it in encouraging argument.

116
Similar remarks apply to Wiggins’s third and fourth marks of truth. The third mark (1980, p. 208) is
that truth be considered independent of a speaker’s means of recognizing whether a statement
possesses it; while the fourth (p. 211) is that true sentences be considered to be true ‘in virtue of
something’. These would appear to be aspects of the required objectivity of truth, on which we
commented a moment ago. The power of truth as a system of incentive depends on the fact that it is
not taken to be able to be bestowed at will. This is what distinguishes the incentive provided by truth
from the incentive provided by a system of boos and cheers. Rival views may share the available
applause, but they cannot share the truth – or so, at any rate, it is useful for us to believe.

So to Wiggins’s fifth claimed mark of truth (1980, p. 211): if two sentences are true then so is their
conjunction. It is difficult to take this as a property of truth, rather than as an account of conjunction
in terms of truth. There is a legitimate question as to why we should have a sentence-forming operator
and such that we are prepared to assert P and Q when and only when we are prepared to assert both P
and Q individually. Given however that we do have such an operator, the fifth mark follows directly
from the disquotational property – from the fact that we are prepared to assert that P when and only
when we are prepared to assert that P is true.

9.3 THE TRUTH OF TRUTH ASCRIPTIONS: SOME OBJECTIONS EVADED

I mentioned above the treatment of truth ascriptions themselves by the present explanatory account of
truth. I want to draw attention to some problems that trouble analytic non-factualists at this point –
problems that again the explanatory approach avoids.

For one thing, the analytic point of view leaves a non-factualist with the following dilemma. Either ‘P
is true’ is sometimes a statement and sometimes not a statement, depending on the status of P itself; or
all such truth ascriptions have the same status. Neither choice is particularly appealing. The former
entails that truth ascriptions are not the semantically homogeneous class their form suggests. While
the latter implies that ‘P is true ‘can be a genuine statement even though P itself is not a statement, or
vice versa; and hence that the equivalence principle fails. From the explanatory perspective, however,
the dilemma looks entirely self-inflicted – a product of the assumption that any utterance lies on one
side or other of a real semantic divide.

Arthur Fine has recently raised a related objection to a treatment of truth on projectivist lines:

The [projectivist] strategy requires a thin, unproblematic discourse on which to base its
projections. To project out ‘truth’, the thin discourse must be truth free. But where are we to
find any such thin discourse? The redundancy property of truth makes truth a part of
discourse that merits the name. In particular, if we go for acceptance or coherence theories of
truth we try to build truth from certain systems of judgements (or the like). But the judgement
that P, and the judgement it is true that P are the same. You cannot have one without the
other. That is to say, you cannot include one in the system without thereby including the other
.... Thus there is no thin discourse from which the [projectivist] explanations of truth talk
could begin. (1986, pp. 169-70)

This objection is like the above one in squeezing a non-factualist account of truth between on the one
side the desire for a unified theory of truth, and on the other the equivalence principle (or redundancy
property). If the projectivist is prepared to concede unification then there is again an easy way out: to
offer separate accounts of ‘thin’ truth and ‘thick’ truth – separate interpretations of truth ascriptions to
utterances that are themselves about, respectively, ‘genuine’ or ‘projected’ facts. (This seems to be the
approach taken by Blackburn 1984, though not without reservations; see particularly pp. 256–7.)

However, I think that in Fine’s case, as in the earlier one, the explanatory approach needs no such
reply. Because it is committed to no prior bifurcation in language, it cannot be turned on itself in the
way that an analytic theory is bound to find embarrassing. True, its functional account of the role of

117
truth, and fact-stating discourse in general, will itself be couched in fact-stating terms. As its
proponents, we will be inclined to assert its truth. But there is nothing more vicious in this than there
ever is in speaking or writing about the origins of language. If there were no more to Fine’s objection
than the observation that the projectivist must use ‘true’ (in effect, at any rate) in offering a
constructive theory of truth, the point would be trivial and irrelevant. Its force therefore rests on the
consequences of bifurcation, to which the projectivist is committed but our theory is not.

9.4 FUNCTIONAL REALISM: NOT IDEALISM ...

What then of the suggestion that the present theory is simply idealism? As we saw in chapter 4, the
idealist takes over the projectivist’s goal of accounting for the realist features of linguistic practice but
aims to do it quite generally. Our approach thus concurs with idealism in rejecting the projectivist’s
bifurcation between real and projected facts, arguing that there is a single appropriate account of truth,
applicable wherever truth is. Given that the idealist takes over the projectivist’s constructive
programme, we also agree on an explanatory as opposed to an analytic approach to truth, and perhaps
on some of the details.

