Beyond The Limits of Thought (G. Priest)
Beyond The Limits of Thought (G. Priest)
Beyond The Limits of Thought (G. Priest)
of Thought
GRAHAM PRIEST
Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street, Cambridge CB2 I RP
40 West 20th Street, New York, NY 1 00 1 1 -42 1 1 , USA
1 0 Stamford Road, Oakleigh, Melbourne 3 1 66, Australia
A catalogue record for this book is available from the British Library
page
Preface xiv
Introduction 1
5 The continuum 31
6 Infinite Qarts 33
7 Aguinas' cosmological argument 35
8 Leibniz' repair 38
9 The QrinciQle of sufficient reason 40
CQP!<lusiQP 41
Introduction 43
Varieties of skeQticism 44
2 Sextus' argument for skeQticism 46
3 AnalYl!is of the argument 47
4 SkeEticism and self-reference 49
5 Protasorean relativism 52
6 The argument for relativism 53
7 Socrates' attack �
8 Nothing is true 56
9 C08!!ition and �radox 58
Conclusion 52
Introdu!<tiQO 61
Anselm's ontoloi!cal argument 61
2 The inconceivabilit� of God 63
3 The characterisation principle 64
4 Berkclefs master argument for idealism 65
5 Anal�sis. stage I 66
6 Anal�sis. sta� II 69
7 Berkelefs response 71
8 Some objections 74
9 Berkel�'s �radox 76
CQDI;lusiQP 12
Ip![()dllcljop 81
I PhenQlnena and noumepa 82
2 The categories of judgment 83
3 The aEplicability of the categories 85
4 Th e law Q f callS3liQP 81
5 The contradictory nature of noumena 88
6 Analogy 91
Conclusion 23
Conlents xi
Introduction 94
The transcendental musion 94
2 The antinomics' their abstract slructu!'C 96
3 First antinomy : the beginning of the cosmos 97
4 Second antinomy: the di v isibili ty of matter 99
5 Third an tinomy : causal chains 101
6 Fourth antinomy : a necessary being 103
7 Kant's solution(s) 106
8 Evaluation 109
9 Fifth antinomy: the limit of thought 110
Conclusion 112
7 Hegel's infinities 11 3
Introduction 113
Hegel's critique of Kant 113
2 Contradiction in the world 114
3 Hegel's dialectic 115
4 The false infinite 116
5 The true jnfinilc 119
Co ncl usjon 121
9 Vicious circles 14 1
Introduction 141
Russell's paradox 141
2 Russell's schema 142
3 Definability 144
4 The inclosure schema 147
S More on the inclosure schema 148
6 Russell's VCP and the theory of orders 149
7 Systematic ambiguity 151
8 The solution that reproduces the problem 154
Conclusion 154
xii Contents
\0 Parameterisation 155
Introduction ISS
Ramse:{s two families ISS
2 Group Bii 157
3 Inclosures and the limits of expression 161
4 Groul! B: Ramse:{s solution 162
5 . . . and its inad egua� 164
6 Parameterisation in general 166
7 . . . and its inadegua� 168
CQDI:IIISiQD 111
Bihliographr ill
Index 2M
Preface
This book had its origin in a series of seminars on the limits of thought at
the University of Western Australia in 1988, given by Uwe Petersen and
myself. For some years Uwe and I had been arguing the relative merits of
Kant and Hegel, and we had both observed striking similarities between
Kant's Antinomies and the paradoxes of self-reference. These themes
were explored in the seminars. The result of the discussions was my
paper 'The Limit of Thought - and Beyond' (Mind 100 (1991), 361-
70). The present book is essentially a much expanded form of that paper.
In the next couple of years I realised that the issues involved have a
much larger import than at first I had realised, indeed, that they stretch
through the whole history of Western Philosophy. In 1990 I had a year's
leave, and spent it investigating these matters and drafting the book. I am
greatly indebted to Cambridge University for providing a most congenial
and stimulating context for the research. I would like to thank Clare Hall
for providing me with a Visiting Fellowship for the year; and I would
particularly like to thank the Sub-Faculty of Philosophy, who not only
provided me with ideal working conditions (a room in the Moral Sciences
Library, somewhere to make tea. and office staff who were always
friendly and helpful well beyond the call of duty), but also allowed me
the opportunity of airing some of the material in a series of lectures in
Michaelmas term. Many parts of the book have been served up to
unsuspecting philosophers at numerous other universities in Britain,
Australia, and North America.
Every philosopher is indebted to colleagues for help and criticism. In
writing this book I am particularly so. I do not pretend to be an historian
of philosophy, and on many occasions I have been grateful to colleagues
for helping me understand historical material and preventing me from
making blunders. I hope that those I have forgotten will forgive me if I
list those whom I have not: James Allen. Myles Burnyeat, Stewart
Candlish, Edward Craig. Jan Crosthwaite, Nick Denyer, Lawrence
Goldstein, John Frow. Kevin Hart, Mark Lance, David Lewis, Uwe
Petersen, Mark Sainsbury. I am also indebted to my colleagues in the
Archytas, according to Euthydemus, put the question in this way: if I
came to be at the edge, for example at the heaven of the fixed stars,
could I stretch my hand or my stick outside, or not? That I should not
stretch it out would be absurd , but if I do stretch it out, what is outside
will be either body or place - (it will make no difference as we shall
discover) . Thus, Archytas will always go on in the same way to the fresh
chosen limit, and will ask the same question. If it is always something
different into which the stick is stretched, it will clearly be something
infinite .
Simpiicius, Physics
Introduction
1 Peter Slezak has suggested to me that there may be an important connection between the
phenomena that will concern us and subjective consciousness too. See Slezak ( 1 983).
4 Introduction
limits of thought are boundaries which cannot be crossed, but yet which
are crossed.
In each of the cases, there is a totality (of all things expressible, describ
able, etc.) and an appropriate operation that generates an object that is
both within and without the totality. I will call these situations Closure
and Transcendence, respectively. In general, the arguments both for
Closure and for Transcendence use some form of self-reference, a
method that is both venerable and powerfu1. 2 Closure is usually estab
lished by reflecting on the conceptual practice in question. In a polemical
context this can appear as an ad hominem argument. Arguments for
Transcendence are of more varied kinds; often they involve applying a
theory to itself. Some of them are more technical; a paradigm of these is
diagonalisation, a technique familiar from the logical paradoxes. This
construction is precisely a boundary-tearing heuristic which, given any
boundary of a suitable kind, can be applied to violate it.
2 Dialetheism
That a contradiction might be true, or that dialetheism (the view that
there are true contradictions) makes sense, may still be abhorrent, and
even threatening, to many contemporary English-speaking philosophers.
More likely than not, even the suggestion of it will be met with a look of
blank incomprehension. How could a contradiction be true? After all,
orthodox logic assures us that for every statement, 0: , only one of 0: and
-'0: is true. The simple answer is that orthodox logic, however well
2 Arguably, it is more than that: it characterises philosophy itself. See Priest ( 1 99 I a).
3 Thus, when Quine « 1 970), p. 8 1 ) complains that someone who maintains that a contra
diction might be true just does not know what they are talking about, this is exactly
analogous to a Newtonian physicist complaining that someone who holds that time
might run at different rates in different frames of reference just does not know what
they are talking about. For further discussion, see Priest ( 1 987), ch . 1 4 .
Beyond the limit 5
of the corresponding four subsets of the set {true, false} . How do the
truth (and falsity) conditions for, for example, negation and conjunction
now work? Exactly as you would expect. ,a is true iff a is false, and vice
versa. Similarly, a /\ j3 is true iff both conjuncts are true, and false if at
least one conjunct is false. In particular, if a is both true and false, so is
,a, and so is a /\ ,a. Hence, a contradiction can be true (if false as well) . 4
Since a logical theory of this kind is so easy to construct, it is worth
asking why dialetheism should be so outrageous to the sensibility of
modern philosophers. The answer is, I am afraid, sociological rather
than rational. Until Aristotle, the idea that a contradiction might be
true was a highly contested one. 5 Aristotle went to great lengths to
argue against the idea in book 3 of the Metaphysics.6 Because of
Aristotle's magisterial authority in the Middle Ages, the subject became
closed; and, unlike most other subjects on which Aristotle pronounced,
was not substantially reopened subsequently - at least until recently. Few
philosophers (with the notable exception of Hegel and some of his intel
lectual descendants) challenged Aristotle. And I know of no historical
defence of the law of non-contradiction since Aristotle worth mentioning.
Is this because Aristotle's arguments were conclusive? Hardly. Arguably,
they do not work at all. But even if they were ultimately correct, they
could not close debate on the issue: his major argument is so tortured that
it is hard to tell what it is, and the others are little more than throw-away
remarks. 7
So what other arguments are there? Apart from an appeal to the
received logical theory, which we have already seen simply to beg the
question, the most common is that a contradiction cannot be true since
contradictions entail everything, and not everything is true. This argu
ment is no better than an appeal to the received semantics. Granted, not
everything is true; but on the alternative semantics sketched above, as
should be clear, the argument from a /\ ,a to an arbitrary j3 need not
preserve truth, and so is not valid. 8 More subtle arguments are to the
4 Semantics of this kind can easily be made rigorous. Details can be found in section I of
the technical appendix to Part 3 of the book. There, I ignore the neither case, for reasons
given in Priest ( 1 987), 4.7.
5 See, for example, Dancy ( 1 975), ch. 3.
6 In the same book, he argues against the possibility of a statement being neither true nor
false, though in other places, for example, ch. 9 of De Interpretatione, he appears to show
a good deal more sympathy with that idea.
7 The arguments are described, analysed and rejected in Lukasiewicz ( 1 97 1 ) and Dancy
( 1 975).
R Similarly, the form of inference usually called the disjunctive syllogism (ex, -- ex V f3 f-- (3)
is invalid. Let ex be both true and false, and f3 just plain false. Then the premises of the
syllogism are true (since a disjunction is true iff one disjunct is) but the conclusion is not.
6 Introduction
effect that some crucial notion or other, for example, truth, validity, or
rationality, requires consistency. I shall not take up these issues here. All
that it is necessary to note is that these arguments, whatever they are, are
neither obvious nor obviously correct. It is thus foolish to close one's
mind to the possibility of dialetheism.
I certainly do not intend to imply that these arguments should not be
discussed: they raise profound issues at the very heart of logic. I have
discussed them in In Contradiction ( 1 987), which contains a defence of
dialetheism, and to which I would refer readers for further details. In a
sense, In Contradiction starts where Part 3 of this book leaves off: with
dialetheism and the paradoxes of self-reference. Perhaps more impor
tantly, the current investigations are an application of the main thesis
of that book to one particular area, a more detailed investigation of one
region of the transconsistent.
9 I suspect that the same is true of Eastern philosophy too, though my knowledge of this is
very limited.
10
This requires a good deal of quoting. All italics in quotations are original unless other
wise specified. In accordance with Cambridge University Press house style, any ellipses at
the beginning or end of a quotation have been omitted.
Beyond the limit 7
scales in my favour, I ask that this be forgiven in the context. The fact
that all of the interpretations fit neatly into a grander scheme says some
thing in favour of each; and I hope that my interpretations of some of the
philosophers involved are interesting in their own right.
style may well cloak this fact from the casual reader (if any reader of
Hegel could be described as casual). The chapter on Hegel is therefore the
lynch-pin of the book. Like the vertex of a light cone, it focuses all the
rays of thought that come before, as well as all those that come after.
Introduction
Four limits of thought in particular will concern us in the book: the limits
of the expressible, the limits of the iterable (the mathematical infinite), the
limits of cognition, and the limits of conception. There is, as we shall see
in due course, a close connection between these things. However, no one
before Kant perceived this clearly. In the first part of the book we will
look at ways that these various limits arose in pre-Kantian philosophy.
This will involve us in looking at a number of topics that may appear,
initially, to have little to do with each other. As I said in the Introduction
of the book, if this is so, patience will, I hope, be rewarded.
The first limit we will be concerned with here is the limit of the expres
sible, which is, perhaps, the most fundamental of the limits that will
occupy us in the book. One might suppose, for a number of reasons,
that features of the (non-linguistic) world are such as to transcend the
ability of language to express them. If this is indeed so, then attempts to
say what these are, are liable to end up in contradiction. For in saying
what those features are, one is liable to say the unsayable.
In this chapter we will look at three reasons for supposing there to be
features of the world that transcend language. Good candidates for
features of the world that are liable to fall into this category are ones
that might be thought to constitute ultimate reality, in some sense; for
example, fundamental matter or God. We will look at an account of each
of these two things that places it in the category of the inexpressible:
Aristotle's account of prime matter, and Cusanus' account of God.
First, however, we will start with a much more radical reason for the
existence of the inexpressible; one that, if it were correct, would show that
the whole of reality is like this: Cratylus' view of the flux of meaning.
12 The limits o f thought i n pre-Kantian philosophy
THEODOR u s. Yes.
SOCRATES. With both the kinds of change we distinguished; both moving and
altering?
THEODORUS. Yes , certainly, they must be, if they're to be completely changing.
SOCRATES. Well now, if things were only moving, and not undergoing altera
tion, we'd be able to say, surely, that the moving things flow qualified in
such-and-such ways. Isn't that right?
THEODOR u s. Yes.
SOCRATES. Whereas since not even this stays constant, that the flowing thing
flows white [say] , but it changes, so that there's a flux of that very thing,
whiteness, and change to another colour, in order not to be convicted of
staying constant in that respect - since that's so, can it ever be possible to
refer to any colour in such a way as to be speaking of it rightly?
THEODORUS. How could it be, Socrates? Indeed, how could it be possible to do
so with any other thing of that kind, if it's always slipping away while one
was speaking; as it must be, given that it's in flux?
Note the phrase 'the flowing thing flows white' in Socrates' last speech;
we will return to this . The above argument is embedded in a discussion of
perception. The dialogue then goes on to illustrate the point by applying
it to this.
SOCRATES. And what shall we say about perception of any given kind, for
instance that of seeing or hearing? Shall we say it ever stays constant in
just that guise, namely, seeing or hearing?
THEODORUS. No, we mustn't, if everything changes.
as not a case of seeing, or as any o ther perception any more than as not that
perception; at any rate , we shouldn't do so if everything is changing in every
way .
THEODORUS. No.
soc RATE s. Then let us consider the absolute, not whether a particular face, or
something of that sort, is beautiful, or whether all these things are in flux. Is
not, in our opinion, absolute beauty always such as it is?
CRATYLUS. That is inevitable.
SOCRATES. Can we, then, if it is always passing away, correctly say that it is this,
SOCRATES. How, then, can that which is never in the same state be anything?
For if it is ever in the same state, then obviously at that time it is not
changing; and if it is always in the same state, and is always the same,
how can it ever change or move without relinquishing its own form.
CRATYLUS. It cannot do so at all .
SOCRATES. No, nor can it be known by anyone. For at the moment when he
city, v, given by the equation v = t, where t is time (and both v and t are
measured in suitable units). Then at t = 1 its velocity is instantaneously l .
Although it is changing, it does not follow that the velocity is not deter
minately 1 at that instant.
Possibly, Plato would have argued as follows. If something is in a state
for only an instant, then, since the instant has no duration, the thing is in
a state for no time, and hence not in that state. However, we now have
(arguably) a better understanding of the relationship between points and
continua than did Plato, and from this perspective we can see that the
subsidiary argument is itself fallacious since it confuses time with dura
tion: a state of affairs can hold at an instant whilst that instant itself has
no duration.
SOCRATES. Yes, Theodorus, except that I said 'so' and 'not so' . One oughtn't
even to use the word 'so ' , because what's so wouldn' t any longer be chan
ging; and, again, one oughtn't to use 'not so', because that isn't a change
either. No , those who state that theory must establish some other language ,
because as things are they haven' t got expressions for their hypothesis: unless
perhaps 'not even so ' , said in an indefinite sense, might suit them best.
THEODORUS. Yes, that would certainly be a more appropriate idiom.
Again, seeing that the whole of nature is in motion, and that nothing is true of
what is changing, [some] supposed that it is not possible to speak truly of what is
changing in absolutely all respects. For from this belief flowed the most extreme
opinion of those I have mentioned - that of those who say they 'Heraclitize' ,
and such a s was held b y Cratylus, who i n the end thought one should say nothing
and only moved a finger, and reproached Heraclitus for saying that you cannot
step into the same river twice - for he himself thought that you could not do so
even once.
If, however, Cratylus does express his views, as, presumably, he did, if
he expressed views about stepping into rivers, then he does seem to be in a
contradictory position. And even if he does not say anything, there still
seems to be a contradiction implicit in his position. For, after all, even to
come to the conclusion that one cannot speak, it is necessary to express
this thought to oneself, and how can one do this if there is no language
(not even a language of thought!) in which this can be expressed?
We meet here the inexpressible, and the contradiction to which it leads,
for the first time. By applying Cratylus' theory to itself it follows that it is
not in the domain of the expressible (Transcendence); but he does succeed
in expressing it, at least to himself (Closure).
Let us scrutinise the argument to inexpressibility again, however. We
saw in 1 . 1 that the extreme Heraclitean view that everything (including
the meanings of words) is in a state of flux is plausible if, but only if, we
take a sufficiently generous time scale. Now if the meanings of Cratylus'
words change very suddenly, say, before a hearer can hear and under
stand them, then I think it fair enough to say that he cannot express his
views. But words change their meanings only over relatively long periods
of time, in which case Cratylus can certainly express his views to a
localised audience (though maybe not in an eternal - though therefore
12
See Guthrie ( 1 9 8 1 ) , p. 227; Dancy ( 1 978), p. 404.
The limits of expression 19
Table I
Now, at last, we come to prime matter. This is just the matter at level l ,
that is, the substance at level o. (I have marked it with an X in Table I.
For here lies, as we shall see, the unknown.) It will be useful to separate
two distinct claims about the nature of prime matter.
The first is that prime matter (considered as the substance at level 0) is
the ultimate bearer of all properties. This is so since properties are
hereditary downwards. Thus, all the properties of Mrs T. (for example,
being in Downing St) are properties of the substance which composes her
(flesh and bones). And the properties of this flesh and bones (being in
Downing St, being constituted as a person) are properties of the sub
stances composing them (the elements); and so on. Let us call this:
Thesis 1: Prime matter is the ultimate subject.
The second claim is that prime matter, residing on the lowest level, is
not itself enformed matter of a lower level (which is why Table I has
blanks there). It therefore has no form, only accidents. Let us call this:
Thesis 2: Prime matter has no essence.
Or, as the medievals put it, prime matter is 'pure possibility'. It can be
anything; but is, in itself, nothing.
13 All quotations from Aristotle in this chapter are taken from Barnes ( 1 984).
20 The limits of thought in pre-Kantian philosophy
Which triggered perhaps the most scathing attack of all on the notion;
that by Berkeley. 1 5
This tradition concerning the unknowability of prime matter does not
get to the heart of the problem, however. Prime matter is unknowable, it
is true. But it is impossible to know what it is precisely because it is
impossible to say what it is: by Thesis 2, it has no form. Prime matter
is, therefore, precisely something beyond the limit of the expressible
(Transcendence). But we can quite happily give its essence; we have
already done so in describing it (Closure) . What makes something
prime matter, as opposed to some other kind of substance, is exactly
that it is the bearer of all properties, whilst it itself has no essence, as
Theses 1 and 2 have it. (Its essence, to put it bluntly, is to be essenceless.)
Thus prime matter exhibits the typical contradiction at the limit of the
expressible, though Aristotle, as far as I know, never observed this.
Let us turn to Thesis 2. Is there any reason to suppose that the ultimate
subject is formless? If the regress goes round in a loop then, as we have
just noted, there are ultimate subjects; and, since they are not bottom,
they have form. The quick answer, therefore, is 'no'. But, even supposing
the ultimate subject to be the substance of level 0, the answer is still 'no'.
We can hold the theory of substance and the theory of change and the
bottoming out of the regress. We simply insist that the level 0 substance
does have both a form and a matter. The Linkage Principle simply does
not apply at the bottom level (though we can still accept it for higher
levels). The form and matter of the level 0 substance cannot be thought of
as inherited from below; they are sui generis. In this way the whole of the
Aristotelian schema is preserved by adding a couple of new entries to
Table 1 : viz. , a matter and form on the bottom line (which are, in any
case, really required by the theory of substance) . Thesis 1 specifies exactly
the form: it is to be the bearer of all properties. It makes no sense to ask
what the matter is, since one can ask only what substances are; and this
matter is not a substance. Of course, if the level 0 substance could cease to
be, and we are to preserve the theory of change, the matter would have to
be a lower level substance. It just follows that it cannot cease to be. (But
this is ruled out for Aristotle anyway: if prime matter could cease to be it
would lose its form; but it hasn't got one, so it cannot cease to be.)
Are there any other arguments for the claim that the level 0 substance
has no essence? In Metaphysics 7 . 3 , Aristotle gives a thought experiment
as to how we can arrive at prime matter (in thought, not in practice, since
matter and its form are not physically separable) . We simply take a body
and strip off all its properties: colour, weight, 'length, breadth and depth';
what is left when all these have been removed is the prime matter. Now if
this is intended as an argument that the level 0 substance has no essence
then it begs the question. For to assume that the form gets stripped off in
the process is exactly to assume that it is a property; but on the present
scheme the essence is not a property - of anything.
In part of the Timeus Plato presents the notion of a receptacle of
qualities, xropcx, which in some ways is similar to Aristotle's notion of
prime matter. 16 Moreover, Plato uses an argument to try to establish that
the receptacle has no intrinsic properties; it cannot have them since,
otherwise, it would 'obtrude them and receive the other qualities
badly'. However good this argument is for Plato, it will not help Thesis
2. For there is no reason to suppose that the form of the level 0 substance
16
For references and discussion, see Sorabji ( 1 988), pp. 32ff. Charlton «( 1 970), appendix)
argues that the traditional notion of prime matter owes more to this passage in the
Timeus than to Aristotle.
The limits of expression 23
1 .8 Cusanus on God
So let us move on to another candidate for something beyond the limits
of the expressible: God. Many have been tempted by the thought that
God is ineffable, and so transcends anything that can be said about
Her. 1 7 There are strong elements of this in the Neoplatonist tradition
of negative theology. But nowhere do we find the idea more graphically
expressed than in the fifteenth-century German philosopher, Nicholas of
Cusa.
Cusanus defines God simply as the Maximum, where 'we speak of a
thing being the greatest or maximum when nothing greater than it can
exist' (Of Learned Ignorance, I, 2 1 8). God, then, is a species of the infinite.
Moreover, according to Cusanus, because God is infinite, there can be no
way that we can comprehend Him. As he puts it (Of Learned Ignorance,
1, 4):
There can be nothing greater in existence than the simple, absolute maximum;
and since it is greater than our powers of comprehension - for it is infinite truth -
our knowledge of it can never mean that we comprehend it.
This may seem a bit swift. Why does the incomprehensibility of some
thing follow from its infinitude? Essentially, it is because, according to
Cusanus, our categories are finite; and no finite category can 'fit' an
infinite object. If this is right, then God is not only beyond comprehen
sion, but facts about God are beyond expression. For any way of expres
sing such facts would require us to subsume God under some category or
other. Hence, facts about God are beyond expression (Transcendence).
This does not stop Cusanus expressing (and presumably comprehend
ing) various facts about God, however. He is quite explicit in attributing
properties required by dogma, such as the Trinity (Of Learned Ignorance,
I, 1 7) . But even if piety did not force him into this contradiction, logic
19 Interestingly in this context, Stace ( 1 9 6 1 ) claims that when mystics are forced into
uttering something contradictory because of their ineffable insights, this is precisely
because those contradictions are true.
20
We will discuss one when we come across Anselm in a later chapter. Some others are
discussed in Alston ( 1 956).