The main difference, I think, concerns the treatment of non-factualist claims. The projectivist aims for
‘quasi-realism’, as Blackburn calls it – for a demonstration on projectivist lines that we are entitled to
talk just as if there were genuine facts of the disputed kinds. Blackburn’s quasi-realist seems not to
countenance differences in usage between the disputed utterances and genuine statements, aiming
instead for the perfect pretence. So when the idealist challenges the distinction between pretence and
reality, non-factualism, having been lashed to quasi-realism, follows it down the slope.

The present theory, on the other hand, goes out of its way to draw attention to some characteristic
peculiarities of the kinds of discourse with which non-factualists have been concerned. It tries to
associate these peculiarities with particular variations in the factors that seem normally to account for
the functional advantages of truth and related features of usage. In the process, as well as
demonstrating its ability to explain these cases, it offers some foundation for the intuitions that have
usually motivated non-factualists. True, this usage-based descriptive–explanatory programme is not
what non-factualists have usually had in mind. But unlike Blackburn’s quasi-realist projectivism, it
does offer a refuge for the non-factualist against the push to generalize the constructive or explanatory
approach to truth.

In summary, I think the present approach concurs with the idealist in seeking to globalize the
projectivist’s constructive approach to certain areas of reality – the attempt to ground our conception
of the world ‘internally’, so to speak, in our own patterns of reaction to our environment. Unlike the
idealist, however, we find room in this for a kind of non-factualism. This aside, the term ‘idealism’
has associations the present account will want to avoid. It is apt in particular to suggest an analytic
approach to truth, in terms of coherence, or some such – an approach quite different from ours and
from that of the projectivist’s opponent. The present theory is best described as functional or
explanatory realism. Its goal is to explain our realist practices; as well, of course, as their non-
factualist variations.

In expressing even this much sympathy for idealism I am likely to be accused of denying common
sense intuitions about the reality of the external world. So I want to make it clear that my views about
most aspects of the external world are the same as those of the most hard-nosed realist. We each hold
true much the same class of propositions about how things are. Our professional differences concern
at most a tiny fraction of such propositions: those that deal with the explanation of the fact that
humans hold such views. For my part, I believe that at least in so far as holding a view about the
world depends on our possession of the notion of truth or correct representation, the fact that we do so
is explicable in the kind of terms here suggested. Hard-nosed realists may have different views on this
topic, or none at all. But this doesn’t prevent us agreeing about almost everything else (in other words,
about almost everything).

118
In particular, it would be a mistake to think that because I have a biological explanation of the fact
that we apply a notion of correctness to judgement I am unable to apply such a notion myself. That
would be as absurd as supposing that a biologist who explains hunger in evolutionary terms should no
longer feel the need to eat. Indeed, it would perhaps be more absurd, for whereas to hold a theory
about hunger is not to feel hungry, a belief about the origins of truth is an exercise of the very
biological capacity it itself concerns. So it demonstrates its author’s continued ability to exercise this
capacity.

9.5 ... NOT INSTRUMENTALISM ...

I should perhaps emphasize also that functional realism is not a form of instrumentalism. Indeed, it is
likely to reject instrumentalism on the same grounds as it rejects projectivism. Both these approaches
ask us to make a distinction between two kinds of superficially factual discourse: the genuinely
factual areas, on the one hand, and the instrumental or projected areas, on the other. The present
theory denies that we can draw such a distinction. The best it can offer an instrumentalist is a place in
the ranks of the reconstructed non-factualists; and this only if the instrumentalist can point to the
relevant sort of difference in usage between the supposed instrumental discourse (theoretical
judgements in science, or whatever) and the canonical assertoric form.

All the same, I think there is a nice analogy between the instrumentalist’s most powerful argument
against orthodox scientific realism and an important argument for the present explanatory approach to
truth. The instrumentalist’s argument is described by Fine, in the paper from which I quoted above.
Fine notes that

the realist must at least allow that, generally speaking, truth does lead to reliability. For
instance, ... the realist offers the truth of a theoretical story in order to explain its success at a
certain range of tasks. If this offering is any good at all, the realist must then allow for some
intermediate connection between the truth of the theory and success in its practice. The
intermediary here is precisely the pragmatist’s reliability. Let us therefore replace the realist’s
‘truth’ with the pragmatic conception, framed appropriately in terms of reliability. Then, if the
realist has given a good explanatory account to begin with, we get from this pragmatic
substitution a good instrumentalist account of the same phenomena. Indeed, the
instrumentalist account we obtain in this way is precisely the intermediate account that the
realist has himself been using. Since no further work is done by ascending ... to the realist’s
‘truth’, the instrumental explanation has to be counted as better than the realist one. (1986, p.
154)

The instrumentalist thus accuses the realist of a redundant metatheoretical commitment.

An analogous argument can be drawn from our account of the role of truth in encouraging argument.
We saw in chapter 7 that if one begins with a conception of truth as a property of belief, and asks
what the point of argument could be – why we should care whether our beliefs have this property –
then at first one may think of truth as an end in itself. It becomes apparent, however, that truth could
at best be an intermediate goal in one’s quest for something like behavioural success. On the other
hand, we have also seen that progress towards behavioural success seems to require, or at least to be
enhanced by, our all behaving as if there were such an end as truth to judgement and enquiry. If there
were no real property of truth, we would be well advised to invent one. In other words, the real or
substantial notion of truth now looks redundant – unnecessary either to account for the utility of
argument, or to explain our use of a notion of truth.

This parallel is hardly surprising. After all, the redundancy argument is the hallmark of the claim to
conceptual economy, itself characteristic of the general philosophical movement to which
instrumentalism and the present theory both belong. Such arguments involve some characteristic
risks, however. As the quasi-realist’s problems illustrate, they tend to backfire, undermining the very

119
positions their advocates wish to preserve. Economy needs to be backed up by stability: the thin
residue must be sturdy enough to stand on its own.

Fine argues, I think rightly, that instrumentalism cannot stand on its own. As he puts it,
‘instrumentalism avoids the inflationary metaphysics of realism only by pumping up either
epistemology or semantics’ (1986, p. 170). For the semantic case, his argument is the objection to a
projectivist theory of truth we noted earlier. I think that like ours, his basic objection to
instrumentalism is that it involves an untenable distinction between ‘genuine’ and ‘instrumental’ areas
of factual discourse.

9.6 ... AND NOT QUIETISM

In the face of these objections, however – on the one side to realism, and on the other to
instrumentalism – Fine recommends that we be satisfied with what he calls ‘NOA ..., the natural
ontological attitude’ (1986, p. 172). He says that

NOA allies itself with what Blackburn ... dismissively calls ‘quietism’. Less pejoratively,
NOA holds a ‘no-theory’ conception of truth .... NOA does not think that truth is an
explanatory concept, or that there is some general thing that makes truths true. (1986, p. 175)

Fine also links NOA to Rorty’s recent views on truth (on which we remarked in chapter 6), saying
that Rorty’s ‘developing neo-pragmatism ... seems quite NOAish’ (1986, p. 173, n. 16).

My use of the term ‘quietism’ has not been exactly Blackburn’s. I used it initially for the view that
declines to theorize about the fact-stating/non-fact-stating distinction, rather than for the analogous
view about truth. But clearly the two views run parallel, and my sympathies are with Blackburn.

NOA is right, I think, in rejecting attempts to analyse truth. And it is right to dismiss recent attempts
to make truth itself an explanatory notion. (We considered these attempts in chapter 6. I argued that
far from supporting a realist or correspondence notion of truth, they depend on no more than the
disquotational or redundancy property.) NOA is wrong, however, to conclude that there is no further
role for a theory of truth. At the very least there remains the not inconsiderable task of accounting for
the fact that we have such a notion at all.

The quietist is simply blind to this, a blindness perhaps explained by the tendency on all sides in these
discussions to ignore non-descriptive uses of language. From this perspective we are liable to
overlook the contrasts in language that draw attention to the remaining explanatory task. To notice the
non-indicatives, and the issues raised by the non-factualists, is to notice not only that truth is not a
universal notion in ordinary language, but also that there is considerable doubt as to the limits of its
application. At this point, I think, to decline to theorize about truth would be an attitude to ontology
with all the naturalness of the ostrich’s famed response to challenges of a different kind.

I have urged that the most profitable approach to this explanatory task involves a conceptual retreat.
We need to step backwards, asking not why truth matters but why it matters that ordinary speakers
should think that it matters. We need to enquire what purpose could be served by our ordinary
conviction that there is an end to enquiry. My account of the purpose of this conviction may not be the
most appropriate one, but I think the stance is right. We need to examine the game rather than the
goal. Concerning truth itself, in other words, I think there is more truth than usual in the pious
sentiments of Grantland Rice.

For when the One Great Scorer comes

120
 To write against your name,
He marks – not that you won or lost –
But how you played the game.

9.7 LANGUAGE AND MIND: EXPLANATORY PROSPECTS

I want to finish with some comments on the scope and potential of the general philosophical approach
exemplified by this functional theory of truth. In particular I want to draw attention to its relevance to
recent work on mind and language – to the philosophical end of cognitive science. Here its
significance lies in the fact that it is at the same time naturalistic and non-reductive. I think it thus
offers an important and undervalued alternative to recent work – a middle way for cognitive science,
between the evils of non-naturalism on the one hand and the intractability of reductivism on the other.
The middle way may not be the right way, but it needs to be recognized as one of the options.