The limits of expression 25
Conclusion
In this chapter we have looked at three Pre-Kantian philosophers whose
views led them to the conclusion that there were things beyond the
expressible, and so into the contradiction typical of the limits of
thought - though only one (Cusanus) recognised and endorsed the con
tradiction. As we have seen, the arguments of the three to the conclusion
that there are things beyond the expressible, cannot be defended.
However, similar arguments of later philosophers are not so easily dis
armed. For the present, though, let us leave the limits of the expressible
and move on to the second of the limits that will concern us. Cusanus, as
we saw, was forced into contradiction because of features of an infini
tude. In the next chapter, we will look at infinity itself.
21
Though we will meet another argument to the effect that finite categories cannot char
acterise infinite objects when we come to Hegel.
2 The limits of iteration
Introduction
In this chapter we will meet a second limit of thought. This arises when
there is some operation that is applied over and over again as far as
possible. I will call this, for obvious reasons, the limit of the iterable:
the most notable case of this is the mathematical (ordinal) infinite.
For reasons that I will not speculate on here, the mathematical infinite
has exercised a primitive fascination for people from earliest times; it is at
the heart of issues in philosophy, mathematics and even theology. Despite
this fact - or perhaps its cause - the notion is a deeply paradoxical one.
For various reasons that we will come to in due course, it has seemed to
people that though there be no greater than the infinite; yet there be a
greater. This is, in fact, the leitmotif of the book.
In virtue of this paradox, the notion of the mathematical infinite has
often been considered problematic. It is commonly claimed that these
problems were cleared up in the nineteenth century, by the writings of
Dedekind, Cantor, and others . For example, Russell: '
For over two thousand years the human intellect was baffled by the problem [of
infinity] . . .
A long line of philosophers, from Zeno to M . Bergman, have based much of
their metaphysics upon the supposed impossibility of infinite collections . . . The
definitive solution to the difficulties is due . . . to Georg Cantor.
2.2 . . . in Aristotle
Let us now turn to Aristotle. What was his theory of the infinite? It can
be summed up very simply: the infinite does not exist. Aristotle's account
of infinity is given in his Physics, mainly in the last three sections of Book
3 . There, he explains that although there clearly are infinity generators,
2 Note that, since an infinity generator must produce an object different from everything
obtained so far, it is better to conceptualise it as an operator on a collection of objects of
the appropriate kind. Thus, instead of the operator I + x, it is better to think of the
operator I + max(X), where X is a finite set of numbers. However, we will ignore this
subtlety for the present.
28 The limits of thought in pre-Kantian philosophy
and these can be applied indefinitely, there are no such things as the limits
these generate. In his words (Physics, 3 . 7 , 206a26-206b 1 6): 3
The infinite exhibits itself in many ways - in time, in the generations of man, and
in the divisions of magnitudes. For generally the infinite has this mode of exis
tence: one thing is always being taken after another, and each thing that is taken is
always finite, but always different. Again, 'being' is spoken of in several ways, so
that we must not regard the infinite as a 'this', such as a man or a horse, but must
suppose it to exist in the sense in which we speak of the day or the games as
existing - things whose being has not come to them like that of a substance, but
consists in a process of coming to be or passing away, finite, yet always differ
ent . . . The infinite, then, exists in no other way, but in this way it does exist,
potentially or by reduction.
2.3 Time
Let us now turn to the question of whether Aristotle's view, that there are
no completed infinities, can be maintained. Aristotle considers some
arguments to the effect that there are actual infinities in Physics 3 .4,
203b l 6-26, and answers them very briefly at 5 . 8 , 208a5-25. By and
large, the arguments do not establish their conclusions. They point to
the existence of various infinity generators, and just assume that there
must be a corresponding infinity. Aristotle has to do little more than
point out that this does not follow.
3 Quotations from Aristotle in this chapter are taken from Barnes ( I 984).
4 Or as the medievals put it, the syncategorematic and categorematic infinite. See Moore
( I 990), ch. 3.
The limits of iteration 29
the General Theory of Relativity to imply that both time and the material
cosmos started together at the Big Bang. The geometry of time would
then be highly non-Euclidean. Whether or not this move is possible, I
leave for modern physicists to argue out; it is not a move that is open to
Aristotle. Once you admit that time past is potentially infinite (as
Aristotle does), it follows that it is actually infinite.
2.4 Motion
The second way in which an infinity can be instantiated in an actual and
terminated process concerns motion. This can be seen, as might be
expected, with the help of one of Zeno's arguments.
Consider a body (which we might identify for present purposes with
the point which is its centre of mass) which moves from A (say 0) to B
(say 1 ) . Consider the sequence of spatial points in the motion 0, 1 /2, 3/4,
7/8 . . . each point (save the first) halving the distance between the
previous point and 1 . The generator that takes us from one point to
the next is clearly an infinity generator. Alternatively, and equivalently,
so is the function that takes us from each half to the next: [0, 1 /2] , [ 1 /2,3/
4], [3/4,
7/8] . . . Moreover, the infinity is completed by the time that the body
arrives at B .
Aristotle discusses this paradox o f Zeno and, naturally, does n o t accept
the conclusion. His solution is to deny that the points (or halves) in
question exist. As we shall see, Aristotle regards it as absurd to suppose
that the continuum is composed of points. In particular, a point is simply
a division of the continuum into two parts, produced by a body at rest.
Hence, if the motion of the body is continuous there are no (actual)
points between A and B and no corresponding halves. As Aristotle
puts it (Physics 8 . 8 , 263a23-30):
In the act of dividing the continuous distance into two halves one point is treated
as two , since we make it a beginning and end [of different halves] . . . But if
divisions are made in this way, neither the distance nor the motion will be
continuous; for motion if it is to be continuous must relate to what is continu
ous; and though what is continuous contains an infinite number of potential
halves, they are not actual halves. If . . . [the body] makes the halves
actual . . . [it will not have] a continuous but an intermittent motion.
5 Sorabji ( 1 983), p. 2 1 3 , notes the unfortunate admission but thinks that Aristotle can get
out of trouble by insisting that the infinitude of potential points is itself a potential
infinite. He cannot, since the infinitude is completed.
6 A similar example is given in Bostock ( 1 973-4), though this will not quite do since the
body stops only instantaneously at each point, and Aristotle requires an object to stop
for a period if it is to define a point. See Sorabji ( 1 983), p. 324.
7 See Vlastos (1 967).
32 The limits of thought in pre-Kantian philosophy
to suppose that a body (i.e. a magnitude) is divisible through and through, and
that this division is possible, involves a difficulty. What will there be in the body
which escapes the division?
To see what Aristotle has in mind, take some finite stretch of conti
nuum (say [0, 1 ]); divide it in half, i.e. remove its mid-point from the set of
points (note that the division here is a conceptual one; Aristotle is not
talking about the application of hacksaws); divide the halves in half; and
so on. The operation of dividing each remaining piece in half is an infinity
generator. Now take this to the limit. What is the result? A set of some
things; but what is each something?
We have a three-way fork. Either (i) what is left has length, or (ii) it does
not; in the second case (iia) it is either nothing or (iib) it is a lengthless
point. The case (i) is contradictory since all positive lengths have been
divided (after n divisions all undivided pieces have length 1 /2ll, which can
be made as small as we please). The case (iia) is contradictory since an
amalgam of nothings is nothing; the case (iib) is contradictory since a
collection of dimensionless points cannot have positive dimension.
With a little rearranging we can put this argument as follows. Let us
call the things that result from division through and through atoms; and
let the totality of all things that can be (conceptually) divided be D. Now,
first, an atom cannot be a member of D (Transcendence), since it is
obtained by dividing through and through and so cannot be further
divided. But an atom must have magnitude, since a magnitude cannot
be composed of non-magnitudes. Hence an atom can be divided, i.e., it is
12 For a review of these see Todd ( 1 980). For further discussion see Sorabji ( 1 983), ch. 1 4.
13 Another reductio used by Philoponus goes as follows. If the number of years, that is, the
number of Januaries, before 1 990 is infinite then the number of months before 1 990 must
be infinite also; and, since there are twelve months to the year, this infinity must be twelve
times the size of the first. Again, we have something (the set of months) being the same
size as a proper subset (the set of Januaries) .
14 See Murdoch ( 1 982), p. 569.
The limits of iteration 35
hence the same size a s c (or i s the same size a s c b y the pairing argument),
and so is in C (Closure).
Now, and to return to Aristotle, if this was the thought that lay,
probably half-formed, behind his remark about Anaxagoras, history
has shown it not to be correct. The story is familiar to anyone with a
first course in set-theory, and so needs no long telling. One needs to
distinguish between one set being a subset of another and one set having
a smaller cardinality that another. l s Once one makes this distinction it
can then be shown that a proper subset of a set can have the same
cardinal size as the whole, and so the Euclidean Axiom fails (though it
is true for finite totalities - from which the original intuition of its cor
rectness was, presumably, drawn). Indeed, it is exactly its failure which
became, for Bolzano, Dedekind, and Cantor, the mark of an infinitude.
The argument to Transcendence therefore fails.
We have now looked at two of Aristotle's arguments aimed at showing
that a completed infinite produces contradictions . Of the two, the second,
being the more general, is historically more important. In a nutshell, the
contradiction it diagnoses is that any finite addition to an infinite mag
nitude does not increase its size, though by the Euclidean Axiom, it must.
In the Middle Ages this problem became known as the 'annihilation of
number' (since the infinite size annihilates any finite addend) or the
problem of 'unequal infinities' . 1 6 As we have seen, advances in mathe
matics, particularly in the nineteenth century, enabled both contradic
tions to be resolved. But the framework in which this could be done, the
general theory of sets, provided new and much more powerful arguments
for paradoxes of the actual infinite. The resolutions had dealt only with
symptoms, not with causes. What these arguments are we will see in
detail in a later part of the book, but we can get a glimpse of them in
another pre-Kantian philosopher: Leibniz. To set the scene, I will start by
considering Aquinas' cosmological argument.
15 Possibly the first person to make this distinction was Gregory of Rimini (see Murdoch
(1 968» , though the earliest definitive statement is by Bolzano in 1 8 5 1 (see sections 1 9-23
of Bolzano ( 1 950» .
16 See Dauben ( 1 979), p. 1 22. For more detailed discussions of the treatments of both
arguments in the Middle Ages see Kretzmann (1 982); Moore ( 1 990), ch. 3; Murdoch
( 1 98 I), ( 1 982).
36 The limits of thought in pre-Kantian philosophy
In the world of se nsible things we find there is an order of efficient causes. There is
no case known (neither is it, indeed, possible) in which a thing is found to be the
efficient cause of itself; for so it would be prior to itself, which is impossible . Now
in efficient causes it is not possible to go on to infinity, because in all efficient
causes following in order , the first is the cause of the intermediate cause , and the
intermediate is the cause of the ultimate cause, whether the intermediate cause be
several, or one only . Now to take away the cause is to take away the effect.
Therefore, if there be no first cause amongst efficient causes, there will be no
ultimate, nor any intermediate, cause. But if in efficient causes it is possible to
go on to infinity, there will be no first efficient cause, neither will there be an
ultimate effect, nor any intermediate efficient causes; all of which is plainly
false. Therefore it is necessary to admit a first efficient cause, to which everyone
gives the name of God.
Now, the last clause, identifying the first cause with God, is a complete
non-sequitur; but our interest lies in the argument used to establish the
existence of a first cause as such. This concerns the generator 'cause of ' as
applied to events or states of affairs. (The distinction is of no importance
here; I will use 'event' and 'state of affairs' interchangeably.) To apply it
we do not need to assume that the cause of something is unique. We need
only suppose that the generator picks out one of the pertinent causes.
Given any event, say the death of Hegel, we can apply the generator. If
we succeed in producing an event, a cause of Hegel's death, this must be a
different event, since no event can cause itself (as Aquinas points out).
There is, however, according to Aquinas, no guarantee that the generator
is an infinity generator. The event may have no cause. The point of the
argument is exactly to establish the existence of such an event.
Does it succeed? In discussing this, it is important to note that when
Aquinas appeals to the notion of cause here, he does have in mind the
notion of cause that would first come to the mind of a modern reader,
viz. , an event temporally prior to the event it brings about (such as
smoking causing lung cancer). Rather, he means an event (or state of
affairs) simultaneous with the event in question such that the relation of
causal dependence between them is, none the less, asymmetric. (So that
when he uses the word 'prior' he is referring to causal, not temporal,
priority Y ) Standard examples are: the motion of the pen is causally
17 See, for example, Copleston ( 1 9 6 1 ), pp. 1 1 7ff. ; Hick ( 1 970), pp. 4 I ff. ; Kenny ( 1 969), pp.
4 1 f.
The limits of iteration 37
18
See Hick (1 970), p. 4 1 , for discussion and references.
38 The limits of thought i n pre-Kantian philosophy
Four things are clear from these quotations. From the first, it is clear
that Leibniz has in mind a temporal regress rather than Aquinas' simul
taneous regress. Next, Leibniz prefers talk of reasons to talk of causes;
and there is good reason fo r this. He is in the process of demonstrating
that there can be a cause for the cosmos. This cause must therefore be
'outside' the cosmos, and so not in time. ' Reason' may therefore be a
better word. None the less, for this chapter I will continue to use 'cause'
for any explanatory state, in or out of time. Third, the second quotation
shows that Leibniz is invoking the Principle of Sufficient Reason (PSR).
What this is, exactly, I will return to, but a traditional way of stating it is
The limits of iteration 39
as follows: for every contingent state of affairs (i.e. state that could (have)
be (en) otherwise), there is a reason why it is as it is.
The fourth, and final, thing is the most important for our purposes:
Leibniz observes that we may ask for a cause of the whole sequence of
Aquinas' causes, even though this is infinite. In our language: the gen
erator 'cause of ' can be applied to the infinity it generates, which is, in
modern jargon, an w-sequence (that is, a sequence isomorphic to the
natural numbers). In other words, Leibniz countenances applications of
such an operator to an infinite sequence. Well, he countenances one such
application anyway; he just assumes that the application of the operator
to the w-sequence will give God. But this is far too swift. First, there is no
reason to suppose that the reason for the w-sequence is extramundane.
For if the events in the sequence occur in Zenonian fashion, for example,
at times t, t - 1 /2, t - 3/4, etc., the cause of the sequence might be some
quite mundane event at t - 1 . But even if the sequence of events extends
back through all time, and so its reason cannot be a state in time, it does
not follow that the state is a divine existence; it does not even follow that
it is a necessary state of affairs. For the cause may itself be contingent,
and so have its own cause.
Leibniz would, I am sure, have replied to this point as follows. If the
cause is contingent, then we can simply apply the operator again, and so
keep going. Eventually, we must come to a point where the operator is no
longer applicable, and so the cause in question is not contingent. This
reply countenances the possibility of infinitude after infinitude. Leibniz
was not averse to such a bold idea however, as is clear from the following
passage (Monadology, sections 36, 37 1 9 ) :
There i s a n infinity of present and past forms and motions which g o t o make up
the efficient cause of my present writing; and there is an infinity of minute
tendencies of my soul, which go to make its final cause [i .e. purpose] .
And as all the detail again involves other prior or more detailed contingent things,
each of which still needs a similar analysis to yield its reason, we are no further
forward : and the sufficient or final reason must be outside the sequence or series
of particular contingent things, however infinite this series may be .
Does Leibniz' strategy work? Let us spell it out a little more carefully.
Consider the sequence generated by repeated applications of the operator
'cause of ' , as far as possible. That is, whenever we have an event we
generate its cause (if it has one); and whenever we have an unbounded
sequence of causes, we generate its cause (if it has one); and we do this as
19 Latta ( 1 898), pp. 237f. The idea that there are infinitudes of different sizes also plays an
important role in his version of the calculus. See Robinson ( 1 966), pp. 2 6 1 -4.
40 The limits of thought in pre-Kantian philosophy
o
far as possible. 2 Call the sequence generated a, and let the limit of a be
either its last member, if it has one, or a itself if it does not. Now note that
the operator cannot be applied to the limit of a to generate a new cause.
For if it could, a would not be the result of applying the operator as many
times as possible. Hence, since the PSR fails to generate a new state, the
limit of a must be necessary rather than contingent. If the limit of a is a
itself then, since it is necessary, it must have some necessary member; for
as Leibniz, in effect, observes in the second quotation above, if all the
members of a are contingent, a is contingent. It follows, therefore, what
ever the limit of a is, that a contains some necessary member. We have
therefore established the existence of a necessary cause. 2! So the strategy
does work.
Whether the necessary cause is the last member of the sequence a (and
so the first cause) depends on what line we take on whether necessary
states of affairs may themselves have causes. If we rule that they do not,
then the necessary cause must be the last member of the sequence, and so
the Ultimate Cause. It should be noted, however, that even with this
ruling, we are a long way from establishing the existence of God, for at
least two reasons. The first is that the necessary cause has not been shown
to be unique. If we had started with a different initial event, or let the
operator 'cause of' pick out different causes, we would have obtained
another sequence; and there is no guarantee that its necessary state would
be the same. Secondly, even if it were, there is no reason to believe - so
far, anyway - that it has any other divine properties (such as omnipo
tence, omniscience, etc.). It must be extramundane since it is not contin
gent; but the fact that it is out of this world does not make it God.
20
Leibniz could not have spelled out what this means exactly, but we know that it means a
transfinite sequence indexed by ordinals. At limit ordinals the operator produces the
cause of the whole preceding sequence.
21
A similar argument can be found in Meyer ( 1 987).
22
See, for example, Hick (1 970), p. 5 1 ; Mackie (1 982), ch. 5 .
The limits of iteration 41
would appear that many unusual things (in the sense that they are not
met with in normal experience) have no cause: the decay of a particular
radio-active atom, and similar quantum transitions, seem to be com
pletely spontaneous. Moreover, to the extent that the purpose of the
argument is to establish a cause of the cosmos, a premise to the effect
that every contingent state has a cause would clearly seem to beg the
question.
However, while we are on the subject of the PSR, let us return to its
formulation. According to the version I gave, every contingent state of
affairs has a reason. But Leibniz sometimes formulates the Principle
rather differently, to the effect that e very state of affairs has a reason.
For example: 23
Naturally, since I think that the weaker version of the PSR is false, I
think this stronger version is false too; but let us re-run the argument of
the last section for a first cause, using this version of the PSR. For a novel
phenomenon then emerges. Consider the sequence generated by repeated
applications of the generator 'cause of ', and let this be rr, as before. (It
can have no last member now.) By the PSR, we can apply the operator to
this to produce a reason for rr. Assuming that nothing can be a reason for
itself, 24 this reason cannot be a member of rr (Transcendence). But the
cause of rr is exactly one of the things generated by applying the operator
in the prescribed fashion. Hence it is in rr (Closure); and hence we have a
contradiction at the limit of the iterable.
Conclusion
Leibniz did not, as far as I know, make the above observation. Had he
done so, he would have been before his time - by at least a century, if
not two. But it serves to bring out the full significance (in the present
context) of Leibniz' observation that it may be possible to apply infinity
Introduction
So far, we have met two limits of thought: the limit of expression and the
limit of iteration (the mathematical infinite). In this chapter we will meet
a third: the limit of cognition . As we shall see, this, too, takes us to the
verge of contradictions of the characteristic kind.
Cognition concerns relationships that arise between agents and the
world that they cognise . Specifically, it concerns the relationships
between thought, or language, and the states these (successfully, one
hopes) relate to; between representations and the things represented.
Typical such relationships, and the ones that will be our primary con
cern, are knowledge, truth, and rational belief.
We have already met various problematic claims to the effect that
certain things transcend our cognition: states in flux ( 1 .3), prime matter
( 1 .6) and God ( l .8). But, as might be expected, the situation is thrown
into sharpest relief by any doctrine to the effect that there are very
definite limits to cognition. Perhaps no doctrine of this kind is more
extreme than one according to which there is no objective knowledge
of the world at all. A number of views of this kind arose in pre
Kantian philosophy. In this chapter we will look at a couple of them.
One way in which the doctrine can arise is in virtue of the claim that there
is no objective truth at all (and hence no objective knowledge). This is
relativism, and is one of the topics we will look at, mainly through the
views of Protagoras. A second way that it can arise is in virtue of the
claim that, though there may be objective truths, no evidence can estab
lish what these are. This is skepticism, and is the other topic we will look
at, mainly through the writings of Sextus Empiricus.
The contradictions at the limit of cognition that we will look at are all
generated in a uniform way. We start with some thesis of the following
form, which, for want of a better name, I will call the Cognition Schema:
Yx(x E � � Cx)
44 The limits of thought in pre-Kantian philosophy
giving this name to what is virtually the sense-presentation . For since this lies in
feeling and involuntary affection, it is not open to question. Consequently, no
one, I suppose, disputes that the underlying object has this or that appearance; the
point in dispute is whether the object is in reality such as it appears to be.
4 For an introduction to the enormous literature see, for example, Pollock ( 1 986) or Lehrer
( 1 990).
46 The limits of thought in pre-Kantian philosophy
this indicates our feeling, whereby we come to end in equipoise because of the
equipollence of the opposed objects [sc. propositions]; and by 'equipollence' we
mean equality in respect of what seems probable to us, and by 'opposed' we mean
in general conflicting, and by 'equipoise' a refusal of assent to either alternative.
But now we have a problem. For we have reason to believe that the
results of applying the criterion are correct only if we have reason to
believe that the criterion itself is correct. And the criterion is not itself
a statement of appearances; hence, if it is justified it must have some
rationale or proof; and now the question arises as to what justification
we have for believing that proof to be correct. Clearly, to appeal to the
criterion at this point would be to beg the question (Outlines of
Pyrrhonism, I, 1 1 7):
The limits of cognition 47
in this way both the criterion and the proof are involved in the circular process of
reasoning, and thereby both are found to be untrustworthy; for since each of
them is dependent on the credibility of the other, the one is lacking in credibility
just as much as the other .
The only other possibility is that we must be able to give another proof
of the correctness of this proof. But the question now arises as to the
justification of this proof. Clearly, we are embarked on a regress; and if
the regress is not to be terminated illicitly by appealing to the very
criterion we were supposed to be justifying, it must go to infinity. But
then it is vicious. For then there is no way that we could ever establish
that the criterion, or any proof in the series, is correct (Outlines of
Pyrrhonism, I, 1 22f.):
i f h e [who i s trying t o justify the criterion] asserts that the proof i s true h e will be
asked for a proof of its truth, and again, for a proof of this latter proof, since it
also must be true, and so on ad infinitum . But to produce proofs to infinity is
impossible; so that neither by the use of proofs will he be able to prefer one sense
impression to another.
Thus, there is no way of justifying the claim that one set of appear
ances, as opposed to another, is a better indication of how things are.
And hence there is no reasonable belief about how things are, as opposed
to appear.
In the case in question, a is of the form 'It appears to be the case that 13' .
Further information, ,,(, may turn up which does give us reason to sup
pose otherwise (that is, is a defeater) . Hence, although we may have
a I- 13, we do not necessarily have a 1\ "( I- 13 - which is the mark of a
non-monotonic logic.
To illustrate the situation further, suppose that a hitherto unknown
painting is discovered. In style and age it is a typical Rembrandt; it also
bears what appears to be Rembrandt's signature. The painting appears to
be a Rembrandt; and, no defeaters being known, we infer that it is a
Rembrandt. But now suppose that a stroke-for-stroke identical picture
turns up, and it is clear that at least one is a forgery, but that there is no
indication as to which it is. Then this information provides a defeater for
the former inference. Notice that the mere possibility of the existence of
the doppelganger is not a defeater; but its actuality is. In a similar way, the
Trope arguments of Sextus certainly establish that given any perception it
is possible for things to appear different to another, or to oneself at a
different time, but this is not sufficient to defeat the inference.
One objection to this kind of account is given by Lehrer «( 1 990), pp .