The goal here is old enough: an understanding of the ordinary concepts of language and mind; of
‘truth’, ‘meaning ‘, ‘belief, and so on. Increasing scientific interest in the mind has merely added
weight to the view that this understanding should be sought in natural terms – in a manner compatible
with the fact that we ourselves are part and product of the natural world. It is hard to object to the
naturalistic constraint so stated.

So stated, however, it is neutral between reductive and explanatory accounts of the concepts of
language and mind. Recent writers have generally taken it to call for the former. The task has been
taken to be to give a naturalistic account of what we are talking about in talking of truth, belief, and
the rest. Hence the philosophical industry devoted to the naturalistic analysis of these notions – an
industry founded in large part on the view that the only alternative would be to give up these notions,
to regard them as superseded by scientific psychology and linguistics. True, opinions differ as to the
acceptability of that alternative. Not everyone agrees with Fodor that

if commonsense intentional psychology really were to collapse, that would be, beyond
comparison, the greatest intellectual catastrophe in the history of our species; if we’re wrong
about the mind, then that’s the wrongest we’ve ever been about anything .... We’ll be in deep,
deep trouble if we have to give it up. (1987, p. xii)

For a less cataclysmic assessment see Stich (1983), chapter 11 or Churchland (1984), chapter 2.5, for
example. All the same, there is a very widespread view that the fate of the folk notions rests on the
success or otherwise of the reductive enterprise.

The functional approach suggests a challenge to this view. We have seen that it offers an account of
the existence of the folk notions of truth and falsity, without seeking to analyse these notions in
naturalistic terms. Its naturalism is that of evolutionary biology: it explains why creatures like us
should have come to talk this way. But now why stop at truth? Why not look for parallel accounts of
the rest of folk psychology and linguistics?

What would such a theory of intentional psychology be like? With one main difference, I think it
might be very like that of Daniel Dennett, perhaps the most notable opponent of the view that the fate
of folk psychology hangs on the possibility of its reduction to a physical basis. Dennett has long been
concerned to save a place for the folk notions, given (as he accepts) that there is no place for them in
neurophysiology. Briefly, his view is that beliefs and desires are ‘idealized fictions in an action-
predicting, action-explaining calculus’ (1978, p. 30; see also 1978, chapter 1, and 1981). Clearly this
has the makings of an explanatory theory in our sense. As Dennett emphasizes, the practical
advantages of such a calculus in ordinary life provide the basis of an evolutionary explanation of the
fact that humans have developed it.

121
The difference is that as the quote suggests, Dennett presents himself as an instrumentalist about folk
psychology. On the present view, however, this reading of such a theory is unnecessary and
unfounded. A sharp realist/instrumentalist distinction would call for a sharp factual/non-factual
distinction; and that, we argued, is not to be had. This suggests that things are better than they seemed
for an account of Dennett’s kind. The explanatory perspective doesn’t call for ontological apologies –
or at least not such profuse ones, the difference between psychology and neuroscience being at most a
matter of degree.

Stripped of its instrumentalism, Dennett’s view thus seems very much in keeping with our general
explanatory approach. However, it may then seem too good to be true. If it isn’t instrumentalist, if it
doesn’t deny that there really are beliefs and desires, then how does it evade the strong intuition that
there ought to be a physical account of what such things are? How in other words does it distinguish
itself from reductivism?

The answer will be in two parts. The first step is to defuse the physicalist intuitions – to appeal to
factors in virtue of which it seems implausible that ascriptions of psychological attitudes could be
equated with ascriptions of physical states. (Not surprisingly the most comprehensive accounts of
these factors are those of the opponents of realism about folk psychology; see, particularly, that of
Stich 1983.) The second step is then to deny that the first calls for an anti-realist view of folk-
psychological attitudes. The case here draws on the argument of this book, in a way that I hope will
by now be familiar. Briefly, the anti-realist construal rests on an application of an untenable fact/non-
fact distinction. All areas of discourse are potentially non-factual to some degree, but the factors
responsible may vary from case to case. The peculiarities of folk-psychological discourse seem to
distinguish it in this respect from talk about neurophysiology, with the result that we shouldn’t expect
a reduction. But only a philosophical misconception about truth could support the conclusion the anti-
realist wants to draw from this: namely, that only the claims of neurophysiology are genuinely factual,
truth-bearing, or whatever.

This approach to the problem of folk psychology in cognitive science thus applies the explanatory
approach at two levels. It applies it first to the folk notions themselves, recommending an explanatory
treatment in the same evolutionary vein as our functional theory of truth. The result, plausibly, is
Dennett’s account of the origins of intentionality. The second application then concerns the
interpretation of this theory of psychology. The issue is the factual status of folk-psychological
claims, and here an explanatory view of truth affords no basis for a sharp distinction. It was our
disenchantment with the possibility of such a distinction that led us to the explanatory approach.

I don’t want to sound too sanguine about the prospects for explanatory accounts of language and
mind. Nor do I want to suggest that intuitions about ontological and theoretical priority can simply be
dismissed out of hand. It would be much more in keeping with the present approach to try to explain
these intuitions and the phenomena of theory change – to explain them in terms of our general
conception of the function of factual discourse. Clearly usage is not immutable in the face of such
considerations, and it would be nice to be able to explain why it changes when it does change.

For the moment, however, I want simply to urge some critical attention to the terms in which the
current debate about the status of folk psychology is being phrased. The questions as to whether there
are ‘really’ beliefs and desires, as to whether ordinary ascriptions of such attitudes have genuine truth
values, are themselves phrased in folk-linguistic terms. Truth, fact and reality are folk notions. Their
status is as much at issue as that of the folk-psychological notions – to which, evidently, they are
closely related. Hence it is a mistake, I think, to expect or assume different kinds of answers in the
psychological and linguistic cases. In particular it is a mistake to allow an unexamined philosophical
conception of truth to govern one’s interpretation of the issue about the status of folk psychology.
There is a kind of consistency constraint here. If we’re physicalists about belief then we ought to
interpret the question of the truth status of belief ascriptions in terms of a physicalist notion of truth.

122
But if we take the explanatory approach in both cases then we deserve to be judged by the standards
of that approach.

It thus turns out to have been important to treat truth first. To accept an explanatory account of truth is
to constrain one’s understanding of any other theory in the same general spirit. Explanatory theories
have in the past been crippled by their failure to apply their insights to truth itself (think of the
problems of projectivism, for example). The promise of the explanatory approach lies in considerable
measure in its ability to explain the terms in which the traditional issues have been phrased.

123
REFERENCES

Adams, E. 1965. The logic of conditionals. Inquiry, 8, 166–97.

—1975. The Logic of Conditionals, Dordrecht: Reidel.

Albury, R. and Slezak, P. (eds) 1988. Computers Brains and Minds: Papers in Cognitive Science,
Dordrecht: Kluwer Academic Publishers.

Austin, J. L. 1962. How to Do Things with Words, Oxford: Oxford University Press.

Bach, K. 1975. Performatives are statements too. Philosophical Studies, 28, 229–36.

Bar-Hillel, Y. (ed.) 1966. Proceedings of the 1964 International Congress for Logic, Methodology
and Philosophy of Science, Amsterdam: North Holland.

Blackburn, S. W. 1971. Moral realism. Casey (ed.) 1971, 101–24.

—1984. Spreadingthe Word,Oxford: Oxford University Press.

—1985. Supervenience revisited. Hacking (ed.) 1985, 47–67.

Bouveresse, J. and Parret, H. (eds) 1981. Meaning and Understanding, Berlin: de Gruyter.

Braithwaite, R. B. 1953. Scientific Explanation, Cambridge: Cambridge University Press.

—1966. Why is it reasonable to base a betting rate upon an estimate of chance? Bar-Hillel (ed.) 1966,
263–73.

Carlstrom, I. and Hill, C. 1978. Review of Ernest Adams, The Logic of Conditionals [Adams 1975]
Philosophy of Science, 45, 155—8.

Casey, J. (ed.) 1971. Morality and Moral Reasoning, London: Methuen.

Churchland, P. 1984. Matter and Consciousness, Cambridge, Mass.: MIT Press.

Davidson, D. 1968. On saying that. Synthese 19, 130–46. Reprinted in Davidson 1984, 93–108.

—1974. Belief and the basis of meaning. Synthese, 27, 309–24. Reprinted in Davidson 1984, 141–54.

—1984. Inquiries into Truth and Interpretation, Oxford: Oxford University Press.

—and Harman, G. (eds) 1972. Semantics of Natural Language, Dordrecht: Reidel.

Dennett, D. 1978. Brainstorms, Cambridge, Mass.: MIT Press.

124
—1981. Three kinds of intentional psychology. Healey (ed.) 1981, 37–61.

Devitt, M. 1984. Realism and Truth, Oxford: Basil Blackwell.

Dummett, M. A. E. 1958. Truth. Proceedings of the Aristotelian Society, 59, 141–62.

—1973. Frege: Philosophy of Language, London: Duckworth.

—1978. Truth and Other Enigmas, London: Duckworth.

Ellis, B. 1980. Truth as a mode of evaluation. Pacific Philosophical Quarterly, 61, 85–99.

Evans, G. and McDowell, J. (eds) 1976. Truth and Meaning: Essays in Semantics, Oxford: Oxford
University Press.