64f.), who argues that for the inference from 'It appears that 13' to 13 to
work (even non-monotonically), there must be some suppressed premises.
For example, to infer that there is a flag on the mast of the building
opposite, I must have information concerning what flags are like, etc.
5 The most eminent exponent of this kind of line was probably Thomas Reid. See his
Essays on the Intellectual Powers of Man, for example, Essay II, ch. 22. A more modern
exponent is Pollock. See Pollock (1 986), ch. 2.
6 For an introduction to non-monotonic logic, see ch. 6 of Genesereth and Nilsson (1 987).
The limits of cognition 49
Thus, 'It appears that (3' can never on its own justify (3. (And the extra
premises are not themselves statements of appearance; so the regress
looms.) This, I think, misplaces the relationship of the role that the
extra information plays with respect to the inference. The information
is not required as a premise for the argument. (A young child, for exam
ple, might not even be able to articulate the information, but still makes
the inference.) Rather, the information is that in virtue of which the
inference is reasonable. In exactly the same way, the fact that a deductive
argument is valid is not itself a premise of the argument; it is the fact in
virtue of which the argument is reasonable (as was wittily exposed by
Lewis Carroll ( 1 895)). If the inference is reasonable then its sole premise
gives a reason for the conclusion. Why the inference is reasonable is a
different question, and not one that needs to be answered before we know
the premise to be a reason (any more than we must have a theory as to
why modus ponens is valid before a and 'If a then (3' provide a reason for
(3). Hence the regre ss is not generated.
It is worth noting that Sextus' argument does establish Carneades'
form of skepticism, at least for certain kinds of claims. Just because the
inference to how things are is non-monotonic, it is possible that a defeater
may turn up, and so that we may have to retract our conclusion. It is not,
therefore, certain. Put differently, if one takes the inference: it appears to
be the case that a; hence it is certain that a; to be itself a non-monotonic
inference, then the mere possibility of alternative perceptions, and all that
this may entail, is a defeater for this inference. Hence it is never valid.
Posterity has, in fact, strengthened this version of the skeptical argument.
For we now have a much better understanding of the theory-dependence,
and consequent fallibility, of observation; but I shall not pursue this
matter here. 7
This is hardly a comprehensive discussion of the topic. However, it at
least indicates where, I take it, the argument for Sextus' form of skepti
cism fails; which is sufficient for the present.
7 For a brief discussion and references, see, for example, ch. 3 of Chalmers ( 1 976).
50 The limits of thought in pre-Kantian philosophy
there are many things which produce the same effect o n themselves a s they
produce on other things. Just as, for example, fire after consuming the fuel
destroys also itself, and like as purgatives after driving the fluids out of the
body expel themselves as well, so too the argument . . . can cancel itself also.
And again, just as it is not impossible for the man who ascends to a high place
to overturn the ladder with his foot after the ascent, so also it is not unlikely that
the Sceptic after he has arrived at the demonstration of his thesis by means of the
argument . . . as it were a step ladder, should then abolish this very argument.
8 Notice that the contradiction does not arise for Carneades and similar skeptics, who take
P to be certitude. The result of applying their skepticism to itself is the claim that
skepticism is not certain. The argument to Closure fails, however. For them, the skep
tical arguments simply show that their skepticism is rationally acceptable. They do not
show that it is certain, for the conclusions are only as good as the premises, and these are
not certain.
The limits of cognition 51
9 See section I of Feyerabend ( 1 977). Similar comments are also made by Hume about
skeptical arguments in the last paragraph of the Treatise, I, 4, 2.
52 The limits of thought in pre-Kantian philosophy
to
See Priest ( 1 987), 4.6.
11
Though it is commonly held to be reasonably accurate. See Kerford ( 1 967).
The limits of cognition 53
We will next be concerned with the arguments put forward for and
against this view. It is important, at the start, to note that the concept of
truth employed here is a relativist one, in the sense that the truth of any
claim is relative to a cognitive agent (true for x). Protagoras need not
claim that there is no objective (agent independent) notion of truth (truth
simpliciter). All that he needs to claim is that nothing satisfies such a
notion.
13 And arguably, the privacy of the object even renders the claim of correctness senseless.
The object is rather like Wittgenstein's beetle in a box (Philosophical Investigations,
section 293).
1 4 See McDowell ( 1 973), pp. 1 37f. McDowell suggests (p . 1 1 8), and I think plausibly, that
the official sense of perception at issue here is non-propositional altogether.
15 The second ( l 6 l c) is a very swift ad hominem argument to the effect that Protagoras' view
is human chauvinist . The third is launched immediately afterwards, and is to the effect
that if whatever anyone believes is true to them, then all views are equally good, and so
there is no such thing as wisdom (which Protagoras professes to teach). After an inter
lude, Protagoras (in the person of Socrates) replies ( 1 66d f.) that wisdom has nothing to
do with truth; the wise person is, essentially, one whose views are most effective in
producing happiness. Again, after an interlude, Socrates objects to this reply ( l 77c ff.),
essentially, that whether or not a person's views are conducive to producing the effect
aimed at is an objective matter, and not one in which each person's view is equally valid.
The limits of cognition 55
Protagoras' doctrine, they must agree that no one ever makes a false
judgment. This, in itself, is said to be incredible ( 1 70c9), though
Protagoras would not have conceded this. (No person makes a judgment
that is false for themself; but they may make a judgment that is false for
x . ) The main argument comes when the observation is applied to
s oc RATE s . So if he admits their opinion is true - that is, the opinion of those
who believe that what he thinks is false - he would be conceding that his own
opinion is false?
THEODORUS. Certainly.
SOCRA T ES . But the others don't concede that what they think is false?
THEODORUS . No .
SOCRATES . And Protagoras, again, admits that judgments of theirs are true, too,
SOCRA TES . Well then, since it's disputed by everyone, it would seem that
Protagoras' Truth isn't true for anyone: not for anyone else and not for
Protagoras himself.
The patient's view of whether they have recovered, for example, is more valid than the
doctor's. I am dubious that this argument works, if Protagoras is prepared to stick to his
guns and insist that judgments about efficacy are relative, too (which he may do, it seems
to me, with just as much plausibility as with judgments about other matters).
56 The limits of thought in pre-Kantian philosophy
to engage in discourse at all [Protagorasl has to assert that something is the case .
It is not just that he is pretending to have a certain social role, that of teacher or
'wise man', which he is not entitled to claim . . . The matter cuts deeper: it is
presupposed in all discourse that some propositions are true, that there is a
difference between being the case and not being the case, and to deny this in
o
discourse is already to presuppose the existence of the difference . 2
Suppose (<p) E II
then <p since being P entails being true
i.e. Vx(x E E� x (j. II)
so (<p) E E � (<p) (j. II
But ( <p) E E
Hence (<p) (j. II
Thus, (<p) E II entails (<p) (j. II; it follows (by the Law of Excluded
Middle or reductio) that (<p) (j. II. Thus, we have established, on no
assumptions, that <p is not P. This would not trouble Sextus (when P is
knowledge), or Protagoras (when P is truth) . But note that if P is truth
then it follows that <p is false, i.e., -Nx(x E E� x (j. II); and thus, argu
ably, that 3x (x E E 1\ x E II) . But this x cannot be <p itself (since <p is not
true), so there must be something else in E. This is paradoxical since, as
far as the logic of the argument goes, apart from containing <p, E was
quite arbitrary. Yet we have a proof that there must be something else in
it. 2 1
20 Passmore ( 1 970), pp. 67-8 . The fact that there is a difference between truth and untruth
does not, note, entail that nothing is both.
21 This paradox is spelled out very nicely by Prior ( 1 9 6 1 ) . For a further discussion, see Priest
( l 99 1 b).
The limits of cognition 59
Conclusion
We have seen that Sextus' and Protagoras' views generate contradictions
of the kind typical of the limits of cognition. These contradictions
depend, ultimately, on their arguments for skepticism and relativism,
which, as we saw, do not work. But we have also seen that in the vicinity
of these contradictions there are others that are produced by applying
instances of the Cognition Schema to itself, and that depend on no such
22 For example, let '�' be some new name, and let <p be the sentence Vx(x = (<p) --> x ¢ II).
We now stipulate that '�' refers to {<p} . See Kripke ( 1 975), p . 693 .
23 Those who think that --> i s relevant will have t o take i t t o be a n enthymematic implication
at this point, as in Meyer ( 1 973), p. 1 72.
24 See, for example, the references to the paradox in Kneale and Kneale ( 1 962).
25 See Kaplan and Montague ( 1 960). Strictly speaking, for the argument to Closure to work
in this case, we need not just ( * ) , but that ( * ) is known to be true. But, since ( * ) is true by
definition, this is unproblematic.
60 The limits of tho ught in pre-Kantian philosophy
Introduction
In this chapter we will discuss the last of the four limits that will concern
us in the book: the limit of conception. It is common enough to suppose
that there are things beyond conception; but prima facie at least, it is
difficult to do so without conceiving them in some sense. Hence the
contradiction at the limit of conception. We will see this phenomenon
arising in two pre-Kantian philosophers. The first is Anselm, whose
concern is God. The second is Berkeley, whose concern is quite differ
ent: idealism. Despite the difference, we will see that Anselm and
Berkeley are tied together by more than just contradiction. They also
share certain views about how objects may be characterised.
Specifically, they both subscribe to the view that if an object is charac
terised as the (or a) thing with such and such properties, then it must have
those properties. If we let 8 be some description operator (so that 8xrp(x)
may be read as 'a/the thing with property rp'), then they both subscribe to
the schema:
rp( 8xrp(x) )
Following Routley ( 1 980), I will call this the Characterisation Principle,
'CP' for short. As we shall see, the CP is not, in general, correct. Despite
this, the contradiction at the limit of conception shines through the work
of these two philosophers.
Is there, then, no such nature as You, for the Fool has said in his heart that God
does not exist? But surely when this very Fool hears the words 'something nothing
greater than which can be thought' , he understands what he hears . And what he
understands is in his understanding, even if he does not understand [judge] it to
exist . . . But surely that than which a greater cannot be thought cannot be only
in the understanding. For if it were only in the understanding, it could be thought
to exist in reality - which is greater [than existing only in the understanding] .
Therefore, if that than which a greater cannot be thought existed only in the
understanding, then that than which a greater cannot be thought would be that
than which a greater can be thought ! But surely this conclusion is impossible.
Hence, without doubt, something than which a greater cannot be thought exists
both in the understanding and in reality.
1 Quotations from the Proslogion are taken from Hopkins and Richardson ( I 974).
2 My own shot is as follows. Let rx be 'x is conceived' . Then God (g) may be defined as
8x..,3y(ry A y > x). (Quantifiers, note, are not existentially loaded.) Let <p(x) be the
second-order condition Ex A \fP(P f. E -> (Px Pg)), where E is existence. Then the
+-->
What, exactly, the structure of Anselm's argument is, is less than obvious.
However, again, it is clear that it depends on the CP, since it relies on the
claim that nothing greater than God can be thought - and, for good
measure, it requires that the thing that is greater than can be thought
is greater than can be thought. It is also clear that the argument assumes
that God can be thought of, or else there is no reason for supposing that
something greater than anything that can be thought of is greater than
God. s
Leaving the cogency of the argument aside for the moment, it is clear
that Anselm is caught in a contradiction characteristic of a limit of
5 The best I can make of the argument is as follows. It requires us to take God, g, to be
'
8x ..., 3 y(ry /I y > x /I y f x ) . Then if g is 8xVy(ry
-> x > y), two applications of the CP
give:
..., 3y(ry /I Y > g /I Y f g) (1)
Vy (ry
-> g* > y) (2)
The logic of the argument also requires the following:
rg (3)
rg* (4)
Now suppose that g f g* .
rg g* > g
-> by (2)
so g* > g by (3)
Hence rg* /I g* > g /I g* f g by (4)
so 3 y(ry /I y > g /I Y f g)
contradicting ( I ). Hence by reductio g = g* . And so:
Vy (ry -> g > y) by (2)
The main problem with this reconstruction is that the clause y f x seems to be redundant
in the definition of g since it is entailed by the clause y > x. But if we delete it, the
supposition g f g* is redundant and so does not get discharged in the reductio. The
only solution I can offer is that Anselm does not take y > x to entail y f x. After all, (2)
' O
and (4) entail that g > g , and presumably it is not the case that g* f g* .
64 The limits of thought in pre-Kantian philosophy
thought. The argument shows that God is not in the set of things that can
be conceived (Transcendence) . Yet Anselm obviously is conceiving God
in putting forward the argument (Closure).
Anselm is, in general, aware of the contradiction involved in saying
anything about the inconceivable. In chapter 65 of the Monologion he
tries to defuse the problem by insisting that our concepts do not apply to
God literally, but only by analogy (per aliud ). This solution cannot be
sustained. For in the claim that God is inconceivable 'inconceivable' must
have its literal sense, or the whole force of the claim is lost. God must
therefore be literally conceivable, if only as the inconceivable. In any case,
the move will not remove this particular problem. As I have just
observed, the logic of the argument itself requires that 'God is conceiv
able' is true (in exactly the same sense that conceivability is used in the
conclusion), so the argument itself locks Anselm into this claim.
6 This is essentially Descartes' version of the Ontological Argument in the 5th Meditation.
The limits of conception 65
such a thing. What Kant actually says is that existence is not a determin
ing predicate. On a reasonably generous reading of the Critique this can
be interpreted as the view that existence is not a predicate that can be
used in a valid instance of the CP, as versions of the Ontological
Argument require.
Leaving Kant aside, there certainly are instances of the CP that are
true: the highest mountain in the world, for example, is, undoubtedly, the
highest mountain in the world. However, the essential point remains: the
CP cannot be assumed in general. And until the instances of it that are
used in Anselm's arguments are legitimised (which no one has ever
succeeded in doing), this is sufficient to sink the arguments of
Proslogion 2 and 1 5. Anselm, then, did not succeed in establishing the
contradiction at the limit of conception. The matter is different when we
turn to Berkeley.
As is clear from the last sentence of this quotation, Berkeley held a much
stronger view: that all things (except, possibly, minds themselves) are
essentially thought. It is not just that there is nothing that is not
thought; there could be nothing. Important as it is in Berkeley's philoso
phy, this stronger view need not concern us here. Hence, slightly inaccu
rately, I will call the claim that everything is thought of, 'Berkeley's
Thesis' .
Berkeley advanced a number of arguments for his Thesis. Most of
these are direct arguments depending on various empiricist assump
tions, and have been well hammered in the literature. 7 He has one
time unseen?
iv. HYLAS . No , that were a contradiction.
which is unconceived?
VI. HYLAS. It is.
vii . P H I LON 0 u s . The tree or house therefore which you think of, is conceived
by you?
viii. HYLAS . How should it be otherwise?
xi. P H I LO N OUS . How then came you to say, you conceived a house or a tree
(0)
T3x-,Tx (1)
9 In particular, there is no sophistic slide between the two notions, contra Wisdom ( 1 953),
pp. 8f. , and Thomson ( 1 956).
68 The limits of thought in pre-Kantian philosophy
Torc (2)
Whether or not this is so, notice that the reductio that Philonous performs
is on (2), not ( 1 ) . Here is the first puzzle, then. Even if (2) implies ( 1 ), it is
clear that ( 1 ) does not imply (2); how, then, is the reductio supposed to
work?
Let us leave this for the time being and ask what the contradiction is, to
which (2) is suppl)sed to lead. It is 'conceiving a thing that is unconceived'
(v). We may reasonably understand this as :Jx(rx 1\ on ) , where this is
clearly meant to follow from:
rc 1\ orc (3)
Now how is (3) supposed to follow from (2)? The first conjunct is sup
posed to follow from the fact that Hylas is doing the conceiving (vii-x).
In particular, to conceive c as being something is, ipso facto, to conceive
of c. Let us call this the Conception Schema:
How is the second conjunct of (3) supposed to follow? On this the text is
silent. Hylas just assumes it, and there seems no reason why it should
follow from (2). It would appear that if Hylas had had his wits about him
he should just have said: look Phil, I know that I conceived that c was
unconceived; but that doesn't imply that c is unconceived, any more than
my conceiving the moon to be blue implies that it is blue. Here, then, we
have the second puzzle: where does the second conjunct of (3) come
from? 1 0
Let us take stock. The argument so far looks as follows, with ? indicat
ing the lacunae. I put the argument in an informal natural deduction
form (essentially in the style of Prawitz ( 1 965» . The bar over a premise
indicates that it is an axiom, not an assumption, and so not a candidate
for being discharged.
10 Pitcher ( 1 9 77), pp. I 1 3ff. , and Tipton ( 1 974), pp. 1 74f. , argue that Berkeley confuses the
properties of the conception with properties of the thing conceived, and so, in effect, fau lt
the argument at this point. There is a sense in which this is right; but, as we shall see, it is
not the end of the matter.
The limits of conception 69
T-'Tc ----> TC ?
TC
TC 1\ -'TC
3X(TX 1\ -'TX)
II
An answer different from the one I shall give is provided in Prior ( 1 955). This is discussed
and rejected in Priest ( l 994a).
12
For the orthodox Hilbertian account, see Leisenring ( 1 969). For a slightly different
account, see Priest ( l 979b) .
70 The limits of thought in pre-Kantian philosophy
..:
T ...,-'T..=C
_ _______ ----T:......,
-= _T'-'C"---t--..!.
- T-=.C :JX..., TX :JX""TX --t ""TC
TC ""TC
The limits of conception 71
not know this. Now the argument involves quantification into the scope
of T, in the form of the c-operator; and the sense of quantification into
intensional contexts is notoriously problematic. Quantification into
intensional contexts must have sense (since we often operate with it
informally), and maybe the premises with T understood intensionally
are true. But until we have a clearer understanding of what this sense
is, we will do best to interpret conception in its extensional sense here. It
is worth noting that even those who hold that conceiving is mainly
(primarily, normally) intensional cannot deny that there is an exten
sional sense, namely, conception-under-some-description-or-other. And
just as we can (and I will) interpret 'conceive' extensionally, we can (and I
will) interpret 'conceivable' extensionally (conceivable-under-some
description-or-other) .
Now, one premise of the argument is: it is conceived that there is
something that is unconceived; or, on the modal interpretation: it is
conceivable that there is something inconceivable. Both of these are
unproblematically true since I, for one, conceive such things. Even to
suppose them as the first step of a reductio requires one to conceive
them (they do not even have to be possible) .
The other premise of the argument is: there is something that is uncon
ceived; or, in the modal interpretation: there is something that is incon
ceivable. The first of these is, again, unproblematically true. It is
obviously difficult to give an example of something that is unconceived
(difficult, though not impossible as we have seen!); but we know that
there are (and were) things going on on planets in galaxies far away in
space and time that no sentient being will ever think of. These are
unconceived things. Moreover, in my life I have only a finite number
of thoughts, and so will conceive of only a finite number of things; and
so will everyone else. The number of things conceived (even by the human
race) is therefore finite. Since there is an infinite number of things
(numbers, points in space, etc.), most of them will not be conceived.
The claim that there are some things that are inconceivable is less
clearly true. It is pretty clear that a medieval monk could not conceive
of a computer programmer, a Hilbert Space or a black hole, at least
under those descriptions: he did not have the concepts available. And
presumably, in exactly the same way, there are things of which we can not
now conceive, but of which future generations will be able to. But recall
that conceivability is extensional here; and it does not follow from this
that these things are not conceivable by us under any description. The
premise is, none the less, true. If something is conceivable there must be
some way of singling it out before the mind. Such singling out may
involve direct perception (acquaintance) or may be purely by descrip-
The limits of conception 73
tion. Now consider, for example, a point in space. Continua being dense,
there is no way that any particular point can be picked out by our finite
perception; and, since there is an uncountable number of points but only
a countable number of descriptions in our language, 1 4 there is no way
that every point can be picked out by a description. The matter is even
clearer with uncountable totalities of mathematical entities, like the ordi
nals, where perception is not even on the cards. There are not enough
names to go around for every ordinal. But even if there were, we have no
way of singling out most ordinals to pin names on them. 1 5
Having established the truth of the premises, let us turn to the rele
vance of this fact. This is that a reductio is illegitimate in the context of
the argument. If the premises of the argument are true, so is the conclu
sion; but if the contradiction entailed by certain assumptions is true, this
entailment cannot be used to reduce those assumptions to absurdity.
How, exactly, one cashes out this insight depends on a variety of fac
tors, such as the exact details of the semantics of descriptions and of the
underlying logic, but the situation is essentially as follows . Let the con
junction of the negation of Berkeley's Thesis and the conceivability of
this be 0: . Let f3 be the conjunction of the other premises of the argument,
namely, the instances of the c-axiom and the Conception Schema
employed. 0: /\ f3 entails a contradiction. If entailment contraposes, we
can get from this to -' ( 0: /\ f3) . But even if we can get here, the argument
from this and f3 to -' 0: is an instance of the disjunctive syllogism; and this
fails in dialetheic contexts of the kind we are in. 1 6
There is a way that Berkeley - or at least, Berkeley fortified by
Hilbert's theory of c-terms - could have short-circuited the reductio to
obtain his conclusionY In Hilbert's c-calculus the inference from
rp(cx-,rp(x) ) to Vxrp(x) is valid; this is just the contraposed form of the
c-axiom. In particular, therefore, the inference from rc to Vxrx is valid.
Now, take the fragment of the repaired argument up to rc, and stick this
inference onto the end. We get:
14 See ch. 8 if you are not familiar with the tenus 'countable' and 'uncountable'.
1 5 For some further relevant discussion, see ch. 5 of Nagel ( 1 986).
16 See the Introduction to the book, fn. 8 .
17 A s both John Bacon and Vladimir Smirnov pointed out t o me.
74 The limits of thought in pre-Kantian philosophy
Thus, Berkeley might have traded in his reductio proof for a direct, or at
least, ad hominem argument. 'Look Hylas', he might have made
Philonous say, 'you agree that you do conceive that something exists
unconceived. Well, it follows from this that everything is conceived! '
Neat a s this argument is, i t fails. The reason i s that though the c-axiom
contraposes in non-dialetheic semantics, it does not do so in dialetheic
semantics. To see why, suppose that some things are <p and only <p, some
things are '<p and only '<p, and some things are both. It follows that
'v'x<p(x) is just plain false (since the universal quantifier acts like an
infinitary conjunction). 'cx.<p(x) ', we are agreed, denotes one of the
things that satisfies '<p, but if it also satisfies <p then <p(cx.<p(x) ) is true
(and false too; but at least true). Hence the contraposed c-axiom fails to
be truth-preserving. 1 8
18 This can be turned into a formal counter-model in the semantics of the appendix to Part
3. I leave the details as an exercise.
1 9 Put to me independently by Thomas Forster, Mark Lance, and David Lewis.
The limits of conception 75
20
For example, if x is a non-empty set of non-empty sets the following is a choice set on x in
ZF {y; for some z in x, y is a (selected) member of y} (see Leisenring ( 1 969), pp. 1 05ff.) . It
is well known that in general choice sets cannot be demonstrated to exist in ZF if only the
usual quantifiers are available.
76 The limits of thought in pre-Kantian philosophy
Let us call this the Core Argument. The Core is an argument to a contra
diction from the negation of Berkeley's Thesis together with the extra
premise, TC. It does not employ the Conception Schema, Prefixing, or
even the operator T. Moreover, the extra premise is true, or at least can
be made true. One can conceive of an object that is not conceived
(inconceivable), in the appropriate sense; I do, and so can you.