Field, H. 1972. Tarski’s theory of truth. Journal of Philosophy, 69, 347–75. Reprinted in Platts (ed.)
1980, 83–110.

—1978. Mental representation. Erkenntnis, 13, 9–16.

Fine, A. 1986. Unnatural attitudes: realist and instrumentalist attachments to science. Mind, 95, 149–
79.

Fodor, J. 1985. Fodor’s guide to mental representation. Mind, 94, 76–100.

—1987. Psychosemantics: The Problem of Meaning in the Philosophy of Mind, Cambridge, Mass.:
MIT Press.

Grover, D., Camp, J. and Belnap, N. 1975. A prosentential theory of truth. Philosophical Studies, 27,
73–125.

Habermas, J. 1970. Towards a theory of communicative competence. Inquiry, 13, 360–75.

Hacking, I. (ed.) 1985. Exercises in Analysis: Essays by Students of Casimir Lewy, Cambridge:
Cambridge University Press.

Hamblin, C. L. 1987. Imperatives, Oxford: Basil Blackwell.

Harman, G. 1986. Change in View: Principles of Reasoning, Cambridge, Mass.: MIT Press.

Healey, R. (ed.) 1981. Reduction, Time and Reality, Cambridge: Cambridge University Press.

Hesse, M. 1978. Habermas’ consensus theory of truth. Proceeding of the Philosophy of Science
Association 1978, 2, 373–96.

Hume, D. 1978. Treatise of Human Nature, ed. L. A. Selby-Bigge, 2nd edn with revisions by P. H.
Nidditch, Oxford: Oxford University Press.

Jackson, F. 1984. Two theories of indicative conditionals: reply to Brian Ellis. Australasian Journal
of Philosophy, 62, 67–76.

—1987. Conditionals, Oxford: Basil Blackwell.

125
Kripke, S. 1982. Wittgenstein on Rules and Private Language, Oxford: Basil Blackwell.

Leibniz, G. W. 1951. Theodicy: Essays on the Goodness of God, the Freedom of Man and the Origin
of Evil, ed. A. Farrer, London: Routledge & Kegan Paul.

Lewis, D. 1972. General semantics. Davidson and Harman (eds) 1972, 169–218. Originally published
in Synthese, 22 (1970), 18–67.

—1976. Probabilities of conditionals and conditional probabilities. Philosophical Review, 85, 297–
315.

Mackie, J.L. 1977. Ethics: Inventing Right and Wrong, Harmondsworth: Penguin.

McDowell, J. 1978. Are moral requirements hypothetical imperatives? Proceedings of the Aristotelian
Society, Supplementary Volume, 52, 13–29.

—1981. Anti-realism and the epistemology of understanding. Bouveresse and Parret (eds) 1981, 225–
48.

McGinn, C. 1977. Semantics for nonindicative sentences. Philosophical Studies, 32, 301–11.

—1982. The structure of content. Woodfield (ed.) 1982, 207–58.

—1984. Wittgenstein on Meaning, Oxford: Basil Blackwell.

Mellor, D. H. 1971. The Matter of Chance, Cambridge: Cambridge University Press.

—1988. How much of the mind is a computer? forthcoming in Albury and Slezak (eds) 1988.

Olshewsky, T. M. (ed.) 1969. Problems in the Philosophy of Language, New York: Holt, Rinehart &
Winston.

Palmer, F. R. 1986. Mood and Modality, Cambridge: Cambridge University Press.

Parkinson, G. (ed.) 1982. Marx and Marxisms, Cambridge: Cambridge University Press.

Pettit, P. 1982. Habermas on truth and justice. Parkinson (ed.) 1982, 207–28.

Platts, M. (ed.) 1980. Reference, Truth and Reality, London: Routledge & Kegan Paul.

Price, H. 1981. The Problem of the Single Case. Ph.D. thesis, Cambridge University.

—1983a. Sense, assertion, Dummett and denial. Mind, 92, 174–88.

—1983b. Does ‘Probably’ modify sense? Australasian Journal of Philosophy, 61, 396–408.

—1986. Conditional credence. Mind, 95, 18–36.

—1987. Truth and the nature of assertion. Mind, 96, 202–20.

Putnam, H. 1978. Meaning and the Moral Sciences, London: Routledge & Kegan Paul.

126
Quine, W. V. 0. 1960. Word and Object, Cambridge, Mass.: MIT Press.

—1970. Philosophy of Logic, Englewood Cliffs, NJ: Prentice-Hall.

Ramsey, F. P. 1978. Foundations, ed. D. H. Mellor, London: Routledge & Kegan Paul.

Rorty, R. 1982. Consequences of Pragmatism: Essays 1972–1980, Brighton: Harvester.