Of the two interpretations of this argument (the original and the
modal), it is, arguably, the modal interpretation that is the more funda
mental. For the truth of the conclusion of the original interpretation is a
merely contingent affair, depending as it does on the fact that someone is
actually doing the conceiving; whilst the truth of the conclusion on the
modal interpretation depends on no similar assumptions. At any rate, in
future chapters I will call the modal interpretation of the Core Argument
Berkeley 's Paradox. 2 1
21
In ( 1 992) Slater comments on what is, in effect, Berkeley's Paradox. In as far as I
understand them, his comments come to the following. Berkeley's Thesis is true, but
innocuous, when properly understood; and hence the argument can be interpreted as a
reductio. Slater gives an independent argument for the Thesis, and attempts to show why
it is innocuous. The independent argument is, essentially, the last one we looked at the
end of 4.7, and as we saw there, it fails. (Slater gives an argument of sorts for the
contraposability of the c-axiom, but this simply begs the question.) The Thesis is sup
posed to be innocuous since a certain description ('a thing not denoted by this term')
refers to everything. Now, this is just not possible according to the standard semantics of
c-terms that I have been operating with. So even if Slater has some non-standard
semantics in mind, this is beside the point. Slater does sketch an independent argument
for the claim that the term in question refers to everything, but this, too, depends on
contraposing the c-axiom.
The limits of conception 77
Conclusion
With these baptisms, the first part of the book ends. The string of pre
Kantian intimations that limits of thought may be contradictory comes
to a climax in Berkeley's argument (the first pre-Kantian argument we
have met that succeeds in establishing a contradiction at the limits of
thought - though we have glimpsed later ones). Berkeley's idealism pre
vented him from realising the contradiction; but idealism is no escape, as
we have seen. In the next part of the book, we turn to the two philoso
phers in whose work the contradictory nature of the limits of thought was
recognised and theorised for the first time: Kant and Hegel.
Part 2
We say: but that isn't how it is! - it is like that though! and all we can
do is keep repeating these antitheses.
Introduction
In the first part of the book we found limits of thought appearing at
various sites in philosophy. We also found numerous intimations that
they could be contradictory, in the form of arguments produced by
various philosophers. None of these philosophers, with the exception of
Cusanus, was happy to accept that they were dealing with a literally
contradictory object; and even Cusanus had little theoretical understand
ing of the situation in which he found himself. This state of affairs
changed radically in the eighteenth century, however, with the work of
Kant and Hegel. In these philosophers, for the first time, we arrive at a
general recognition of the contradictory nature of the limits of thought,
together with a theorisation of how and why this occurs. In this part of
the book we will review the relevant work of these two philosophers,
starting with Kant.
We have met four limits of thought: the limit of conception, the limit of
expression, the limit of cognition, and the limit of iteration (the generated
infinite). By and large, these limits arose in quite different contexts, and
had little to do with each other. When we turn to Kant we find all these
limits occurring. The first three arise in connection with Kant's views on
the Categories; the last in connection with his account of the Antinomies.
Moreover, in Kant these limits are not at all independent, but interact in
an important way, as we shall see. In this chapter I will deal with the first
topic (the Categories), saving the second (the Antinomies) till the next.
Due to the interaction of the various notions, however, neither chapter
stands completely on its own.
The principal text we will be concerned with is the Critique of Pure
Reason (to which all citations refer, unless otherwise stated ! ). The aim of
this text is to chart the logical geography of the most general features of
I All quotations from the Critique in this chapter and the next are taken from Kemp Smith
( 1 933).
82 The limits of thought in Kant and Hegel
thought and, as its name suggests, spell out their limits. Un surprisingly,
then, it is forced to confront the contradictions inherent in the limits of
thought. Interpreting the Critique is a very sensitive issue, however. This
is so partly because of the abstract and jargonated way in which Kant
expresses himself. More importantly, it is because it is a text composed
over some ten years, during which time his views changed. The interpre
ter's nightmare is exacerbated by the fact that Kant made major changes
to the second edition of the Critique; and, to make matters worse, pub
lished a short and 'popular'! version, between the two editions,
Prolegomena to any Future Metaphysics.
The result is a set of documents that contain numerous internal contra
dictions, not of the kind that we are looking at - which are due to the fact
that the subject involved is inherently contradictory - but which are due
simply to the patchwork of their construction. (All of this is meticulously
documented in Kemp Smith ( 1 923).) Thus, though what I shall say seems
to me to be a correct interpretation of the considered Kant, it would be
foolish to deny the possibility of other interpretations; and to take on
other commentators here would certainly result in obscuring the wood
for the trees, so I shall not attempt it. If I am right, Kant should get the
credit for being the first person to note that contradictions of the kind we
are discussing are, in some sense, inherent in the limits of thought. If I am
wrong, the credit merely goes to someone else.
internal sensations) . Many would argue, however, that not all things in
(space and) time are phenomena. For there are many physical entities,
including those that are responsible for our perceptions (such as photons
and electromagnetic radiation), which are not themselves perceivable.
It is therefore important to note that Kant has a somewhat idiosyn
cratic view about what sorts of things phenomena are. For Kant thinks
that objects in themselves cannot be perceived, or intuited in his jargon;
what are perceived are our mental representations of such objects. He
calls this view 'Transcendental Idealism' . As he explains it (A 1 09):
Appearances are the sole objects which can be given to us immediately, and that
in them which relates immediately to the object is called intuition. But these
appearances are not things in themselves; they are only representations, which in
turn have their object - an object which cannot itself be intuited by us, and which
may, therefore, be named the non-empirical, that is, transcendental object = x.
Table 2
concept horse). The precise details are not too important, but how they
are obtained is. Kant abstracts them from what he took to be the logical
forms of judgments, or statements as we might now put it. In the neo
Aristotelian logic he endorsed, every judgment has a quality, quantity,
relation, and modality. And it may have each of these in one of three
ways. Corresponding to each of these ways is a Category; these can be
displayed as in Table 2. 3
To illustrate: consider, for example, the judgment 'Some capitalists
may not be compassionate. ' This has particular quantity, negative qual
ity, categorical relation, and problematic modality. It thus deploys the
Categories of plurality, negation, substance, and possibility. Or again, the
statement 'If a piece of metal is heated then, necessarily, it expands' has
universal quantity, affirmative quality, hypothetical relation, and apodic
tic modality. It thus deploys the Categories of totality, reality, cause, and
necessity.
The exact details of the taxonomy do not now withstand much
thought. A judgment such as 'all capitalists are exploiters or some work
ers are exploited' is both particular and universal; the judgment 'Oswald
3 I take the table from the Prolegomena, section 2 1 , except that I have reversed the order
of the three quantities, following Bennett (1 966), p. 77. Quotations from the
Prolegomena are all from Beck (1 950). It is perhaps stretching the point a little to
say that the category of modality is a matter of logical form in the modern sense,
for Kant takes this to be semantic rather than syntactic. See A 74 B 1 00 fT. I will ignore
=
this subtlety.
Noumena and the categories 85
killed Kennedy' is categorical, but clearly causal; and so on. But the
underlying weakness here is in no small measure due to the inadequacy
of the theory of logical form in traditional logic. It would be an interest
ing exercise to draw up a similar table of Categories based on a modern
notion of logical form (one which, for example, gave a distinctive form to
agentives), but I shall not attempt this here. The only points we need to
note are that the Categories are abstracted from the logical forms of
judgments, and, crucially, that each judgment deploys one or more
such Category, as Kant himself remarks in the following corollary
(A245 B302):
=
Kant comes back to this point again and again in the Critique (for
example, A95, B 1 47, A 1 39 B 1 78 , A239 B298). His arguments for it
= =
We do not need to follow the argument through all its tortuous turns;
essentially, it runs as follows. It is a feature of each individual conscious
ness that it has a unity. (The perceptions of the depression of the keys on
which I type and of the keys' clicking belong to one consciousness.) How
is this possible? It is possible, according to Kant, because the objects of
consciousness themselves have a unity. (The depression of the keys and
the sound of the clicks form part of the one thing: my typing.) How is this
unity possible? The answer, again according to Kant, is that it is precisely
my judgments deploying the Categories that unify these disparate things.
(The depression of the keys causes the clicking sound. ) Thus, the applic
ability of the Categories is guaranteed by, ultimately, the unity of my
consciousness. It follows that the Categories are mental features that are
- like space and time - constitutive of my perceptions, mental representa
tions. And thus that they apply just to phenomena. As Kant sums it up
(A I 1 1 ) :
The a priori conditions o f a possible experience i n general are a t the same time
conditions of the possibility of objects of experience. Now, I maintain that the
Categories, above cited , are nothing but the conditions of thought in a possible
experience just as space and time are the conditions of intuitions for that same
experience. They are fundamental concepts by which we think objects in general
for experiences, and have therefore a priori objective validity. This is exactly what
we desired to prove.
For present purposes we need not discuss the part of Kant's argument
to the effect that it is the Categories that make the unity of my conscious
ness possible. For, even if that is right, the final stage of the argument is a
patent non-sequitur. From the fact that the Categories just are a certain
kind of condition of my experience, it does not follow that they cannot
function in another way; that they are 'nothing but' this, as Kant puts it
in the quotation. Similarly, from the fact that a screwdriver just is that
tool which propels screws, it does not follow that it does not have other
uses - which I leave to the imagination. And nothing in the
Transcendental Deduction - that I can see, at any rate - appears to
guarantee this.
A second, and better, reason that Kant gives for supposing that the
Categories apply only to phenomena goes as follows. Kant observes that
to apply a Category it is necessary for us to have some criterion, or
schema in his jargon, of its applicability. In the 'Schematism of the
Pure Understanding' Kant gives what he takes to be the criteria of the
applicability of the Categories. He does not deny that, logically, there
could be other criteria; but, as a matter of fact, these are the only criteria
that we have, or that beings constituted like us could have.
Noumena and the categories 87
Now, it turns out that the criteria for all the Categories involve time.
To give a couple of the simpler examples (A 1 43 = B 1 8 3 ff.): 'the schema
of substance is permanence in real time', 'the schema of necessity is
existence of an object at all times' . It follows that it makes sense to
apply the criteria only to those things that are in time: phenomena. As
Kant puts it (A I 45 = B I 84 ff.):
W e thus find that the schema of each Category contains and makes capable o f
representation only a determination of time . . . The schemata of the pure con
cepts of the understanding are thus the true and sole conditions under which these
concepts obtain relation to objects and so possess significance. In the end, there
fore, the Categories have no other possible employment than the empirical.
5 Ibid., p. 369.
88 The limits of thought in Kant and Hegel
But our further contention must also be duly borne in mind, namely, that though
we cannot know these objects as things in themselves, we must yet be in a position
to at least think them as things in themselves; otherwise we should be landed in the
absurd conclusion that there can be appearances without anything that appears .
The pure categories, apart from formal considerations of sensibility, have only
transcendental meaning; nevertheless they may not be employed transcendentally,
such employment being in itself impossible, inasmuch as all condition of any
employment in judgments are lacking to them, namely, the formal conditions
of the subsumption of any ostensible object under these concepts. Since, then,
as pure categories merely, they are not to be employed empirically, and cannot be
employed transcendentally, they cannot, when separated from all sensibility, be
employed in any manner whatever.
7 See Wolff ( 1 963), p. 1 52 n., and Lukasiewicz ( 1 953), p. 72.
90 The limits of thought in Kant and Hegel
Understanding accordingly limits sensibility, but does not thereby extend its own
sphere. In the process of warning the latter that it must not presume to claim
applicability to things-in-themselves but only to appearances, it does indeed think
for itself an object in itself, but only as transcendental object, which is the cause of
appearance and not itself appearance.
And this is but the tip of the iceberg. l O When Kant says that noumena
may be supposed to exist (A253 = B309) he deploys the Category of
existence; when he says that they are not in time, he deploys the
Category of negation. Even the statement that the Categories cannot
be applied to noumena deploys the Categories of possibility and nega
tion! Hence unless Kant is to accept that his own theory is meaningless it
must be possible to express such things (Closure).
Yet it is not (Transcendence). For, as Kant puts it (A679 = B707):
There are no concepts available for any such purpose; even the concepts of reality,
substance, cause, nay, even that of necessity in existence, lose all meaning, and are
empty titles for [ possible] concepts, themselves entirely without content, when we
thus venture with them outside the field of the senses.
But this is not yet an end of the matter. For not only is Kant's own
theory beyond the limit of the expressible, noumena themselves are
beyond the limit of the conceivable ( thinkable). If I can think about
=
certain objects, then I must have some way of fixing on them mentally;
and I can use this fact to make them the subject of some assertion. Hence
I can make a judgment about them. Since one cannot make judgments
about noumena, one cannot, therefore, think of them. Hence noumena
are precisely objects beyond the conceivable (Transcendence). Yet clearly,
5.6 Analogy
In 4.2 we noted that Anselm, when faced with a similar situation in which
our concepts could not apply to God, tried to defuse the contradiction
implicit in making assertions about God by claiming that apparent
applications were to be understood only analogically. Could Kant
make a similar move?
92 The limits of thought in Kant and Hegel
58) Kant says that although we cannot (literally) make sense of the claim
that one noumenon, such as God, caused the world (another), or some
thing in it (a phenomenon), we can regard things as if He did
(A672 B700 ff.). This is not a happy way of putting it. If the claim cp
=
Conclusion
We have now explored all the ways to avoid contradiction that are, as far
as I can see, open to Kant. None of them is successful. The problem is
posed for anyone who holds that the Categories do not apply to noumena
and, at the same time, wants even to consider propositions about them.
There are therefore only two possible solutions.
The first is to give up all talk of noumena. How much of the Kantian
project would be left if one were to do this, I leave to Kantian scholars to
argue about; not much I suspect. The central idea of Transcendental
Idealism is that the things we normally take ourselves to be familiar
with are representations. The very notion of things of which these are
the representations - one kind of noumenon - therefore seems built into
Kant's very problematic. 1 1 Nor is this solution open to anyone else who
holds that there are - or even could be - objects that we do not perceive,
be they God, photons, black holes, or numbers.
The other possible solution is to ditch the claim that the Categories
apply only to phenomena. After all, we have seen only one reasonable
argument for this so far; and we noted in 5.3 that it is not beyond doubt.
This solution is, I think, correct. But it takes us out of the frying pan and
into the fire, as we shall see in the next chapter.
11
See the quotation from Kant given in 5 . 1 , and the second quotation in 5 . 5 .
6 Kant' s antinomies
Introduction
In this chapter, we will concern ourselves with the other limit of thought
in Kant, the limit of the iterable. This rears its head, and plays a central
role, in the Dialectic of Pure Reason. The concerns of the previous
chapter are lurking beneath the surface, however, as we shall see.
Reason makes this demand in accordance with the principle that if the conditioned
is given, the en tire sum of the conditions, and consequently the absolutely uncondi
tioned . . . is also given.
The resultant totality does not itself possess conditions of the appropriate
kind - or it would not be the totality of all such conditions. This is why
Kant calls it the unconditioned. It is therefore a noumenon (if it is any
thing at all): any phenomenon (which is a representation for Kant; see
5 . 1 ) must have conditions of the appropriate kind (space, time, and
cause), since these are constitutive of experience. 2
The unconditioneds are exactly the Transcendental Ideas. According to
Kant, there are three ways of totalising, corresponding to the three kinds
1 Bennett « 1 974), section 84) suggests that a condition of something is a condition of its
possi bili ty.
2 See, for example, A508 B536.
=
Kant's antinomies 95
If w e assume that the world has no beginning i n time, then u p t o every given
moment an eternity has elapsed, and there has passed away in the world an
infinite series of successive states of things. Now the infinity of a series consists
in the fact that it can never be completed through successive synthesis. It thus
follows that it is impossible for an infinite world-series to have passed away, and
that a beginning of the world is therefore a necessary condition of the world' s
existence.
Though this argument is better than that for the Thesis it is still highly
suspect. The first reason is that it appeals to the Principle of Sufficient
Reason (PSR) . As we noted in 2.9, this will not take the required weight.
Note, moreover, that even if the PSR is correct, the fact that nothing
physical was happening does not imply that there was no reason why the
cosmos started - at least if you believe in a God of a certain kind. The
reason might just be the intention of God to start the cosmos at that
instant. (God's intention does not have to be formed at that instant. ) One
may ask why God should have intended the cosmos to start then. To
which the answer might just be: if God was to start the cosmos, She had
to start it at some time; all times were equally good, so She chose that one
at random. As Buridan's ass discovered, choosing one of a number of
equally good alternatives might be a perfectly sufficient reason.
The second reason that the argument fails is that it assumes that the
topology of time is Euclidean, and, in particular, that before any time
there is another. This may, in fact, not be so. Nor is this a mere academic
possibility. The General Theory of Relativity assures us that the structure
of space/time and gravitational fields are mutually dependent, and if the
cosmos started as a singularity, then it may be that it makes no sense to
talk of either time or space prior to this singularity. There was no prior. 7
7 See, for example, Hawking ( 1 988), pp. 42ff.
Kant's antinomies 99
The cruxes of this argument are, first, the claim that it is always
possible to decompose in thought, and, second, that it is impossible to
decompose 'through and through'. The first seems reasonable enough: it
always seems logically possible that spatial parts could be arranged in
some other way. The second is reminiscent of Aristotle's argument to the
same end (2. 5), and is more problematic. Its correctness here depends on
what Kant means by 'substance ' . In the Schematism Kant's criterion for
being a substance is permanence in time (see 5.3), i.e., existence through
all time. If this is what he means I see no hope of pushing the argument
through. More plausibly, by substance, he means something that is
indestructible (and hence must exist through all time). Then the argu
ment looks a lot better. For anything that is made up of parts could
(logically) cease to exist; for the parts could come apart. Hence if all
parts of the object have parts, all its parts could cease to exist, in
which case it, too, would cease to exist.
Whatever one makes of this argument in the end, modern science seems
to have finessed it, since it would now appear that matter is not a sub
stance in the required sense. The modern notion of matter is still, per
haps, a moot one (especially when we take quantum theory into
account 1 0) . Yet it is clear that matter can be destroyed by turning it
into energy, and so is destructible. Matter/energy may well be con
served, but energy is not spatially extended, and so the spatial decom
position involved in the argument seems to get no grip on it at all. Hence
the argument fails.
The argument for the Antithesis is much more straightforward:
Suppose that there were a simple part. This would occupy space. But
all spaces are divisible. Hence this simple would be correspondingly
divisible. By definition, this is impossible. As Kant puts it (ibid.):
Assume that a composite thing (as substance) i s made u p o f simple parts. Since all
external relation, and therefore all composition of substance, is possible only in
space , a space must be made up of as many parts as are contained in the compo-
10
See Priest ( l 989a).
Kant's antinomies 101
site which occupies it. Space, however, is not made up o f simple parts, but of
spaces . Every part of the composite must therefore occupy a space. But the
absolutely first parts of every composite are simple. The simple therefore occu
pies a space . Now since everything real, which occupies a space, contains in itself
a manifold of constituents external to one another, and is therefore composite;
and since a real composition is not made up of accidents (for accidents could not
exist outside one another, in the absence of substance) but of substances, it
follows that the simple would be a composite of substances - which is se1f
contradictory.
Let us assume that there is no other causality than that in accordance with the
laws of nature. This being so, everything which takes place presupposes a preced
ing state upon which it inevitably follows according to a rule. But the preceding
state must itself be something which has taken place (having come to be in a time
in which it previously was not); for if it had always existed, its consequence also
would have always existed, and would not have only just arisen . The causality of
the cause through which something takes place is itself, therefore, something that
has taken place which again presupposes, in accordance with the law of nature, a
preceding state and its causality, and this in a similar manner a still earlier state,
and so on. If, therefore, everything takes place solely in accordance with laws of
nature, there will always be only a relative, and never a first beginning, and
consequently, no completeness of the series on the side of the causes that arise
the one from the other. But the law of nature is just this, that nothing takes place
without a cause sufficiently determining a priori. The proposition that no caus
ality is possible save in accordance with laws of nature, when taken in unlimited
universality, is therefore self-contradictory; and this cannot, therefore, be
II
regarded as the sole kind of causality.
II
There is then another paragraph which explains that the first member must arise other
than through natural causation. But this is not relevant here.
Kant's antinomies 103
Another reason the argument fails is that it does not establish that the
event has no sufficient reason; merely that it has no sufficient reason in
the chain of causes. But as Leibniz noted, this leaves open the possibility
that the chain itself, and hence each event in it, has a sufficient reason .
The argument for the Antithesis is rather different, and goes essentially
as follows: if the chain has a first member, then this has no causal
determination, which is impossible by the 'Law of Causality': every
event has a cause. As Kant puts it (ibid.):
As usual, the argument fails. This is simply due to the failure of the 'Law
of Causality' . (See 5.4.)
12
There is another paragraph explaining that there is therefore no 'causation of freedom'.
But that is not relevant here.
1 3 See Kemp Smith ( 1 923), pp. 49 5ff. ; Bennett ( 1 9 74), p. 24 1 .
1 04 The limits of thought in Kant and Hegel
I f we assume that the world itself i s necessary, o r that a necessary being exists in
it, there are then two alternatives. Either there is a beginning in the series of
alterations which is absolutely necessary, and therefore without a cause, or the
series itself is without any beginning, and although contingent and conditioned in
all its parts none the less, as a whole, is absolutely necessary and unconditioned.
The former alternative, however, conflicts with the dynamical law of the deter
mination of all appearances in time; and the latter alternative contradicts itself
since the existence of a series cannot be necessary if no single member of it is
necessary . 1 4
14 There is another paragraph of the argument where Kant argues, in effect, that there are
no other candidates for a necessary being. This is irrelevant here.
Kant's antinomies 105
15 The passage omitted argues that the necessary being must be in the world in a certain
sense. This is not relevant here.
16 If this exists it must be a distinct object. It cannot be the sequence itself: parts of this
depend for their existence on other parts, but the whole does not depend on the whole, as
Leibniz, in effect, pointed out (see the first quotation in 2.8).
1 06 The limits of thought in Kant and Hegel
the generator can be no further applied, it follows that the limit of (J" (that
is, its last member if it has one, or it itself, if not) has no ground, i.e. , is
necessary.
is essentially that of Aristotle: deny that the limit exists. (Thus, the
infinities in question are only potential infinities, in Aristotle's terms.)
Like all things in Kant, however, it has knobs on.
Table 3
Kant claims that if the objects in the series of conditions were noumena
then the limit would exist (for example, A498 B526 f.). This appears to
=
have been so obvious to him that he never bothered to argue for it. ( Why,
I am not sure. Possibly, he thought it evident that in the cases we are
dealing with the infinity must be completed.) But in any case the objects
in the regress are phenomena; and this means that the limit does not exist.
The reductio arguments can be taken to demonstrate this. Since they are
perfectly sound, it follows that both the Thesis and Antithesis are false,
but the only way that statements of the form 'S is P' and 'S is not P' can
both be false is if S does not exist (A506 B53 1 ff.) .
=
As best I can understand it, it goes as follows. If the limit exists it is either
a phenomenon or a noumenon. But it is clearly not a phenomenon, as we
have already seen (in 6. 1 ) . 1 8 On the other hand, it cannot be a noumenon
either, for noumena exist independently of all experience; yet the limit of
all experiences of a certain kind obviously does not exist independently of
experience. Such an idea is self-contradictory according to Kant. Hence
the limit does not exist.
The second reason as to why the Antinomic arguments do not establish
true contradictions, Solution 2, is rather different, and is supposed to
apply only to the Third and Fourth Antinomies. At the risk of spoiling
the over-all pattern, Kant produces this solution because Solution 1 is
inconsistent with other fish that he has to fry. In the Fourth Antinomy
the limit object, a necessary being, is (what else?) God; and Kant does not
want to deny His existence. Similarly, when tracing back the chain of
causes in the Third Antinomy, Kant wants it to be possible that one
should come to something which initiates it, and so make room for
free will. Thus, he cannot endorse the claim that the Theses are false.
Solution 2 also turns on the distinction between phenomena and nou
mena. As we noted in 5 . 6 Kant distinguishes between phenomenal causa
tion and noumenal causation; the latter he calls, rather misleadingly,
'freedom' (A532 B 560, A538 B566). In phenomenal causation a phe
= =
18 This is not quite right: in the Fourth Antinomy the limit could be the last member of the
sequence, and, as such, unproblematically a phenomenon; but presumably Kant would
rule out this possibility on the ground of the Law of Causality.
108 The limits of thought in Kant and Hegel
6.8 Evaluation
The next question is how good Kant's resolution of the contradiction is in
his own terms. The simpler solution to deal with is Solution 2; so let us
take that first. Consider, first, solution 2p. Since Kant nowhere explains
where the arguments break down on this disambiguation, and nor is this
obvious, there is, in a sense, no solution to evaluate here: the contra
diction is unresolved. What solution 2n is, on the other hand, is clear. But
is it not clear that it works. The arguments for the Antitheses in the Third
and Fourth Antinomies appeal to the 'Law of Causality' to establish that
the generator can always be applied again. Now, the Law of Causality
does not apply to the noumenal generator, it is true; but the PSR, to
which Kant subscribes, does. Every noumenon must have its (noumenal)
reason; hence the generator can always be iterated, and the solution
fails ?O
So let us turn to Solution 1 , according to which the limit does not exist.
First, let us deal with Kant's independent argument for the non-existence
of the limit. The argument for this, as I have interpreted it, hinges on the
claim that the limit must be dependent on experience, since each of the
objects generating it is. This is simply a fallacy of composition of sorts.
Limits often do not share the properties of the things of which they are
the limits. For example, a limit of finite numbers does not have to be
finite, and a limit of rational numbers does not have to be rational. In this
case the limit is a noumenon; as such it is an object of thought, not
experience, and so its existence does not depend on experience. It does
not share this property with the things of which it is a limit, and the
argument collapses.
But we still have the reductio argument against the existence of the limit
(at least in the case of the first three Antinomies) . Clearly this depends on
the claim that there are no truths of the appropriate kind about non
existent objects. This is obviously a debatable point, but fortunately there
is no need to debate it here. For the Solution fails: if the arguments are, as
19 This is the argument I referred to in 5 . 3 . Since the Antinomic arguments fail anyway, it
would seem to be no better than the other two arguments discussed there.
20
It might be argued that the PSR cannot be applied to noumenal states of affairs since
these might be 10gicaJly necessary, but it is unlikely that Kant would have endorsed this.
For one reason: the main candidate for a necessary state of affairs is God's existence, but
the Critique is famous for its demolition of the Ontological Argument (A592 B620 ff.),
=
Kant maintains, sound then they lead to contradiction anyway. For the
reductio arguments in the Theses (with the possible exception of the First)
never actually depend on the existence of the limit (the completed infi
nity)! but only on the existence of the indefinite regress (the potential
infinity), which is entailed by the non-existence of the limit (and to which
Kant is committed anyway). For example, suppose that the limit in the
Second Antinomy, the simple, does not exist. Then certainly matter is not
composed of simples; but this is precisely the supposition that Kant has
reduced to absurdity in the argument for the Thesis. Similarly, suppose
that in the Third Antinomy, the limit does not exist; then there is a chain
of causes stretching back to infinity; and this is precisely the supposition
that Kant has reduced to absurdity in the argument for the Thesis. Hence
Kant's Solution 1 does not work either.
formal representation will become evident in due course.) Start with any
object, say, the Critique of Pure Reason, and apply the generator itera
tively, to produce, at each stage, the thought of the previous object: the
Critique; the thought of the Critique; the thought of the thought of the
Critique. And when we have an unbounded sequence of thoughts we next
produce the thought of all of them, and keep going. (Clearly, we never
produce a self-referential thought in this process since the object of
thought is 'pre-defined' in each case.) Let this procedure be performed
as often as possible (and so, presumably, into the transfinite). Consider
the totality of all thoughts generated in this way, T. Clearly, T has no last
member, since for any object there is a thought of it (though maybe it is
so complex that no person could actually entertain it). And ex hypothesi,
the generator can be no further applied to T. So T cannot be thought of
(Transcendence) . But you can think of T: you have just done so
(Closure). Contradiction. 2 1
For future reference, and with apologies to Kant, I will call this the
Fifth Antinomy. It is exactly of the Kantian form. For we have an
infinity generator applied to a certain object, together with arguments
that the limit produced in this way has inconsistent properties. This can
be illustrated as in Table 4, which is a fifth row for Table 3 in 6 . 7 .
The contradiction can b e put i n a slightly different but equivalent form.
This will turn out to be useful in later chapters. The equivalent contra
diction is that the thought of T both is (Closure) and is not
(Transcendence) a member of T. The argument for equivalence goes as
follows. Suppose that one can and cannot think of T. T is the set of all
things generated by applying the generator as far as possible. Since one
can think of T, the thought of T is just such a thing. Hence it is in T. But
the thought of T is distinct from any thought in T (since T is distinct from
any of its members); hence the thought of T is not in T (Transcendence).
Conversely, suppose that the thought of T both is and is not in T. Since
the thought of T is in T, one can think of T. But if one could think of T
Table 4
21 Actually, the claim that T has no last member is inessential to the argument. If, for some
reason, one takes T to have a last member, we can run exactly the same argument with
respect to that.
1 12 The limits of thought in Kant and Hegel
Conclusion
I will return to possibilities for defusing contradictions of this kind when
we have further examples of them. For the present, I just note that Kant's
claim that antinomy is inherent in aspects of our reasoning about the
limits of thought, is quite correct. For what to make of this, we turn to
Hegel.
7 Hegel' s infinities
Introduction
In this chapter we move from Kant to Hegel. As we saw in the last
chapter, Kant argued that contradictions are, in some sense, inherent
in our thinking about generated infinities. Despite this, he drew back
from a dialetheic conclusion: employing the distinction between phenom
ena and noumena he argued that the Antinomic arguments do not, in
fact, establish contradictory conclusions . Hegel criticised the ground of
the distinction between phenomena and noumena - with predictable
consequences: the infinities of the Antinomies are contradictory. In this
chapter we will first look at this matter, and then move on to look at
Hegel's own positive account of the nature of infinity. Un surprisingly, the
account explicitly incorporates the view that generated infinities are truly
contradictory.
Hegel also observes (as we did in 5 . 5) that Kant's very claim that we
cannot make judgments about noumena is self-inconsistent (Lesser Logic,
section 60, p. 9 1 ):
It argues a n utter want of consistency t o say, on the one hand, that understanding
only knows phenomena, and, on the other, assert the absolute character of this
knowledge, by statements such as 'Cognition can go no further' . . . No one
knows, or even feels , that anything is a limit or a defect until he is at the same
time above and beyond it.
the essence of the mind, it does seem strange to hear how calmly and confidently
the modest dogma has been advanced by one and repeated by o thers, that
thought or Reason, and not the World, is the seat of contradiction.
2 There is one exception: there are four sub-categories of judgment. This is somewhat
ironical, since these sub-categories - or at least their sub-categories - are essentially
Kant's Categories .
1 16 The limits of thought in Kant and Hegel
' Cx = ' ,Cx entails CXI\,CX. For details, see Priest ( l 9 89b).
6 Discussions can be found in numerous commentators, for example Findlay ( 1 958) or
Inwood ( 1 983).
7 Taylor «( 1 975), p. 1 1 4) calls it 'off-beat'.
Hegel's infinities 1 17
Something is infinite (in the false sense) if it is not finite. The finite and
the false infinite are therefore mutually complementary and bounding
concepts.
Hegel calls this notion of the infinite false, and objects to it on the
ground that it is not really infinite, since it does not manage to 'free itself
from the finite' . He gives two distinct arguments for this. The first is that
since it is bounded by something (viz. , the finite), it is, itself, finite. As
Hegel puts it in the Logic (pp. 1 39f. 8 ):
This contradiction [between the finite and the false infinite] occurs a s a direct
result of the circumstance that the finite remains as a determinate being opposed
to the infinite, so that there are two determinatenesses; there are two worlds, one
infinite and one finite, and in their relationship the infinite is only the limit of the
finite and is thus only a determinate infinite, an infinite which is itself finite .
The point that is being made here, when stripped of all its Hegelean
obscurity, is a simple one: a potential infinity, at any stage of its existence,
is, after all, never more than finite. Hence it is not truly infinite. The point
is well made. The notion of the potential infinite does not deliver an
understanding of the way the infinite behaves; it delivers an understand
ing of the way that certain finitudes behave. One can deny, as did
Aristotle (but Hegel most certainly does not, as we shall see), that the
actual infinite exists. But the pertinent point here remains: What is it,
then, of which one is denying the existence? It is the true infinite.
Hegel's infinities 1 19
The true infinite is the notion of an object whose finitude is its infinitude,
and which is therefore both finite and infinite.
It is probable that the meaning of this will be quite opaque to the
reader; a simple example can help to make it much clearer. The example
is the Fifth Antinomy of 6 . 9 . Recall that T is the totality of all thoughts
generated by starting with some object and applying the generator
thought of x as far as possible. T bounds all thoughts generated, but is
transcendable by an application of the generator. It therefore has exactly
the properties of the Hegelean true infinite. First, it is bounded (on the
9 For a brief summary of the section, see Findlay ( 1 958), pp. l 74f.
10
Note that modern analysis actually defines the infinite sum as this limit. But Hegel is
writing before Cauchy and Weierstrass and has a more geometric understanding of the
infinite sum.
1 20 The limits of thought in Kant and Hegel
The thesis and antithesis and their proofs therefore represent nothing but the
opposite assertions, that a limit is, and that the limit equally is only a sublated
one; that the limit has a beyond, with which however it stands in relation, and
beyond which it must pass, but that in doing so there arises another such limit,
which is no limit.
great stress is laid on the limitations of thought, of reason , and so on, and it is
asserted that the limitation canno t be transcended . To make such an assertion is
to be unaware that the very fact that something is determined as a limitation
implies that the limitation is already transcended . For a determinateness, a limit ,
is determined as a limitation only in opposition to i t s other in general, that is, in
opposition to that which is free from the limitation; the other of a limitation is
precisely the being beyond it.
fall asunder, the result is that some becomes other, and this other is itself a
somewhat, which then as such changes likewise, and so on ad infinitum. The
result seems to superficial reflection very grand, the grandest possible . But such
a progression is not the real infinite . This consists in being at home with itself in
its other, or if enunciated as a process, in coming to itself in its other. Much
depends on rightly apprehending the notion of infinity, and not stopping short at
the wrong infinity of endless progression.
The true infinite is the coming together of the two moments in the shape
of an absolute totality, which yet can be broken out of; a limit which,
none the less, can be transcended; a bounded unbounded.
Hegel's infinities 121
Conclusion
A s we have seen, not only does Hegel observe that certain kinds o f limits
behave in a contradictory fashion; he actually fashions a contradictory
category to think them. This bold piece of philosophy brings to a close
this part of the book. The contradictory nature of the limits of thought,
of which numerous philosophers before Kant had had intimations, finally
found a realisation in Kant's work, though his 'tenderness for things in
the world' finally obscured this from him. It was left to Hegel to bring
this realisation to its conclusion .
This is hardly an end of the story. Galileo is justly famous for (amongst
other things) his recognition and formulation of the notion of inertia. In
his work, the notion is a fairly rudimentary one, however, at least in
comparison with the way in which it was articulated and elaborated by
subsequent generations of physicists, starting with Newton. So it is with
Hegel's theory of the infinite. Though essentially correct, the theory is
rudimentary and incomplete in many ways.
Some examples: Hegel had a very limited understanding of limiting
processes. He had no idea, for example, of transfinite sequences.
(Though this is not as important as one might think in the present
context; it affects only the question of how 'big' infinity is.) Next,
Hegel had only a rudimentary understanding of the boundary-tearing
mechanism which transcends limits. Though he had the notion of a
boundary being transcended, or aufgehoben, he could give
(understandably, given the historical context) no detailed account of
mechanisms (such as diagonalisation) by which this occurs. Thirdly,
Hegel's infinite is a special case of a more general structure. Hegel had
no conception of this, or of the many areas in which this structure would
turn up (or of the lengths to which people would go to maintain that it
does not!). Many of these details were worked out in subsequent devel
opments in the foundations of set-theory. In the third part of the book we
will look at these.
Part 3
Introduction
In the last chapter we looked at Hegel's crucial ideas on infinity. Hegel's
philosophy in general exercised a profound fascination and influence on
succeeding philosophers. For some reason, however, his ideas on the
infinite were quietly forgotten. Further developments had to wait nearly
1 00 years to come; and when they did come, they came from a very
different direction, one which might well have surprised Hegel. This
was the work of Cantor on the foundations of set-theory.
Cantor was not a philosopher in the way that all the people we have so
far met were. However, his contribution to our understanding of the
infinite was, perhaps, greater than any other person before or since;
almost single-handedly he found an intricate and beautiful form in an
area that had hitherto been thought formless. In this chapter we will
consider his work.
I will not be concerned with all of Cantor's work on the infinite: only
that part which is relevant to the theme of this book. I will start by
looking at his notion of transfinite sequences. Next we will turn to the
subject of diagonalisation. We will then look at the paradoxes of absolute
infinity and Cantor's response to them. Finally, in preparation for a later
chapter, we will look at his remarks on potential and actual infinity.
In this chapter and the others of this part, the material is more tech
nical than in other chapters. I do not, in general, intend to go into
technical proofs where these are standard. There are numerous text
books which can be consulted. ! I will, however, give enough technical
detail for those unfamiliar with the area to be able to follow the discus
sion; and, in as far as I can, do this in a way that is intelligible to non
logicians.
I For example, the excellent texts Fraenkel, Bar-Hillel, and Levy ( 1 973), and Bell and
Machover ( 1 977).
1 26 Limits and the paradoxes of self-reference
°
0, 1
0, 1 , 2
0, 1 , 2, 3
and still we can keep going with a sequence which has another member
after all these:
° . . w . . . 2w . . . 3w . . . w
.
2
To see what this means, note that a typical transfinite quantity is one of
the (non-finite) ordinals, generated as in 8 . 1 . Recall that for any such
ordinal there is a greater. The totality of ordinals generated in this way is,
on the other hand, an absolute infinite; this can be made no greater. As
Cantor put it on another occasion (Hallett ( 1 984), p. 44):
The transfinite with its plenitude of formations and forms necessarily indicates an
Absolute, a 'true infinite' whose magnitude is capable of no increase or diminu
tion, and is therefore to be looked upon quantitatively as an absolute maximum.
As is clear from this quotation, the Absolute is just the infinity that had
concerned previous generations of phi losophers . The transfinite is an
important middle ground coming between the finite, properly so called,
and the Absolute. The existence of this was a highly important discovery;
but as far as debates about the existence and nature of the infinite go, an
absolute irrelevancy. 6
What did Cantor have to say about absolute infinity? For a start, in
private, he was wont to wax lyrical about it (Dauben ( 1 979), p . 1 3) :
What surpasses all that i s finite and transfinite . . . i s the single completely indi
vidual unity in which everything is included, which includes the 'Absolute' incom
prehensible to the human understanding. This is the 'Actus Purissimus' which by
many is called 'God'.
6 The similarity between the traditional infinite and the Cantorian Absolute is noted by
Maddy ( 1 983). Moore « 1 989), pp. 1 98f.) also notes that ancient problems of the infinite
replay themselves in the Absolute.
Absolute infinity 1 29
of the Absolute. The Absolute can only be recognised, never known, not even
approximately.
8.3 Diagonalisation
This is diagonalisation. First, let me set the scene. Given that it makes
sense to extend the ordinal numbers (i.e., those that index order) into the
transfinite, the next question is whether it makes sense to extend, simi
larly, the cardinal numbers (i.e., those that index magnitude). To do this
we need a criterion of size comparison. Cantor and, at much the same
time, Frege, suggested that two sets, x and y, are the same cardinal size
(x � y) iff their members can be paired off one for one. That is, there is a
function which maps each member of x to a member of y; and each
member of y is mapped to by a unique member of x. Next, a set x is at
least as big as y (x 2:: cY ) iff there is a subset of x which is the same size as
y. (The subscript 'c' is there to remind that the ordering is cardinality
ordering, which is different from the ordering of ordinals.) x is bigger than
y (x 2:: cY) iff x > cy and it is not the case that x � y. It is not difficult to see
that these definitions of size capture our notion of cardinal size for finite
collections as a special case; one can also show that they have all the
properties one would want of a size-ordering. 9
Clearly, there are finite sets of increasing size. Also, clearly, the collec
tion of natural numbers (w) is larger than any finite set. A set is called
denumerable (or countably infinite) if it is the same size as w . In an
ingenious proof Cantor showed that the collection of all real numbers
is not denumerable (Hallett ( 1 984), pp . 75f.). It quickly follows that it is
bigger than w . The next question is whether there is a set of largest size. In
a generalisation of his earlier proof, Cantor showed that there is not. For
any collection, there is a bigger collection. This is now called Cantor 's
Theorem.
8 Hallett himself (( 1 984) p. 47) notes that this sort of situation is one in which 'there is
certainly some tension'. This is rather an understatement.
9 The ordering is easily seen to be transitive. It is more difficult to show that it is a partial
ordering. (This is essentially the content of the Schroder-Bernstein Theorem). Using the
Axiom of Choice, one can show that it is a total ordering (that is, that any two sets can be
compared in size), though Cantor himself was never able to establish this fact satisfacto
rily.
Absolute infinity 131
Table 5
Xo T F T F
Xl F F T F
X2 T T F T
X", F F F F
1 32 Limits and the paradoxes of self-reference
Xl ¢: f(Xl ) , SO set Xl in z; this ensures that z is not f(Xl ) ' And so on. Hence
we ensure that z is not on the list at all; but since it is, ex hypothesi, we
have a contradiction.
The essence of Cantor's proof is as follows. Given a list of objects of a
certain kind (in this case, subsets of x), we have a construction which
defines a new object of this kind (in this case z ) , by systematically destroy
ing the possibility of its identity with each object on the list. The new
object may be said to 'diagonalise out' of the list.
8.4 Paradox
We may now return to the subject of paradox. The contradiction at the
heart of the Absolute was bound to surface sooner or later. In fact, it was
sooner. It did so in a number of paradoxes that were discovered. The first
is entirely implicit in the discussion of 8 .2. Consider the collection of all
ordinals, On. By construction, On is an ordinal; the least ordinal greater
than all the ordinals. Hence it is not an ordinal. This paradox was
published by Burali-Forti in ( 1 897), though apparently Cantor himself
discovered it at least a year before this, 1O which is hardly surprising in the
context.
One further example of the contradictions that came to light will do for
the time being. This is a version of a paradox discovered by
Mirimanoff. 1 1 Starting with the empty set, we generate a transfinite
sequence by applying the power-set at successor stages and collecting
up at limit stages; that is, at limit stages we form the union of all pre
vious stages. Let us call the union of all the sets in the sequence R.
In modern set-theory R is called the cumulative hierarchy (since every
member of the sequence can be shown to be a proper subset of any later
member), and the stages of construction are called ranks. R has some
important properties. First: anything in it is also a subset of it. (Since
its members have lower rank.) Next, call any sequence (finite or
infinite) Xo, X l . X2 . . . such that . . . X2 E Xl E Xo a regress from Xo. Call
a set, x, wellfounded if there is no infinite regress from x. The members of
the cumulative hierarchy are all well founded; as a matter of fact, the
cumulative hierarchy comprises all and only the well-founded sets.
Now consider R. Since all its members are well founded, it, too, is well
founded. Hence it is in R. But then it is not well founded since the
following is an infinite regress: . . . R E R E R E R. Hence it is not in R.
IO
See Hallett ( 1 984), p. 74.
II
Ibid. , section 4.4.
Absolute infinity 1 33
Table 6
12
See Dedekind ( 1 888), Theorem 66.
1 34 Limits and the paradoxes of self-reference
the next member after any sequence, x (whether or not it has a last
member), by forming the least ordinal greater than all the members of
x. I will write this as log (x) . For R, it is not difficult to show that the next
member after the sequence, x, can be generated by applying the operator
U{P(y);y E x} . I will write this as UP(x) . For T the sequence can be
generated by always applying the operator: thought of the last member
of x, if it has one; or the thought of x otherwise. I will write this as t(x).
This explains the bracketed entries in the Generator column of Table 6.
The similarities between the Kantian Antinomies and the paradoxes of
absolute infinity have been noted by a number of people, including
Hessenberg, Zermelo and Fraenkel. 1 3 Perhaps the clearest statement is
given by Martin. Speaking of the dialectic between defining from below
and conceiving as a unity from above, he says (( 1 9 55), p. 55):
This conflict between concluding and beginning anew, between forming a totality
and using this totality as a new element, is the actual ground of the [set-theoretic]
antinomy. It is this conflict that gives the connection with the Kantian anti
nomies. Kant saw quite clearly that the antinomies rest on this antithesis between
making a conclusion and going beyond the conclusion. In principle, this had
already been seen by Archytas, when he wanted to go to the end of the world
and stretch out his arm.
of identifying an ordinal with the set of all previous ordinals). Why this is
supposed to follow is unclear.
Light is thrown on the matter by a remark in a later letter to Jourdain,
where he says (Hallett ( 1 984), p. 286) :
an inconsistent multiplicity, because it cannot be understood as one whole and
thus cannot be considered as one thing, cannot be taken as an element of a
multiplicity.
16
For example, van Heijenoort in his introduction to Cantor (1 899); Hallett ( 1 9 84), section
8.3.
Absolute infinity 1 37
The situation with (2) is somewhat different. It says that given some
unending progression (of ordinals) there is a set containing them. This is
an exceptionally important principle; but there is some problem about
stating it. Since the success of set-theoretic reductionism this century, the
natural way of understanding the notion of an unending progression is
exactly as a series, i.e., a set of a certain kind. On this understanding the
principle is tautologous . But this is obviously not its intent.
If the sequence in question is not to be understood as an already
completed infinity (a set), then it must be understood as something in a
state of generation, i.e., a potential infinite. And to speak of the set
containing the members of the progression is exactly to speak of the
(actual) infinity thus generated. What this principle comes to, therefore,
is this:
Following Hallett (( 1 984), p. 7) let us call this the Domain Principle. I take
it to be a formulation of the Kantian insight that totalisation is concep
tually unavoidable (6. 1 ), though stated in a much more satisfactory way
than Kant ever managed to achieve.
But what, exactly, is a potential infinity? The idea is, intuitively, clear
enough. ( We have been working with it since chapter 2.) Moreover,
traditionally, it has been thought unproblematic; it is the other member
of the pair, the actual infinite, that has been thought problematic. It is
therefore somewhat ironic that in modern mathematics the notion of the
actual infinite is relatively well understood (any infinite set is an actual
infinity); it is a precise analysis of the nature of potential infinity that still
awaits us (as Hart ( 1 976) observes) . I shall not try to offer one here; 1 7 it
will suffice, for our purpose, to leave the notion at an intuitive level .
Whatever else it is, a potential infinity is some kind of variable quan
tity, whose variation can go beyond any preassigned bound of a certain
kind. As Cantor puts it (Hallett ( 1 984), p. 1 2) :
17 Some suggestions employing possible worlds of increasing domain can be found in 1 1 .2.
Hart (1 975--6) offers two precise ways of understanding the notion. One is very similar to
that suggested in 1 1 .2. The other is simply to identify the notion with that of a proper
class (which we will also meet in chapter 1 1 ) . I am dubious about this suggestion since it
entails that no potential infinity has a corresponding actual infinity, which seems far too
strong to be an adequate analysis of the traditional notion. It must be possible, for
example, to conceptualise the natural numbers qua potential infinity.
1 38 Limits and the paradoxes of self-reference
at least one value . ' This is true if z may be complex; false if z must be real.