Schiffer, S. R. 1972. Meaning, Oxford: Oxford University Press.

Stalnaker, R. 1970. Probability and conditionals. Philosophy of Science, 37, 64–80.

Stenius, E. 1967. Mood and language-game. Synthese, 17, 254–74.

Stich, S. 1983. From Folk Psychology to Cognitive Science: The Case Against Belief, Cambridge,
Mass.: MIT Press.

Strawson, P. F. 1949. Truth. Analysis, 9. Reprinted in Olshewsky (ed.) 1969, 610–23.

—1950.Truth. Proceedings of the Aristotelian Society, Supplementary Volume, 24, 129–56.

van Straaten, Z. (ed.) 1980. Philosophical Subjects: Essays Presented to P. F. Strawson, Oxford:
Oxford University Press.

Wiggins, D. 1980. What would be a substantial theory of truth? van Straaten (ed.) 1980, 189–221.

Williams, B. A. 0. 1973. Problems of the Self, Cambridge: Cambridge University Press.

Woodfield, A. (ed.) 1982. Thought and Object, Oxford: Oxford University Press.

127
INDEX of 1st Edition

Adams, E., 85, 99

Adams’s hypothesis, 85, 99–112, 185

analytic approach see assertion, explanatory approach, marker view, non-factualism

anti-realism, 1
see also non-factualism

approval problem, 76, 82

see also confidence problem

argument
advantages over agreement per se, 156–7
behavioural utility of, 15, 134–5, 145, 151–5, 156–7
role of truth in, 15, 134–5, 153–5, 180–1
see also disagreement, dissent

assertion
belief-based analysis of, 14, 41–2, 63–4, 69, 75, 85–6, 99
primacy of, 7, 203
supposed characteristics of, 43–4, 57–9
truth-based analysis of, 13, 19–40, 93
see also non-factualism, non-indicatives

Austin, J. L., 50, 55

Bach, K., 52

Blackburn, S., 67, 76, 77–82, 127, 209, 210, 213–14

Braithwaite, R.B., 72

Carlstrom, I., 102, 107–11, 112–13

Chomsky, N., 193

Churchland, P., 216

commands see assertion, marker view, non-indicatives

conditionals, 99–114, 185–9

subjunctive, 188–9
see also Adams’s hypothesis, factualism, non-factualism

confidence problem, 72, 74, 82

see also approval problem

Davidson, D., 8, 9, 33, 35, 36, 37

Dennett, D., 216–18

128
desire
catalytic view of, 90
in cognitive theory, 94–9
Humean viewof, 87–94

Devitt, M., 138–42

direction of fit, 181

disagreement
evaluative,173–80
evaporative, 161, 195–8
evidential, 158, 173
inferential, 158, 173
probabilistic, 159–60, 162–7
substantial, 161
see also argument, dissent

dissent
and negation, 149
triggersof, 146–9
see also argument, disagreement, truth

Dummett, M., 21, 29, 30, 31–2, 43–5, 57, 131, 193

Ellis, B., 155–7

emotivism, 1, 6
see also moral judgement, non-factualism

ethics
see moral judgement, non-factualism

Evans, G., 8, 36

explanatory approach, 14–16, 118–19

contrasted to analytic approach, 4, 118–19
evolutionary character of, 15, 125–30, 216
to folk psychology, 215–18
to linguistic distinctions, 119–20, 122–30
naturalistic character of, 4, 15, 215
prospects for, 215–18
to social class, 120–2
see also non-factualism, truth

explicit performatives, 44, 49, 50–3

factualism
about conditionals, 99–114
about meaning, 11
about moral judgement, 64–5, 76–7
about probability, 63–5, 72–5

129
 radical, 75, 77, 82, 84, 88, 99
universal, 43, 44, 49, 53, 56, 59

factuality see assertion, explanatory approach, factualism, non-factualism

fatalism, 103–7

Field, H., 137–44

Fine, A., 208–9, 212–14

Fodor,J., 94–5, 98, 216

folk psychology, 16, 94, 216–18

Fregean general semantics

affinity with marker view, 12
conflict with non-factualism, 3, 7–11, 20, 202–4
primacy of assertion in, 7–10
sense /force distinction in, 9, 34–5
treatment of non-indicatives, 8–9, 12
use of truth in, 7– 8

functional realism, 211

naturalism of, 215–16
not idealism, 210–11
not instrumentalism, 211–13
not projectivism, 210
not quietism, 214
see also explanatory approach

Habermas, J., 157

Hamblin, C., 53

Harman, G., 163–5

Hesse, M., 157

Hill, C. see Carlstrom, I.