Now consider some statement about a potentially infinite variable; by
the observation, if this has determinate sense, its domain of variability
must be determinate. Can this domain not itself be a potential infinity
(variable)? Perhaps it can; but then the sense of the original claim is not
determinate unless the domain of this variable is itself determinate. Could
that, itself, be a variable? Perhaps. We are obviously off on an infinite
regress. Moreover, this regress is vicious. If it went on forever, the sense
of the original statement would never become determinate. Given that the
statement does have determinate sense, the regress must bottom-out
somewhere. And thus the original variable must find its ultimate ground
in an actual infinite domain.
Cantor's argument can be put in contemporary form (bypassing the
issue of what, exactly, a potential infinity is) by a consideration of mod-
Absolute infinity 1 39
there is a proof of some statement of the form aCt), together with a proof that the
object denoted by t belongs to the domain [of quantification] .
Conclusion
Returning to the main theme of this chapter: as we have seen, Cantor's
theory of the transfinite did not succeed in removing the paradoxes of the
infinite. It merely relocated them. This should not be held against him,
however. The contradictions are there, inherent in the object of discourse;
so getting rid of them is hardly a sign of an adequate account! The great
virtues of Cantor's work from the present perspective are twofold. First,
it made it clear with an unprecedented rigour that, and where, the contra
dictions occur. This rigour, moreover, narrowed down the field of pos
sible solutions to a very small and surveyable class . Secondly, the notion
of diagonalisation provided the corner-stone of an adequate understand
ing of boundary-transcendence. We will take up these issues in the sub
sequent chapters of this part.
9 Vicious circles
Introduction
The paradoxes of absolute infinity that we met in the last chapter (8 .4)
were but the leading edge of a number of paradoxes that turned up in set
theory at the turn of the century. No one investigated these more thor
oughly than Bertrand Russell, and in this chapter 1 will discuss several of
the things he had to say about them. As we will see, he found the key that
holds the family of paradoxes together (or almost, anyway); we will also
have a look at his proposed solution, which was less successful.
this case there is a one-to-one mapping between V and P(V), viz. , the
identity map (the map that maps everything to itself). But as we saw in
8 . 3 , Cantor showed that for any set, x, there can be no such mapping
between x and P(x). This contradiction is now called ' Cantor's paradox' .
Cantor did, indeed, discover it, though he never published it officially?
Cantor's paradox can be stripped down to its essence. As we saw in 8 . 3 ,
Cantor's proof that there i s n o one-to-one correspondence between x and
1 It is traditional wisdom that V is just the cumulative hierarchy, R (8.4). This is called the
Axiom of Foundation. However, it is quite possible for there to be pure non-well
founded sets. See Aczel (1 988).
2 See Fraenkel, Bar-Hillel, and Levy (1 973), p. 7.
1 42 Limits and the paradoxes of self-reference
Condition ( 1 ) I will call Existence. 6 Conditions (2a) and (2b) are clearly
versions of Transcendence and Closure, respectively. When this schema
(or the generalisation of it that we will meet later in the chapter) is at
issue, I will use these names for these versions. 7 I will call any function, 0,
that satisfies the schema a diagonaliser (with respect to <p). A diagonaliser
need not be defined literally by diagonalisation; but, as we shall see, it is
always defined systematically to ensure that the result of applying it to
any set cannot be identical with any member of that set. Diagonalisation
proper is the paradigm of such a procedure. A diagonaliser can be
thought of as a transfinite generalisation of what I have so far called a
generator. We can define a transfinite sequence of members of 0,
( wa ; a E On) by repeated application of the diagonaliser (transfinite
recursion), thus: Wa = o( { w/J; ,8 < a } ) . (If a = 0, the argument of the
function is just ¢.)
Let us now see how the paradoxes fit this schema. In the case of
Russell's paradox, the property <p(y) is just Y E V, and so 0 is V. The
function o(x) is just Px ' Transcendence and Closure obtain, as I observed
in the previous section; and the contradiction is that Pv E V! 8
For Burali-Forti's Paradox, <p(y) is 'y is an ordinal' and the diagona
liser, o(x) , is log(x). Transcendence holds by definition, as does Closure.
In this case, the collection 0 is just On, as is 0(0) (by the von Neumann
convention of 8 . 1 ) . Hence the contradiction is that On E On! .
I n Mirimanoffs Paradox <p(y) i s ' y i s well founded' and the diagona
liser, o(x), is UP(x) . Transcendence and Closure hold simply in virtue of x
being well founded. The collection 0 is the cumulative hierarchy, R.
Moreover, it is not difficult to see that oCR) is just R itself. Hence, the
contradiction in this case is that R E R ! .
Although the Fifth Antinomy is not a standard set-theoretic paradox,
it fits the same pattern. <p(y) is y E T, so 0 is T. o(x) is just t(x). The
argument for Transcendence is as follows.
6 Meinongians who take sets to be non-existent objects are invited to read 'existence' as
'being', here and throughout; similarly with their cognates.
7 In case the inferences used to establish Transcendence and Closure are not always
dialetheically valid, we may define the conditions more cautiously. Whenever a is dedu
cible classically there is some formula, {3, such that a V {3! is dialetheically deducible. (See
Priest (I 987), Theorem 0, p. 1 49.) Hence we can formulate Transcendence and Closure as
8(x) 1- xV3pp! and 8(x) E 0 V 3pp!, respectively (or something similar using the truth
predicate if one does not believe in propositional quantification). The ultimate contra
diction is then 8(0) E O ! V 3pp !, or simply 3pp ! .
8 There i s a related way of fitting Russell's paradox into the schema. F o r <p(y) w e take
'y E V /\ Y 1- y', so that 0 is just py; the function 8(x) is simply the identity function id(x).
Transcendence and Closure hold, as may easily be shown, and the contradiction is now
that py E py!
1 44 Limits and the paradoxes of self-reference
First, note that if x, y are in T and x -=f=. y then t(x) -=f=. t(y) . For, since x
and y are distinct objects (as are their last members if they have one), one
can (at least in principle) entertain the thought of one without entertain
ing the thought of the other. These thoughts must therefore be distinct. 9
Hence any thought is of the form t(x) for a unique x. Call a thought t(xo)
unfounded if there is a sequence < Xi; i E w > such that t(Xi + I ) E Xi for all
i E w . Every thought in T is founded. The proof is by a (transfinite)
induction. This is obviously true of the first thought in the sequence
t({the Critique}). Suppose it is true for all members of the sequence x,
but that the next thought, t(x), is unfounded. Then there is a sequence:
x, Xo, X I . . . of the appropriate kind. But then t(xo) E X and t(xo) is
unfounded, which is impossible. It follows that if t(x) is any thought in
T t(x) � x (or else x, x, x . . . would be an infinite regress).
The argument for Closure is simpler: T is the sequence of thoughts
generated by applications of t to any sequence of objects generated; T is a
sequence of objects generated; so t(T) E T. The contradiction is that t(T)
both is and is not in T.
We can tabulate the way the paradoxes fit the schema as in Table 7.
(Compare this with Table 6 of 8 . 5.)
Table 7
Russell Px y EV V Pv E V!
Burali-Forti log (x) y is an ordinal On On E On!
MirimanofT UP(x) y is well founded R R E R!
5th Antinomy t(x) y ET T t (T) E T!
9.3 Definability
The paradoxes of absolute infinity were not the only paradoxes of self
reference that turned up in the period in question. Another important
9 Moreover, t(x) is a thought of x (or its last member) and t(y) isn't.
Vicious circles 145
10
He produced the construction as part of a proof that the Continuum Hypothesis is false.
However, in virtue of the similarity of this part of the argument to recognised paradoxes,
the proof was never generally accepted.
11
After its formulator; see Russell ( 1 906), pp. 1 28f.
12
For further details, see Richard ( 1 905).
13
Complete formalisations of Berry's Paradox can be found in Priest ( 1 983) and ( 1 987), 1 . 8 .
1 46 Limits and the paradoxes of self-reference
IL
TC
TC 1\ ""TC
If the reader experiences a certain sense of deja vu at this point, this is
quite understandable. So far in this chapter - and in fact for the last three
chapters - we have been discussing only one of the kinds of limit that
have concerned us throughout the book: the limit of the iterable. Another
has now reappeared.
The above argument is essentially the same as Berkeley's Paradox (see
4.9), as a simple comparison will show. There are three differences, two
formal and one informal. The first formal difference is that Koenig's
argument uses a definite description operator ('the least ordinal such
that'), whilst Berkeley's uses an indefinite description operator. This is
a trivial difference of no moment. In the case of Koenig'S argument we
are simply making use of the (inessential) fact that ordinals are well
ordered. The second formal difference is that in Koenig's argument
there are independent sub-arguments (� and II) for the two axiomatic
premises of the Core Argument. This is not terribly important either: in
chapter 4 I gave informal arguments for these claims anyway. But it does
mean that 3X""TX is no longer even a target for reductio, as Berkeley
wished it to be, since it is no longer an independent premise.
The informal difference is that in Berkeley's paradox T is interpreted as
'is conceivable' ; whereas in Koenig's paradox it is interpreted as 'is
definable' . Now these notions may not be identical, but they would
seem to be pretty well co-extensional. For if there is a noun-phrase
that refers to something, that thing can be conceived, simply by bringing
the noun-phrase before the mind; and if something can be conceived,
there must be some way of singling it out, and hence some way of
referring to it (for example, with 'the object being thought about by.
GP at time, date').
Hence, the limit of the conceivable is essentially the limit of the defin
able, and Koenig's Paradox and Berkeley's Paradox are, to all intents and
purposes, the same. For the same reason, the definability paradoxes pro
vide further examples of contradictions at the limit of the conceivable. 14
14 And just as there is a variant of Berkeley's Paradox which uses the notion of conceiving,
rather than conceivability, so there are versions of the definability paradoxes which use
the notion of definition (or better, referring) rather than definability. Corresponding to
Koenig's Paradox, for example, is the paradox of the least ordinal that has never been
referred to; similarly for the other definability paradoxes.
Vicious circles 1 47
Given that these conditions are satisfied we still have a contradiction. For
since 'lj;(n) , we have 8(n)� n ! . I will call any n that satisfies these condi
tions (for an appropriate 8) an inclosure. 1 6 The conditions themselves, I
will call the Inclosure Schema, and any paradox of which this is the
underlying structure, an inclosure contradiction. The main difference
between the Inclosure Schema and Russell's is that in the Inclosure
Schema the diagonaliser is guaranteed to function only on a sub-family
of the power set of n, viz. , those sets that satisfy 'lj;.
Bearing this in mind, it is clear that Russell's schema proper is just a
special case of the Inclosure Schema, where the family is the power-set
itself, i.e., 'lj; is the universal property (AXX = x). Thus, the Inclosure
Schema still encompasses the paradoxes of absolute infinity. But by
choosing 'lj; appropriately, it also encompasses the definability para-
15
And conversely, the mere fact that a class of ordinals is definable does not imply that
every member is definable.
16
The English prefix 'in-' is ambiguous. It can have the sense of within, as in 'income' or,
indeed, 'inclose' ; this is often corrupted to 'en-' in later English. It can also have the sense
of a negative particle, as in 'incapacity' or 'indisposition'; later English prefers 'un-' .
1 48 Limits and the paradoxes of self-reference
Table 8
Koenig's /l-YY f/. x x in definable y E DOn DOn ( /l-YY f/. DOn) E DOn!
Berry's /l-YY f/. x x E DN90 Y E DN99 DN99 ( /l-YY f/. DN99) E DN99 !
Richard's diag(x) x is definable y E DR DR diag(DR) E DR!
Berkeley's e YY f/. x x is conceivable ry C eYY f/. C E C!
The issue is much more subtle than this, however. We require of the
Inclosure Schema not merely that it be a pattern into which the contra
dictions fit. Patterns are cheap enough: given any finite set of data (such
as the received paradoxes of self-reference) there is an infinite number of
Vicious circles 1 49
patterns that it can be made to fit. We want not just any old pattern, but
the essential pattern. How to cash out this notion here is an interesting
question. If we were dealing with a scientific, empirical, matter, what we
would be after would be a pattern that is a lawlike, not just accidental,
generalisation (however, in the end, one wishes to cash out the distinc
tion). But it is not clear (at least to me) how to transfer the distinction to
the present case. Still, for genuine satisfaction of the Schema we need the
fact that a contradiction fits the pattern to explain why the contradiction
arises. It is clear that in the example of the last paragraph this cannot be
so. For the fact that the contradiction arises is used to establish that the
pattern is satisfied. That the pattern is satisfied can hardly, therefore, be
used to explain why the contradiction arises. 1 7 This is certainly not to say
that every example of the Inclosure Schema where ,¢(x) is 'x = 0' is
pathological . But for legitimate cases we need, at the very least, some
understanding of why it is that, given the totality 0, 8(0) is able to 'lever
itself out' . Once one understands how it is that a diagonaliser manages to
operate on a totality of objects of a certain kind to produce a novel object
of the same kind, it becomes clear why a contradiction occurs at the limit.
This kind of explanatory value is present in the instances of the Inclosure
Schema that we have noted in the previous section, and in those which we
will go on to note. 1 8
17
In a similar way, one can show that, in a certain sense, any paradox that satisfies the
Inclosure Schema fits Russell's original schema. Given any diagonaliser that satisfies the
Inclosure Schema, 8, we define a new diagonaliser, 8 ' , as follows. If x is a subset of n:
8' (x) = 8(x) if 1jJ(x)
= 8(n - x) otherwise
where 8 is a choice function on the (non-empty) subsets of n. 8' can easily be seen to
satisfy Russell's schema. As is clear, however, 8 ' is a gerrymandered, grue-type entity.
The reduction therefore has little intrinsic value.
18
One way to get some handle on the issue formally might be to note that in the case of the
bona fide diagonalisers that we have met, there is a genuine functional dependence of the
value of the function on its argument: the argument is actually used in computing the
value of the function. This is clearly not the case with the pathological example we noted.
Unfortunatley, the issue of when the value of a function really depends on its arguments
is a tricky and unresolved one. For some discussion of the problem, see Anderson et al.
(1 992), section 70.
1 50 Limits and the paradoxes of self-reference
Russell ( 1 906); a fuller account was given in ( 1 908); which was more or
less simply incorporated in the introduction to Whitehead and Russell
( 1 9 1 0) (to the second edition of which all subsequent page references
refer).
According to Russell, if an object is defined in terms of some totality,
then, if that object were in the totality, we would have a vicious circle.
Hence this cannot arise. Russell sums up these thoughts in the Vicious
Circle Principle (VCP), which he states as follows ( p . 37): 1 9
19 Russell gave three formulations of the VCP, and these are not obviously equivalent (see
Goedel (1 944)). These subtleties are not relevant here.
20
For details, see Quine (1 963), ch. I I .
21
For full details, see Chihara (1 973), ch. I .
Vicious circles 151
22
For an account of the reduction, see Chihara ( 1 973), ch. 1 .
1 52 Limits and the paradoxes of self-reference
This may solve the problem formally, but raises the question of the
justification of the axiom, to which Russell never really produced a
convincing answer. This was not an end to the problems for this solu
tion, however. For even given the Axiom of Reducibility (and an axiom
of infinity), there is much set-theory that cannot be demonstrated given
the restrictions imposed by orders. For example, the existence of the
cardinal �w and, a fortiori, larger cardinals cannot be established.
I do not wish to discuss these issues further. 2 3 I put them here mainly
for the record . Russell's solution founders on a much less technical but
more fundamental problem. According to the theory, every variable must
range over one order of propositional functions. No variable can there
fore range over all propositional functions. For the same reason, no
variable can range over all propositions. This pushes many claims
beyond the limit of the expressible. Take, for example, the law of
excluded middle: every proposition is either true or false . Since this has
a quantifier over all propositions, it cannot be expressed. Or, closer to
home, consider the Axiom of Reducibility itself. This is supposed to hold
for all functions, f. Russell's very statement of it (above) therefore vio
lates the theory of orders. Even decent statements of the YCP cannot be
made without violating the YCP since they must say thatfor any function,
f, any propositional function which 'involves' f cannot be an argument
for f. Such statements are impossible by Russell's own admission.
To add insult to injury, the very theory of orders cannot be explained
without quantifying over all functions, and hence violating it. For to
explain it, one has to express the fact that every propositional function
has a determinate order. Hence, the theory is self-refuting. 24
Russell was well aware of at least some of these difficulties. To solve
them he formulated his theory of systematic ambiguity. This is most
easily understood with respect to an example. Take the statement of
the Axiom of Reducibility. According to Russell, this has to be under
stood (for fixed i) as indicating not a single proposition, but an infinitude
of propositions: one for each order that f might be. Russell, himself,
explains the point with respect to a similar example as follows ( p . 55):
In some cases, w e can see that some statement will hold of 'all nth-order proper
ties of a', whatever value n may have. In such cases, no practical harm results
from regarding the statement as being about "all properties of a" provided we
remember that it is really a number of statements, and not a single statement
23 Discussion of the various issues can be found in numerous places, for example, Chihara
( 1 973), ch. I ; Fraenke1, Bar-Hillel, and Levy ( 1 973), 3. 5; Quine ( 1 963), ch. I I .
24 As was well pointed out by Fitch ( 1 946).
Vicious circles 1 53
which could be regarded as assigning another property to a, over and above such
properties. Such cases will always involve some systematic ambiguity.
Conclusion
In this chapter we have seen that two of the limits of thought that have
concerned us in the book, the limit of iteration and the limit of concep
tion, are intimately related to the Inclosure Schema. One of the other
limits, the limit of expression, also appeared briefly under that guise. The
fourth limit that has concerned us, the limit of cognition, has been absent
from the discussion for a number of chapters now. In the next chapter it
will return. And as we will see there, it, too, is intimately connected with
the Inclosure Schema.
10 Parameterisation
Introduction
As we saw in chapter 9, the limits of iteration and conception (definition)
come together in the shape of paradoxes of self-reference; in particular,
they both instantiate the same form: the Inclosure Schema. An obvious
question is whether the Schema captures the other limits that we have
met. This is the first topic we will investigate in this chapter.
We will do this by looking at some of the ideas of Ramsey, whose essay
'The Foundations of Mathematics' ! provided the most acute early com
mentary on Russell's ideas on paradox. Ramsey suggested that the para
doxes of self-reference are of two kinds. I think it no understatement to
say that this suggestion has provided the framework for all subsequent
work on the paradoxes. A crucial question we will have to address is the
extent to which this division is justified. Ramsey also suggested solutions
to the paradoxes of each kind. One was similar to that of Russell; the
other was more novel. This and its adequacy constitute the second topic
we will take up in this chapter. Finally we will look at some of the more
modern descendants of Ramsey's solution.
1 Ramsey ( 1 925); subsequent page references are to this - the reprinted edition - unless
otherwise specified.
1 55
1 56 Limits and the paradoxes of self-reference
tion, and divided the paradoxes neatly into two families. It is difficult to
state the criterion for the distinction more clearly than did Ramsey
himself. I will simply, therefore, quote him ( p. 1 7 1 ) :
A. (1) The class of all classes which are not members of themselves .
(2) The relation between two relations when one does not have
itself to the other.
(3) Burali Forti's contradiction of the greatest ordinal.
B. (4) 'I am lying. '
(5) The least integer n o t nameable i n fewer than nineteen sylla-
bles.
(6) The least indefinable ordinal.
(7) Richard's contradiction.
(8) Weyl's contradiction about 'heterologische'.
Despite the collapse of the effective basis of the distinction, the distinc
tion itself, like the grin of the Cheshire Cat, survives and continues to
provide an hypnotic fixation for work on the paradoxes. To make mat
ters worse, it is now customary to draw Ramsey's distinction in even
more misleading terms. The paradoxes in Group A are usually called
set-theoretic paradoxes; those in Group B are usually called semantic
paradoxes. Putting the classification in these terms seems to leave no
room for, for example, the Fifth Antinomy or Berkeley's Paradox at all.
From the present theoretical perspective, the distinction between
Groups A and B looks even more arbitrary and misleading. To see
this, distinguish, first, between two sub-classes of Group B. First there
are those of its members that explicitly involve the notion of definability
(for example, (5) (7) in Ramsey's examples). Call this Bi. Next are the
-
others ((4) and (8» . Call this Bii. Now, the paradoxes in Group A are the
paradoxes of the limit of iteration (though the Fifth Antinomy uses the
notion of thought, and so Ramsey would have had to classify it in Group
B); 2 and the paradoxes in Group Bi are the paradoxes of the limit of
conception (definition) . But, as we saw in chapter 9, these paradoxes have
a common underlying structure, the Inclosure Schema, and so belong to
the same family. All the paradoxes in this group depend on a diagonaliser
being applied to a limit totality.
2 We have not met Ramsey's paradox (2) before, but it is not difficult to see that it fits
Russell's schema. Let <p(y) be 'y is a relation' , so that n is the set of all relations. If x<;::: n
then 8(x) is the relation rx, defined as follows: (y,z) E rx iff y,z E x and (y,z) � y. Closure
is immediate. If rx E x then (r"rx) E rx iff (rx, rx) rt. rx. Hence, we have Transcendence.
1 58 Limits and the paradoxes of self-reference
a(a) E a =} (o: � a) E a
=} (o: � a) E Tr
=} o: � a (by the T-schema)
=} a(a) � a
3 This observation appeals to the naive notion of set, and would be challenged, for
example, by those who subscribe to ZF or some similar form of set theory. We will
see the inadequacy of such theories in the next chapter. The naive notion of set is also
defended in Priest ( 1 987), chs. 2 and 1 0 .
Parameterisation 1 59
a(a) E a ::::} a E a
::::} a E Tr
::::} ((3 E Tr) E Tr
::::} (3 E Tr (by the T-schema)
::::} (a � a) E Tr
::::} a � a (by the T -schema)
::::} a(a) � a
(v E a) E a =} (v E a) E Ret
=} -,( (v E a) sat (v E a)
=} (v E a) f- a (by the Satisfaction Schema)
Table 9
5 In fact, we have met one already in 9.8, but this was essentially just an artifact of Russell's
solution to the paradoxes.
6 The observation is made, essentially, by Pollock ( 1 970).
1 62 Limits and the paradoxes of self-reference
7 I am assuming here that each state of affairs can be about only a finite number of
ordinals, i.e., that states of affairs are what the Tractatus will call 'atomic' . If this is
not obvious, just let 'IjJ(x) be 'x is a definable set of states of affairs, each of which is about
only a finite number of ordinals'.
8 A construction of the following kind is given in Grim ( 1 99 1 ), ch. 3.
Parameterisation 1 63
variable occurring in it. This, in effect, takes all the functions, of whatever
order (in Russell's sense), whose arguments are of the same order and
makes them a single class, over which a variable can range. It is clear that
this removes the need for an Axiom of Reducibility (and so the problem
of justifying it) at a stroke. (This was Ramsey' s main aim.) In what
follows, it will be important to keep clear the difference between
Ramsey's definition and Russell's. I will retain the word 'order' for
Russell's definition, and use the word 'type' for Ramsey's. Hence,
Ramsey's hierarchy comprises objects (type 0), function of these (type
1 ), functions of functions (type 2), etc.
The hierarchy of types still provides a solution to standard Group A
paradoxes. The totality of all ordinals (or whatever) of type i, still turns
out to be of type higher than i, making it impossible to demonstrate
Closure. And the totality of all ordinals, all propositions, etc., still does
not exist. But, just because of this, the solution inherits the major weak
ness of Russell's original solution: it renders its own explanation beyond
the limit of the expressible. I need say little more about this here.
The situation is very different for the paradoxes in Group B, however.
There is still no set of all definable ordinals. But the collapse of a vertical
section through Russell's orders into a single type, means that a version
of Koenig's Paradox can be pushed through for the set of definable
ordinals of any given type. Similar comments apply to the other para
doxes of Group B. To understand why, exactly, and what Ramsey's
solution to the paradoxes of Group B was, it is best look at an example.
Ramsey himself, gives the fullest explanation of his solution using the
Heterological Paradox (pp. 1 9 1 ff.), so let us stick with this. His version of
the paradox is somewhat different from the version I gave in 1 0.2. A
major difference is that he formulates the paradox in terms of the con
dition of heterologicality (---, x sat x), rather than the heterological set. To
discuss what Ramsey would have said about the version I gave would
raise the question of his views about sets (which were, in fact, very similar
to Russell's). From the point of view of a solution to the Heterological
Paradox, this would both complicate matters, and be absolutely irrele
vant. Moreover, Ramsey's discussion of the issues is fairly intricate, and I
do not want to introduce further complications. Hence I will discuss the
paradox in essentially the form he gave it.
Ramsey cashes out the notion of satisfaction in terms of properties
(propositional functions). I will use lower-case letters for variables of
type 0, and upper-case letters for variables of type 1 . Now let us write
xRY for 'x denotes Y' or 'x means Y' as Ramsey puts it. The condition
'---, x sat x' can then be written 3Y(xRY /\---, Yx). Let us write this as Fx.
We then reason as follows:
1 64 Limits and the paradoxes of self-reference
to
Perhaps the first person to suggest this possibility was Parsons ( 1 974c). The idea has been
developed at greater length by Burge ( 1 979). Its most sophisticated technical presentation
is in Barwise and Etchemendy ( 1 987), Part III, who call the contexts 'situations' and give
a set-theoretic analysis of these.
11 I think that the originator of this view is Buridan. (See Hughes ( 1 982).) For others who
have run the line, see Smiley ( 1 993), p. 25.
12
See, for example, the discussion of the solutions of Kripke, Gupta, and Herzberger in ch.
I of Priest ( 1 987).
1 68 Limits and the paradoxes of self-reference
13 There is one subtlety worth noting here concerning Tarski' s case. The argument that the
Liar sentence at level n diagonalises out of the true sentences of the language of level n
uses the T-scheme for level n. But this is valid only for sentences of level n- 1 . Hence, it
may be thought, the argument fails since the Liar sentence is itself of level n. However, it
is only half of the T-schema that is used, T ( a ) --> a (as may be seen by consulting the
argument of 10.2) and if a is not a sentence of level n- l , it is not a true sentence of level
n- l ; hence, assuming classical (or at least intuitionist) principles concerning the condi
tional in the T-scheme - which all these solutions do - this half of the T-scheme is valid
for all formulas, and not just those of level n- 1 .
Parameterisation 1 69
indeed, there is evidence to show that it is not. 14 Nor is there any reason
to suppose that the extensions of words like 'true' are context-dependent,
in the way that, for example, 'past' is. Parsons ( 1 974c) argues that some
of the predicates in question may naturally be analysed in terms of
quantifiers Gust as Ramsey had suggested), and that quantifiers - or,
better, their domains - are contextually determined. It is certainly true
that the domains of some quantifiers are contextually determined
('everyone has had lunch'); but, equally, those of others are not ('every
natural number is odd or even'), and Parsons gives no reason indepen
dent of the paradoxes to suppose that the quantifiers in question are
context-dependent. Finally, the claim that different sentences of the
same non-indexical type can have different truth values is patently ad hoc.
Next, and even assuming that it is correct to parameterise <p, it is
normally difficult to find independent reasons as to why the parameter
must change from c to c' to give consistency. 1 5 One might argue that the
fact that 8 is a diagonaliser demonstrates that -'<p(c,8(Oc»; and since
8(Oc) is clearly <p in some sense or other, there must be a c' such that
<p(c ',8(Oc»; but this does not establish that c is not c ' , which is just what
is at issue. 1 6 Tarski obtains the fact that the Liar sentence at level n is true
at level n + 1 , and not at level n, purely by definition: the way the hier
archy is defined, the sentence just is a sentence of level n + 1 , and not n.
But, unless this is pure legerdemain, the question remains as to why things
should be defined in this way.
Thirdly, and crucially, the parameterisation does not avoid the para
dox, but merely relocates it. For we can define the logical sum of all the
parameterised relations as follows: <p* (y) iff 3c<p(c,y) (y is true at some
level, in some context, etc.). Let 0* be the corresponding totality. One
now shows, by the standard argument, that the relevant 8 is a diagona
liser with respect to <p*, and hence that -'<p*(d(O * » . But, as usual, since
8(0 *) has been shown to be <p in some respect, <p*(8(0*» . For example,
consider the claim that this sentence type is true in no contextjtokening.
Suppose that this is true in some contextjtokening, then it follows that it
is not true in that contextjtokening. Hence it is true in no contextjtoken
ing. I.e., it is true in this contextjtokening, and so in some contextjtoken
ing. 1 7
There are a few ways that one might try to avoid this conclusion. The
first (Burge, ( 1 979), p. 1 92) is to claim that cp* cannot be defined, since
one cannot legitimately quantify into the parametric place. This is not
only false, it is self-refuting (if one wishes to be consistent) : even to
explain the view that cp is parameterised one needs to say that things
can be cp in some respect, and not others, and hence quantify into the
parametric place.
Another possible suggestion is to attack the proof that 8 is a diagona
tiser with respect to cp * . For example, in the case of Bi paradoxes, one
might deny that :3yy .;. 0 * , on the ground that everything is definable
(definable in less than 99 words, etc.) relative to some parameter.
Exactly how plausible this is, depends on the paradox and the under
standing of the parameter. It may not be implausible to suppose, for
example, that every natural number is definable in less than 99 words
in some context. But it is most implausible to suppose that every ordinal
is definable in some context, at least if a context is anything that is
humanly accessible. In any case, this move will obviously not help with
paradoxes in Group Bii.
A third possibility is to deny Closure, possibly on the ground that
8(0*) does not exist. 1 8 This, however, brings other, and familiar, pro
blems. For the very solutions offered always provide the wherewithal to
construct 8(0*) and demonstrate that it is cp in some sense. (For example,
the least ordinal definable in no context is defined in this context; and
'this sentence is not true in any language in the Tarski hierarchy' is a true
statement of some language.) But now we are faced with a choice: either
this sense is one of the parameters, and hence we have a contradiction; or
it is not, and since we had enumerated all possible parameters, this fact is
not expressible at all: it follows that the solutions themselves are beyond
the limit of the expressible. 1 9
In 9 . 8 I observed that Russell's solution to the paradoxes merely
reproduces the problem; and, more specifically, that it instantiates the
Inclosure Schema. We are now in a position to see that exactly the same is
true of parameterisation. In the Schema, take for cp the logical sum of the
parameterised properties just defined, cp*; take for 0, 0*; let 'l/J(x) be
*
'x = 0 or for some c, x = Oc' ; and let 8 be the appropriate diagonaliser.
Then, clearly, 'l/J(O*); and as we have just seen, if x is any 'l/J subset of 0*,
8(x) diagonalises out of x, but has the property cp * . Hence, these quan
tities satisfy the Inclosure Schema. Again, parameterisation is less of a
18
This line is taken in Barwise and Etchemendy ( 1 987).
19
For a critique of Barwise and Etchemendy on these grounds, see Priest (1 993).
Parameterisation 171
Conclusion
In this chapter we have looked at a number of modern solutions to the
paradoxes of Ramsey's Group B. As we have seen, they have the same
general failings as did Ramsey's own solution. The other sorts of modern
paradox-solution are aimed primarily at Ramsey's Group A. We will turn
to these in the next chapter. As we will see there, they fare little better.
11 Sets and classes
Introduction
As we have seen in the last two chapters, all the paradoxes of self
reference (including Berkeley's Paradox and the Fifth Antinomy) are
inclosure contradictions; that is, they all instantiate the Inclosure
Schema, which, to remind the reader, concerns properties <p and t/J, and
a function () such that:
One might depict these conditions as in Figure 1 . The large oval is n, the
set of all <p things. x is any subset of n satisfying �, and 8 applied to this
takes us out of x but into n. Applying 8 to n takes us both into and out of
..-_-. lleQ)
",ex)
Figure 1
Sets and classes 1 73
I There is something rather appropriate about this. The boundary is, in its own way,
paradoxical, both joining and separating the inside and the outside.
2 It is often noted that the pseudo-paradox of the barber is structurally similar to Russell's
paradox (for example, Sainsbury ( 1 988), p. I I 0). It is therefore interesting to note how
this relates to the Inclosure Schema. Let rp(y) be 'y is a person', so that !1 is the set of all
people. Let 1f;(x) be 'x comprises the people of a group where those who do not shave
themselves are shaved by a unique person'; and let 8(x) be that unique person. Closure is
immediate, and if 8(x) E x then 8(x) does and does not shave themself. Hence we have
Transcendence. The paradox fails simply because 1f;(!1) is false, and so Existence fails,
though not because !1 fails to exist.
3 ( 1 908); the translation is from van Heijenoort ( 1 967), p. 202.
1 74 Limits and the paradoxes of self-reference
This axiom, together with some more basic axioms of set existence
(such as those for power-set and union) can be used to establish the
existence of most sets standardly appealed to in mathematics. But, cru
cially, the absolutely infinite sets, such as V, On, and R, involved in the
paradoxes of absolute infinity cannot be shown to exist. If they could be
shown to exist, the paradoxical arguments would show the theory to be
inconsistent. 4 It should be noted that there are other sets that cannot be
shown to exist in ZF, whose existence is consistent with the axioms of ZF,
for example, non-well-founded sets. Though the standard interpretation
of ZF is the cumulative hierarchy (8.4), there are other interpretations in
which there are non-well-founded sets. 5
Of course, the mere fact that the contradiction-producing sets cannot
be proved to exist hardly, of itself, solves the paradoxes. No one doubts
that if the proof-theory of set-theory is sufficiently emasculated the
theory can be rendered consistent. The question is what reason there is
to suppose that the sets do not exist. If the set of all (pure) sets could be
shown to exist, then the existence of all the paradox-producing sets would
follow by the Aussonderung Axiom. The crucial question is, therefore,
why one should suppose that V does not exist. Zermelo shows that there
is no universal set as follows: 6
THEOREM . Every set M possesses at least one subset Mo that is not an element
of M .
Proo/ [ If ] M o i s the subset o f M that, i n accordance with Axiom 3 , contains all
those elements of M for which it is not the case that x E x, then Mo cannot be an
element of M . For either Mo E Mo or not . In the first case, Mo would contain an
element x Mo for which XEX, and this would contradict the definition of Mo.
=
4 For details of all this, see virtually any modern book on set-theory, for example,
Fraenkel, Bar-Hillel, and Levy ( 1 973), ch. 2, section 3 .
5 See, for example, Aczel ( 1 988).
6 ( 1 908), Theorem 1 0; see van Heijenoort ( 1 967), p. 203 .
Sets and classes 1 75
is true, which is a nice way of expressing the thought that all sub-classes
of the domain are potential sets.
A minor problem with this construction is that, as specified so far,
there is no reason to suppose that the structure is a model of ZF. This
can be rectified by supposing that for the base world, Rm a is an inac-
10
Mayberry is sympathetic to an intuitionist approach, though the exact details of his
approach are more complex. For further discussion, see Priest ( 1 987), p. 47.
II
See Dummett ( 1 977), pp. 55ff.
12
The following is inspired by Parsons ( 1 97 1 ), fn. 1 0, and Parsons (1 974b). The material in
the rest of this section and the next uses various technical notions not explained in the rest
of the book, and can be skipped over without loss of continuity.
Sets and classes 1 77
13 It is not necessary to know what an inaccessible cardinal is here - for details, see
Fraenkel, Bar-Hillel, and Levy ( 1 973), section 6.3; all that it is necessary is to know is
that if 0 is inaccessible, Ro is a model of ZF.
14 Lear ( 1 977) gives a construction similar to o/i , except that he gives intuitionist truth
conditions for the logical constants. In particular, a universal quantifier over all sets is
effectively involved in the truth conditions of the universal quantifier. His construction
therefore faces the same objection.
1 78 Limits and the paradoxes of self-reference
The intuitive concept of ordinal number, like that of cardinal number and set, is
an indefinitely extensible one.
So far, so good; but the onus is now on someone who holds this
position to explain the things we take to be true about ordinals, sets,
and other absolute infinities. Given that we are dealing with variable
extensions, and the notion of being an ordinal or a set is not itself
vague, the only obvious suggestion is that the definite truths are those
that are true however the extensions of the relevant notions are fixed;
whilst the statements of indefinite truth value are those that may vary
from fixture to fixture. But how to make this thought precise? The
natural way of cashing out the idea is precisely as the model l1lt that we
met in the previous section: we now think of each possible world as one of
the ways of fixing extensions. The determinate truths are exactly those
that are necessarily true at the base world.
It follows that the suggestion in question is little better than the modal
one, and suffers from the same major problem. Even to conceptualise
how the model is supposed to work we have to suppose a totality of all
ordinals, or sets, and quantifiers that range over them. Hence we seem to
be back with the original problem.
Could we avoid it by interpreting the metalinguistic machinery as
vague in the same way? Possibly, but it is then dubious that it could
serve the explanatory function for which it was proposed: the very
sense of the construction is drawn from the contrast between the abso
lute totality and its fragments. If the absolute totality is itself variable, the
fragments become otiose. In any case, the suggestion to interpret the
metatheoretic machinery in this way would be disingenuous: this is not
the way that the model was intended. Its intent was precisely to show how
partial fragments could approximate an absolute whole. And one must
have a conception of an absolute whole for this idea to make sense. As
Kant saw so well, given a notion like that of set or ordinal, reason forces
us to conceive of the totality of all things satisfying it. Totalising is part of
our conceptual machinery - like it or not.
Despite the fact that I have used a quotation from Dummett to illus
trate the suggestion I have been discussing, his view of the matter is rather
different. It is explained in ( 1 99 1 ) pp. 3 1 3- 1 9 . The purpose of this passage
is to argue that quantification over all infinite totalities, including, but
not exclusively, absolutely infinite ones, must be intuitionist. As he puts
it, the point is to reject the claim that (p. 3 1 3):
For these reasons, his views do not resolve the present problem, as he
himself acknowledges. 1 5
15 'Abandoning classical logic is not, indeed, sufficient by itself to preserve us from contra
diction if we maintain the same assumptions as before; but, when we do not conceive
ourselves to be quantifying over a fully determinate totality, we shall have no motive to
do so' (P. 3 1 6, fn.).
1 80 Limits and the paradoxes of self-reference
do not exist in ZF, such as R, On, and V, are proper classes (as the same
arguments that show them not to exist in ZF demonstrate) .
I should note that this description, though correct, is, strictly speaking,
anachronistic for three reasons. The first is that von Neumann took
I-objects to include not only sets but also urelements (non-sets); but
this makes no difference here. The second is that von Neumann took
I-objects and II-objects to be the characteristic functions of collec
tions, rather than the collections themselves. However, in virtue of
the interdefinability of collections and their characteristic functions,
this is of no importance. The third is that von Neumann did not,
himself, identify I-objects with the sets of ZF and II-objects with sub
collections of this domain. Rather, the identification came to be
natural at a later time in virtue of results showing that these notions
were technically equivalent. ! 6
Now, a proper class cannot, b y definition, b e a member o f anything.
But this does not give a criterion for when a collection, specified by some
condition, is a proper class. The criterion that von Neumann proposed,
and that is now standard, is that a class is proper when it is at least as big
as (and therefore the same size as) V. This is encapsulated in his Axiom
IV.2: ! 7
A II-object a i s not a I-II-object i f and only if there is a [function] b such that for
every I-object x there exists a y for which both y E a and bey) = x.
There are . . . some qui te essential differences [between Zermelo 's system and
mine] . That we speak of 'functions' rather than ' sets' is no doubt a superficial
difference ; it is essential, however, that the present set theory deals with sets (or
'functions') that are 'too big', namely those II-objects that are not I-II-objects.
Rather than being completely prohibited, they are declared incapable of
being . . . [members] (they are not I-objects!). This suffices to avoid the anti
nomies.
This solution is reminiscent of Kant's Solution 2 (6.7): the limit exists, but
the operation that gives contradiction, viz . , membership of something or
other, cannot be applied.
The crucial question now would seem to be whether there are any
grounds, except definitional legerdemain, for supposing that there is
nothing for proper classes to be members of. As we have just seen, von
Neumann calls the collections 'too big'; but it is difficult to see why size,
as such, should have anything to do with being a unity, i.e., the possibility
of being a member (see 8 . 6). Indeed, to refer to a collection, even if it is a
proper class, is, in a certain sense, to treat it as an entity, a unity; a
fortiori, if one quantifies over it. Any collection (set or class) must, there
fore, be a candidate for membership. 1 8
This is all bad news; but worse is to come. Consider the paradoxes of
set-theory as they appear in propria persona in von Neumann's theory.
Take the Burali-Forti paradox as an example. Instead of considering the
collection of sets that are ordinals, we can consider the collection of
classes that are ordinals (which is, intuitively, On U {On}), and try to
run the paradox for this. When we do so, we find that it fails, and for
exactly the same reason that it fails in ZF: there is no such collection. (If
there were, On would be a member of it.) Similar remarks apply to
Russell's paradox and Mirimanoffs: neither the collection of pure
classes, nor the collection of well-founded classes exists. Von
Neumann's solution is ultimately the same as Zermelo's.
Unsurprisingly, then, it shares with Zermelo's the same fundamental
problem. Note that the variables of von Neumann's theory range over the
domain of all collections (I-objects and II-objects) . The totality of all
18
Lewis ( 1 99 1 ) achieves some of the effects of classes with the device of plural quantifiers
ranging over sets. It might be thought that this would resolve the problem, since in this
case we no longer have to quantify over classes themselves, and so treat them as entities.
It has been argued by Resnik ( 1 988) that plural quantification over sets is equivalent to
simple quantification over classes. If this is so, then nothing is gained by this move. In
fact, I doubt that it is so. However, avoiding this Scylla drives the suggestion straight
back into Charybdis in the present context. Just because plural quantification does not
recognise classes as entities, Lewisian set-theory is no advance on ordinary ZF as far as
the present concerns go : the quantifiers (singular and plural) range over a totality which
in no sense exists .
1 82 Limits and the paradoxes of self-reference
collections is not, therefore, V (the collection of all sets), but V', the
collections of all sets and classes. And it is this, and similar collections,
whose existence cannot be consistently admitted in von Neumann's the
ory. Hence, the theory violates the Domain Principle just as much as
Zermelo's. (Sometimes, von Neumann's theory is represented as a two
sorted theory with different kinds of variables ranging over I -objects and
II-objects. This makes no essential difference to the point being made: the
Domain Principle is violated by the variables ranging over II-objects.)
And, as for ZF, this inadequacy shows itself in the fact that von
Neumann's theory does not provide an adequate basis for category
theory. One might, for example, be able to form the category of all
sets; but one cannot form the category of all classes. 1 9
It might be thought that this problem could be solved by supposing
that proper classes can be members of some sort of collections that bear
the same relationship to them as they do to sets, hyper-classes. 2o But this
just repeats the problem at the next higher level. An obvious thought is to
iterate this construction into a hierarchy so that a collection of any order
can be a member of something at the next higher level . This gives us, in
effect, the set-theory of Ackermann, 2 1 which can be thought of as iterat
ing this procedure to some transfinite ordinal. 22 A little thought shows
that this construction does not solve the problem either, but, again,
merely transfers it. For in such a theory there are variables that range
over all the collections in the hierarchy; yet on pain of the usual contra
diction, the domain of this variation can be proved not to exist in the
theory.
However one works out the details, all the solutions we have discussed
ultimately take the Aristotelian way out. The limit (the totality of all
collections countenanced by the theory) is said not to exist; and hence
the Domain Principle is violated? 3
1 9 Or, for that matter, even the functor category of the category of all sets - see Priest
( 1 987), p. 42.
20 See Fraenkel, Bar-Hillel, and Levy ( 1 973), p. 1 42.
21 Ibid., pp. 1 48ff.
22 See Grewe ( 1 969).
23 A somewhat different kind of solution is suggested by Maddy ( 1 988). She constructs a
theory of classes on top of a theory of sets by applying the techniques of Kripke ( 1 975), to
obtain proper classes which may be members of proper classes. The paradoxes are
avoided since the logic of proper classes is three-valued. The problem with this construc
tion is the same as that with Kripke's itself. It remains consistent only because certain
legitimate - indeed, necessary - notions are not expressible in the theory. Exclusion
negation, or equivalently, genuine complementation, cannot be expressed, or Russell's
paradox would reappear.
Sets and classes 183
1 1 .7 Unified theories
Even had Ramsey been right, and the paradoxes of Groups A and B been
distinct, contemporary attitudes to the two families and their solutions
Sets and classes 185
1 1 .8 Curry paradoxes
Although it is a digression, this would seem to be the right place to say a
word about another kind of paradox that is related to some of the
paradoxes I have been discussing, and which also cuts across Ramsey's
division: Curry paradoxes. Some of the paradoxes I have discussed pro
ceed by establishing a sentence of the form a +-t ...., a . (All the paradoxes in
Group Bii do this, and some of the paradoxes in Group A, notably
Russell's, but not the paradoxes in Bi or the other paradoxes in group A.)
For each paradox of this kind, we can form a new paradox by replacing
,a uniformly with a -+ 13, where 13 is an arbitrary formula; or, more
simply, with a -+ ..1 , where ..1 is some logical constant entailing every
thing. Using the Absorption Principle ( a -+ ( a -+ 13) f-- a -+ 13) we can
then infer a -+ ..1 , and hence a, and hence ..i. (See Priest ( 1 987), 6.2.)
Do such paradoxes fit the Inclosure Schema? Yes and no, depending
on what -+ is. If it is a material conditional then, in most logics, a -+..1 is
logically equivalent to ,a, and so the curried version of each paradox is
essentially the same as the uncurried form. If, on the other hand, -+ is a
non-material conditional (for example, a strict conditional), then a -+ ..1
and , a are quite different notions. (Evaluating the truth of the first at a
world requires a consideration of what is happening at other worlds;
evaluating the truth of the second does not.) In this case, the curried
versions of the paradoxes belong to a quite different family. Such para
doxes do not involve negation and, a fortiori, contradiction. They there
fore have nothing to do with contradictions at the limits of thought.
Conclusion
To return from the digression: In the last two chapters we have, inter alia,
completed a review of all main contemporary solutions to inclosure
contradictions. As we have seen, the solutions are not adequate, even
in the limited domains for which they were generally designed.
Moreover, not only do they tend to be incompatible, but the piecemeal
approach initiated by Ramsey flies in the face of the PUS and the fact
that all such paradoxes instantiate a single underlying structure: the
Inclosure Schema. The only satisfactory uniform approach to all these
paradoxes is the dialetheic one, which takes the paradoxical contradic
tions to be exactly what they appear to be . The limits of thought which
are the inclosures are truly contradictory objects. 28
The notion of an inclosure plays a central role in the paradoxes of self
reference. But it would be a mistake to suppose that the relevance of this
notion is restricted to that area. Another area in which it plays an
important role is the philosophy of language . Why this is so, is an
issue that will occupy us in the next, and final, part of the book . That
part is not concerned wi th logical technicalities in the way that much of
the present part of the book has been (many readers will be relieved to
28 Interestingly enough, Hegel held just this view of the Liar paradox. See his remarks on
Eubulides in volume I of his Lectures on the History of Philosophy.
Sets and classes 1 87
hear). However, before we leave the technicalities behind, and for the
sake of interest, I will give a few details of the model theory of inclo
sures. These will be of interest only to logicians, and, strictly speaking,
constitute a digression from the main purpose of the book. Hence I
relegate them to an appendix, which can safely be omitted.