Hume, D., 6–7, 67, 76, 86, 87–94

IDA (imagined dislike argument), 88–9, 90

imperatives see assertion, marker view, non-indicatives

indexicality
and conditionals, 107
and probability, 106–7, 167–9

instrumentalism, 211–13, 215–16

130
interrogatives see assertion, marker view, non-indicatives

Jackson, F., 112–13, 189

Kripke, S., 11

language of thought, 95–7, 128

Leibniz, G., 104

Lewis, D., 44, 50, 52, 102– 7, 111–13

McDowell,J., 8, 33–6, 38, 40, 91

McGinn, C., 51, 98, 138, 140–4, 193

Mackie, J. L., 65–6

marker view
affinity with Fregean semantics, 12
analytic presuppositions of, 12
characterization of, 11
recognitional requirement, 28 sceptical argument against, 43–59
truth-based version of, 19, 22, 28

meaning see Fregean general semantics, radical interpretation, truth

Mellor, D. H., 73, 96, 104

moral judgement
disagreement about, 173–80
functional role of, 16, 76–7
regulative role of, 16, 179
supervenience of, 77–82
see also disagreement, factualism, non-factualism

negation
and dissent, 149
as a mood indicator, 23

NOA (natural ontological attitude ), 213–14

non-cognitivism, 1
see also non-factualism

non-factualism
about aesthetics, 173–80
about causality, 196
about conditionals, 85, 99–114, 185–9
about everything, 191–5
about folk psychology, 217–18
about meaning, 11
about modality, 196

131
 about morality, 76–7, 173–4
about presupposition, 190–1
about probability, 63–4, 159–60, 162–7, 177–8
about secondary qualities, 196
about theory, 197
about truth, 180–1, 208–9
analytic version of, 2
characterization of, 1–2
conflict with Fregean semantics, 3, 7–11, 20
explanatory approach to, 136, 161–2, 172, 184
history of, 6–7
sources of, 195–8
and universal factualism, 61–2

non-indicatives
deception by means of, 44, 45–7
indicative paraphrases of, 49, 54–6
non-applicability of truth to, 135
significance of, 199–201
see also marker view

normativity, 131–4
see also moral judgement, truth

Northern Line, 123

Palmer, F., 11–12

paratactic analysis, 35

Pettit, P., 157

presupposition, 190–1

Price, H., 37, 63, 101

probability
disagreement about, 159–60, 162–7, 177–8
functional role of, 70–2, 165–6
and non-final judgement, 165
and rationality, 72–4
supervenience of, 82–4
see also disagreement, factualism, non-factualism

projectivism, 67–8, 79, 181–3, 208–9, 210

and idealism, 68–9, 79, 209–10

Putnam, H., 138, 144

quasi-realism, 210, 213

queerness, argument from, 65–6

132
questions see assertion, marker view, non-indicatives

quietism, 14, 133, 213–14

Quine, W. V. 0., 26, 193, 194–5

rabbit, 163

radical interpretation, 32–3, 37, 39, 202–3

empirical grounds for, 37–8

Ramsey, F. P., 26, 85, 139

rat see rabbit

realism see factualism, functional realism, non-factualism

relativism, 173–80, 183

Rice, G., 215

Rorty, R., 133, 214

RTM (representational theory of mind), 95–8

SBP (same boat property), 152–3, 159, 160, 167, 177, 178, 182–3, 185

Schiffer, S., 51–3

SEA (simple ethical ascription), 64

see also factualism, moral judgement, non-factualism

SPA (simple probability ascription), 63

see also factualism, non-factualism, probability

statement of fact see assertion, explanatory approach, factualism, non-factualism

statementhood, 2
see also assertion, explanatory approach, factualism, non-factualism

Stenius, E., 22–4, 28, 29, 30

Stich, S., 96–7, 216, 217

STM (syntactic theory of mind), 96–7

Strawson, P. F., 26, 31, 39, 130, 137

subjunctives see conditionals

supervenience see moral judgement, probability

133
Tarskian truth theory, 32, 137
see also Fregean general semantics, truth

truth
amen theory of, 26, 39, 130–1, 137
disquotational view of, 26, 137–44, 203
equivalence property of, 26, 39, 205–6
and evaporative argument, 170–1
explanation by means of, 137–4
explanatory approach to, 4, 15–16, 119–20, 130–1, 137, 205–7
functional role of, 4, 15–16, 130–1, 132, 135–6, 145, 205–7
marks of, 20, 40–1, 202, 206–8
normativity of, 131, 203, 205–6
redundancy view of, 26
role in argument of, 134–5, 153–5, 180–1
substantial theory of, 21
truth-like nature of, 205–7
see also explanatory approach, non-factualism

universal factualism see factualism, universal

Wiggins, D., 19–20, 31, 32, 40–1, 206–7

Williams, B., 22, 26, 31

Wittgenstein, L., 11, 29, 193

134