Technical appendix
Introduction
The notion of an inclosure is obviously a strange one. Since its introduc
tion in 9.4, we have seen various arguments to the effect that there are
inclosures. However, since the notion is an inconsistent one, one might
doubt that the arguments work, simply because the notion is incoherent
in some sense. The purpose of this technical appendix is to assuage some
of these doubts.
One way to show a notion to be coherent is to produce models of it.
This is precisely the strategy I will adopt here. I will show that there are
models, in the standard sense of model-theory, of the inclosure condi
tions. Since the conditions are inconsistent, these are not models of
classical or intuitionist logic. They are, and have to be, models of a
paraconsistent logic.
There are many paraconsistent logics. 1 It will suffice for my purposes
to take the simplest of these, the system LP of Priest ( 1 987), chapter 5. In
particular, I will interpret the � in the inclosure conditions as a simple
material (and, in paraconsistent contexts, non-detachable) conditional.
The task of showing that the inclosure conditions have non-trivial mod
els where the conditional is a relevant, or at least, detachable, one is more
complex, and still needs to be undertaken. (That this can be done success
fully, I have little doubt.)
The techniques used to produce the models in question were initiated
by Bob Meyer. His original work is still unpublished, though it is
abstracted in ( 1 976). Perhaps the clearest presentation of the techniques
is in Meyer and Mortensen ( 1 9 84) . They involve, essentially, taking a
model of a theory and collapsing it to produce an inconsistent model.
In the case of Meyer and Mortensen, the model collapsed is the standard
model of arithmetic, and the collapsed models are non-standard, finite
models.
1 For a review of them see Priest, Routley, and Norman ( 1 989), ch. 5 .
188
Technical appendix 1 89
In what follows I will briefly describe the logic LP. I will then state the
relevant collapsing theorem. 2 Finally, I will apply this to construct mod
els of the inclosure conditions. I will make no attempt to make the details
intelligible to non-logicians, and this appendix can be skipped by the
general reader without 10ss. 3
1 Semantics of LP
The language of LP is a first-order language with connectives ..., and A ,
and quantifier Y. :3 and V are defined i n the usual way. Note that ep -t 'IjJ
is defined as ""ep V 'IjJ. There are constants and predicates, including the
identity predicate. In principle, there is no problem about the language
containing function symbols. However, the Collapsing Lemma, in the
form that we will need it, fails if the language contains them. Hence,
we will assume that they are not present. This is no loss, as we will see.
An interpretation, dII , for the language is a pair (D,I ) , where D is the
non-empty domain of quantification; I is a function which maps each
individual constant into D and each n-place predicate, P, into a pair
n
< I + (P), r(p) > , where 1 + (P)UnP) = D . (Note that I + (p)nnp) may
be non-empty.) I will call the members of the pair the positive and
negative extensions of P, respectively. For identity to have the right
properties we also require that 1 + ( = ) = { (x,x) ; XED} . The language of
dII is the language augmented by a set of individual constants, one for
each member of D. For simplicity we take the set to be D itself, and
specify that for all dED I(d) = d.
Every formula in the language of dII , ep, is now assigned a truth value,
v(ep) , in the set { { l } , {O} , { I ,O} } by the following recursive clauses.
1 E V(Pt l . . . tn) <=> (I(t l ) . . . I(tn )) E I + (P)
° E V(Pt l . . . tn ) <=> (I(t l ) . . . I(tn )) E I - (P)
1 E v(..., ep ) <=> ° E (v)
° E v(""ep)<=> 1 E v( ep)
2 This is more general than the Meyer collapse in one sense, since it applies to arbitrary
models and equivalence relations. However, the Meyer collapse is stronger in another
sense, since it handles not only extensional connectives, but also a non-extensional con
ditional.
3 The material in this chapter is part of Priest ( 1 992).
1 90 Limits and paradoxes of self-reference
Where <p(x/d) denotes <p with all free occurrence of 'x' replaced by 'd' .
Let IlIt (D,I) be an interpretation; then IlIt is a model for <p (1lIt � <p) iff <p
=
is true in 1lIt, i.e. , I E v(<p) . The set of sentences in the language of illt true in
IlIt is called the theory of 1lIt .
A couple o f facts about L P are worth noting. The first i s that every LP
interpretation is a model of every truth of classical first-order logic (with
identity). 4 (The consequence relations of LP and classical logic are, of
course, different.) The second is that any standard classical interpretation
is isomorphic to an LP interpretation in which all atomic formulas (and
so all formulas) take the value {O} or { I } . Consequently I will simply
identify such interpretations. Thus, models of classical theories are just
special cases of LP models.
Collapsing Lemma
Let <p be any formula; let v be I or O. Then if v is in the value of <p in 1lIt, it
is in its value in 1lIt - .
I n other words, when IlIt i s collapsed into 1lIt - , formulas never lose truth
values: they can only gain them. The Collapsing Lemma is the ultimate
downward Loewenheim-Skolem Theorem. If theory T has a model, 1lIt,
then T has a model of every cardinality less than that of IlIt (wherein, note,
= has its standard positive extension) . To obtain a model of cardinality K,
(less than that of 1lIt) we simply choose an equivalence relation on the
domain of IlIt that has K, equivalence classes. (A moment's thought shows
that there is always such an equivalence relation .) The result then follows
by the Collapsing Lemma.
I will not prove the Collapsing Lemma here. The proof i s by a straight
forward induction over the complexity of formulas, and can be found in
Priest ( 1 99 1 c). 5 It should be noted that if the language were to contain
function symbols the proof would not work: there is no way of defining
the interpretations of the function symbols in 011 - to ensure that if term t
denotes d in OIl, it denotes [d] in 011 - , which is required for the rest of the
proof. 6
Now, let OIl be an LP interpretation with at least a two-membered
domain. Let - be any equivalence relation on the domain of OIl, and let
011 - be the interpretation constructed by collapsing. T and T- are the
theories of OIl and 011 - , respectively. Then the Collapsing Lemma assures
us that T � T - . Of course, T- may properly extend T. If - is anything
other than the trivial equivalence relation that relates every object of the
domain only to itself, there will be distinct objects, a and b, such that
a = b E T - . T- will therefore be inconsistent, even if T is not. On the
other hand, provided that - is not the trivial equivalence relation that
relates every object to every other, there will be distinct objects, a and b,
such that a = b \t r. Hence r will be non-trivial.
fore always trivially satified. Thus, in effect, we are modelling not only
the Inclosure Schema, but Russell's schema. Next, in every model, we will
choose some object, a, in its domain and take rp(y) to be 'y E a ' . Hence n
will just be a, and Existence will be trivially satisfied; we may therefore
ignore it.
Finally, we must eliminate function symbols, since we cannot guaran
tee collapse if these are present. The function symbol 0 in Transcendence
and Closure can be replaced, in a standard way, with a functional relation
symbol . If we replace it in this way, these conditions become:
(i) Vx:::J !yDxy
(ii) VxVy(x � n 1\ Dxy --+ y � x)
(iii) VxVy(x � n 1\ Dxy --+ y E n)
(where ' ! ' has its usual meaning here). Hence we need to find a model of
these three conditions.
Let 0/1 be any (classical) model of ZF with domain S. Let On(x) be the
formula of ZF that expresses the claim that x is an ordinal. I will refer to
anything that satisfies On(x) in 0/1 as an ordinal; and when I refer to the
ordering of these, I will mean simply the ordering in 0/1. We will be
concerned with equivalence relations on S that leave all non-ordinals
alone, but identify certain ordinals.
Let Dxy be log(x,y), the formula of ZF expressing the fact that y is the
least ordinal greater than every ordinal in x. Let n be any ordinal, a.
Then (i), (ii) are true in every model of ZF, and hence any collapsed
model. (For (i) recall that every set of ordinals is bounded above.) We
therefore need only choose an equivalence relation that makes (iii) true in
the collapsed model.
Let - be the equivalence relation that identifies every ordinal greater
than a with a. Let c and d be any members of S, and let B(y) be:
c � n 1\ log(c, y) --+ yE n
If d is not an ordinal then ..., l og(c,d) is true in 0/1 . Hence B(d) is true in 0/1,
and hence 0/1-. If d is an ordinal < a then d E n is true in 0/1, and so 0/1 - .
I f d i s an ordinal 2: a then for some ordinal j3 , d E j3 i s true i n 0/1, s o d E n
is true in 0/1 - (since a has been identified with j3). Hence, VyB(y) is true in
0/1 - ; (iii) follows. It is worth noting that in 0/1 - , n is exactly the set of all
ordinals.
Conclusion
The above are just a sample of models for the Inclosure Schema that can
be constructed by collapsing; others can be produced in numerous ways.
For example, we could identify, instead of ordinals, stages of the cumu
lative hierarchy. Alternatively, there are known to be (LP) models of ZF
plus the naive Comprehension Schema. 7 If we start with one of these
models instead of a simple model for ZF, then we will obtain collapsed
models that satisfy Comprehension as well as ZF and the inclosure
conditions. I suspect that a more systematic study of collapsed models
will yield a number of interesting results.
Part 4
To say that 'everything is the Tao' almost gets the point, but just at the
moment of getting it, the words crumble into nonsense. For we are here
at a limit at which words break down because they always imply a
meaning beyond themselves - and here there is no meaning beyond .
Introduction
The issue we are concerned with in this book is that of the limits of
thought. We have seen how contradictions at the limits of thought that
we have been concerned with take a very sharp form in the inclosure
contradictions of self-reference. The story does not end here, however.
As we saw in the second part of the book, the classical formulation of the
problem was given by Kant, who put the issue in terms of the limits of
reason. The twentieth century, for its own reason, has taken to formulat
ing philosophical issues in terms of language. It is therefore unsurprising
that the issue should appear again in contemporary form as a concern
about aspects of language. In this part of the book, we turn to this.
It is natural for a theory of language to have implications about what
can and what cannot be expressed. We will see that modern theories of
language always seem to render some very important things - usually
themselves - beyond the limit of expression. The contradictions at the
limits of thought therefore appear in a new guise here. In this part of the
book we will examine a number of modern accounts of meaning with an
eye on the matter. The main concern will not be to evaluate these the
ories; rather, it will be to examine their consequences.
In this chapter we will start by looking at one of the foundational
figures in contemporary philosophy of language: Frege. We will then
move on to the Wittgenstein of the Tractatus. Our major concern
throughout the chapter will be the unity of thought.
had never before been achieved. Ironically, this aside was to be one of the
major influences on twentieth-century philosophy.
We will not be concerned with all of Frege's views on language: only
those that locate the site of a certain aporia. The crucial papers are
'Function and Concept', 'On Sense and Reference', and, particularly,
'On Concept and Object'. ! The contents of these papers are so well
known that a lengthy exposition is hardly necessary?
Frege takes the traditional distinction between subject and predicate,
and refashions it for his own ends. Instead of the category of subject,
Frege proposes the category name. This is wider than the traditional
category, since it includes those noun-phrases that occur within the pre
dicate as a grammatical object. But it is also narrower than the traditional
category, since it excludes quantifier phrases such as 'all men'. We are left
with proper names and definite descriptions. Instead of the category of
predicate, Frege proposes the category concept-expression. A concept
expression is what is left when names are deleted from a sentence.
Thus, in 'Oswald was framed for the murder of Kennedy', 'Oswald'
and 'the murder of Kennedy' are names and 'was framed for' is a con
cept-expression.
Frege also reshaped the traditional distinction between two notions of
meaning: connotation and denotation. He distinguished between the
sense (sinn), of a linguistic unit and its referent (bedeutung). According
to Frege, all linguistic units have both a sense and a reference
(denotation). The denotation of a name is an object; the denotation of
a concept-expression is a concept. The denotation of a statement is a
truth value (true or false). The sense of a linguistic unit is, in general,
that which determines which object/concept/truth value is the correct
referent. In the case of a statement, this is the (objective) thought
expressed by it (the proposition expressed by it).
How much of this is defensible is an issue that we need not go into
here. 3 About the next of Frege's insights there can be little doubt. This is
that the meaning of a compound linguistic expression is, in some sense, a
function of the meanings of its parts. (This is usually called
'compositionality' .) Frege thought that, by and large, the referent of an
expression was a function of the referents of its parts, and the sense of an
expression was a function of the senses of its parts.
1 All to be found in Geach and Black ( 1 9 60), to which page numbers refer unless otherwise
indicated.
2 A concise, but clear, summary can be found in the introduction of Furth ( 1 9 67). A
lengthy discussion can be found in Dummett's mammoth ( 1 973).
3 Detailed discussion can be found in Dummett ( 1 973).
The unity of thought 1 99
4 A mathematician might say that 0 is the application of the function sin to 7r: Ap (sin, 7r) .
But this does not solve the problem: it just transfers it to this new functional expression.
5 A number of the following points are made at greater length in Black ( 1 968).
200 Language and its limits
the three words 'the concept "horse' " . . . designate an object; but on that
account they do not designate a concept as I am using that word.
6 The predicate 'is a function' is problematic for the same reason. This just makes the
problem worse.
7 See also the last paragraph on p. 55.
The unity of thought 20 1
I admit that there is a quite peculiar obstacle in the way of an understanding with
my reader. By a kind of necessity of language, my expressions, taken literally,
sometimes miss my thought: I mention an object when what I intend is a concept .
I fully realize that i n such cases I w a s relying on the reader who would b e ready t o
meet m e half-way - who does not begrudge a pinch of salt .
But he was not embarrassed enough. It is one thing for mystics, such as,
perhaps, Cusanus, to hold views that they also hold to be ineffable; it is
quite another for a man of science, such as Frege was.
In a later and, at the time, unpublished, essay (c 1 892) Frege comes
back to the issue and offers a solution to it. He suggests that we may refer
to a concept with a 'what' clause, as in 'what "is a horse" stands for'. The
point of such clauses is that they can themselves be used predicatively.
Consider, for example, the sentence: Frege is what I am, a philosopher
logician. In this way, we can say, for example, Phar Lap is what 'is a
horse' stands for.
The construction does strain the ear somewhat. But in any case, the
trick cannot do the required job. For if a what-clause must be construed
predicatively then the claim, for example, 'what "is a horse" stands for
was considered by Frege' is nonsense. Ditto the claim: 'what "is a horse"
stands for is not an object' , etc. And so we cannot paraphrase away all
the things we need to say.
More importantly, Frege's suggestion does not address the fundamen
tal problem, which was, as we saw, that he has at his disposal no pre
dicate that applies to concepts. Dummett, in his discussion of the
problem ((1 973), pp. 2 1 1-22) suggests that 'is a concept' should be
eschewed as a pseudo-expression. We need, instead, an appropriate sec
ond-order predicate, i.e., a phrase, like a quantifier phrase, that fits
together with an ordinary predicate to make a sentence. His suggestion
is 'Everything either is . . . or is not . . . '. If we use upper-case variables
for concepts, this is Vy(XyV.Xy) . The suggestion will not do the job,
however - if the job is to express Frege's theory. Consider, for example,
the claim that every concept is unsaturated. This becomes:
VX(Vy(XyV.Xy) -+ X is unsaturated), or more simply VX(X is unsatu
rated). This raises the question of the intelligibility of second-order quan
tification in the present context. But setting that aside, we still have the
problem of the predicate 'is unsaturated' , which is an ordinary first-order
predicate; and so this sentence is nonsense.
Maybe some way could be found to rejig the predicate as a suitable
second-order predicate (though I doubt it) . But worse is in store.
Consider the claim that concept-words refer to concepts, i.e., Vx(x is a
concept-word -+ :lY x refers to Y). The same problem arises with respect
to the predicate 'refers to' ; and this time there is certainly no way of
202 Language and its limits
rejigging it, simply because it must be legitimate to say that names refer to
objects, and so 'refers to' must be a predicate that is first order in both its
arguments. If it be retorted that the reference relations for names and
concept-words must be distinct, then we cannot say with Frege « c 1 892),
p. 1 1 8):
T o every concept-word , or proper name, there corresponds as a rule a sense and a
meaning [reference] , as I use these words.
the aim of the book is to draw a limit to thought, or rather - not to thought, but
to the expression of thought: for in order to be able to draw a limit to thought, we
should have to find both sides of the limit thinkable (i.e. we should have to be able
to think what cannot be thought). It will therefore be only in language that the
8 Translations and page numbers are from the Pears and McGuinness edition, ( 1 96 1 ) . I will
refer to propositions of the Tractatus by prefixing 'T' to their number.
The unity of thought 203
limit can be drawn and what lies on the other side of the limit will simply be
nonsense.
particular fact fit together, like pieces of a jigsaw puzzle (or links in a
chain, to use Wittgenstein's simile), according to possibilities intrinsic to
the objects. It is tempting to think at this point of Frege and the way that
concepts and objects are supposed to fit together. However, this is not
what Wittgenstein has in mind, as we shall see. It is doubtful that
Wittgenstein even took concept-expressions to refer to objects. 1 1 At
any rate, the objects in a fact are articulated into a determinate struc
ture, and the way that the objects are structured is called the form of the
fact.
On the other side of the fence, language is composed of propositions.
These are all truth-functional compounds of atomic propositions, and
hence their truth values are determined by the truth values of the atomic
propositions they contain. An atomic proposition itself expresses a pos
sible (atomic) fact, and says something about the objects comprising this
fact. Propositions are composed of names which correspond to (refer to)
objects. Like facts and objects, propositions are no mere congeries of
names, however. In particular, within a proposition the names are
related to each other in a certain way. The way they fit together is the
form of the proposition. An atomic proposition is true iff there is a fact
whose objects are the objects named in the proposition, and which has
the same form as the proposition. In other words, the proposition is true
if the map which maps names to objects is an isomorphism. Or, as
Wittgenstein puts it, the proposition pictures the fact.
II
For a start, see T3 . l 432.
The unity of thought 205
the proposition, one would not know what it said. But, even then, one
could see something about the proposition, for example, that it has a
certain form. Wittgenstein says that the proposition shows its form in this
way. As he puts it (T4. 1 2 1 ):
Propositions show the logical form of reality. They display it.
In a similar way, and quite generally, all structural facts are shown: a
proposition shows that it is a proposition, shows what its constituents
are, etc.
One should note that the word 'show' in English has both a proposi
tional use and a non-propositional use. In its propositional use, 'show' is
followed by a that-clause (she showed that she could play cricket); in its
non-propositional use it is followed by 'what' , 'how', etc . , or even a
simple noun-phrase (she showed him the bat/how to use it/where he
could put it, etc.). Structures in the world and language show in both
these senses. For an example of the latter, see the previous quotation. We
will have examples of the former in due course.
It is worth noting that there is historical evidence to suggest that
Wittgenstein observed that Russell's theory of types forced him into
supposing that there were things that could not be said (as we saw in
9 . 7), and that reflection on this played some role in the genesis of his
notion of showing. 1 2
12
See the discussion in 3 . 2 of Griffin ( 1 964).
206 Language and its limits
13 See, for example, Fogelin ( 1 976), section 1 . 5 . But see also Candlish ( 1 978).
14 The actual text appears to generate the regress by saying that the sense of the analysans
depends on the truth of the analysandum and hence, presumably, its sense. This detour
through truth strikes me as both incorrect and unnecessary.
The unity of thought 207
The regress is vicious since, if it arose, there would be nothing 'holding all
the constituents together' . The form of a fact must be a quite different
sort of thing from the objects that constitute it. Hence, a fact cannot be
about its own form in the way that it can be about those objects.
15 On the distinction between the two notions of being meaningless, see, for example, Black
( 1 964), p. 1 60.
208 Language and its limits
It might be thought that although a fact cannot be about its own form,
some other fact can be. But we can soon see that this is not true either.
For the form of a fact is not an object at all. As we have just seen, it is the
way that objects are put together, and, as such, a quite different sort of
thing. But if it is not an object, then there are no facts about it, and
therefore, no proposition that expresses such facts. As Wittgenstein
puts it (T4. 1 2 1 ) :
We see, then, that since propositions are not objects they cannot be the
constituents of a fact any more than a form can.
We have now examined two examples of structural claims; these are
quite typical of all similar claims. What we have seen is that attempts to
express them produce claims which violate the canons of logical gram
mar. We are forced to treat as objects things that cannot possibly be
objects, since they have quite different functions (form binds; proposi
tions state). Thus, structural 'facts' cannot be expressed. Attempts to do
so produce something unsinnig. As Wittgenstein summarises the matter
(T4. l 2 1 2):
16 For example, objects are simple ( 1 2.6), but propositions are obviously complex.
17 This argument also goes back to Greek philosophy. Negating can be thought of as a
sense-reversal. And as Aristotle observed (see Geach ( 1 972), p. 45f.) it makes no sense to
negate a name. (See also Ryle (1 990).)
The unity of thought 209
What any picture, of whatever form, must have in common with reality, in order
to depict it - correctly or incorrectly - in any way at all, is its logical form.
When something falls under a formal concept as one of its objects, this cannot be
expressed by means of a proposition. Instead it is shown in the very sign for this
object. (A name shows that it signifies an object, a sign for a number that it
signifies a number, etc.)
210 Language and its limits
18 Ironically enough, Wittgenstein even seems to concede this in the introduction to the
Tractatus, since he says ( p . 4) that the thoughts expressed by the Tractatus are unassail
ably and definitively true - and so not nonsense.
The unity of thought 21 1
Tractatus understands what its statements show, not what they say. This,
however, is quite unsatisfactory. For if the statements of the Tractatus
show anything, then some of the things that are shown are propositional
(see 1 2 . 5). In other words, they must show that something is the case. Let
a be a statement of the Tractatus; and let us suppose that it shows that p.
What is p meant to be?
Perhaps it is a nonsense 'proposition', such as one of the statements
from the Tractatus itself. But this will not work. For if p is nonsense, so is
the suggestion that a shows that p. In this it is no worse than p; but it is
no better either, so any discomfort felt about p will be felt equally about
the claim that a shows that p. Perhaps, then, it is some proposition that is
literally ineffable. This suggestion just does not do justice to the facts.
There is no such proposition; or if there is I have not the remotest idea
what it is - which I should have, since as far as I can tell I have under
stood the Tractatus. Naturally, if p were some ineffable proposition, and
I knew what it was, I could hardly tell you. But that is not the situation. I
literally have no idea as to what should fill the gap if it is an ineffable
proposition. And neither, as far as I know, has anyone else. 19
But of course, we know what must fill the gap: a quite effable proposi
tion. We have seen that it is quite possible to state what can be shown. In
this chapter I have said many such things myself, and in 1 2 . 8 we saw
Wittgenstein doing exactly the same.
Conclusion
Let us take stock. In the first part of the chapter we saw that to account
for the unity of thought Frege postulated the existence of concepts,
unsaturated entities whose distinctive properties account for this unity.
The fact that concepts are not objects, rendered his claims about them
beyond the limit of the expressible. In the second part of the chapter we
saw that the Wittgenstein of the Tractatus accounted for the unity of
thought (the proposition) by invoking the notion of form. The distinc
tive properties of this account for the unity, but also drive statements
about it beyond the expressible.
1 9 Moore «1 990), see especially 1 3 .4 and 1 5 .3) tries to defuse the violation of the Domain
Principle by ZF (see 1 1 .2) by applying a version of the saying/showing distinction. His
solution is to the effect that, although the claim that there is a set of all sets is false, it is all
we can do to articulate an ineffable insight that we have when we reflect on our practice
of quantifying over all sets. Comments similar to the ones in the text apply to his
suggestion. If the insight is not that the collection of all sets quite literally forms a set
(or is a unity, as Moore is inclined to put it), then I have not the foggiest what it is
supposed to be; and neither, I suspect, has he.
212 Language and its limits
Both Frege and Wittgenstein were dealing with the same fundamental
problem here: the unity of thought. 20 In this respect, Wittgenstein's
notion of the form of a proposition is just the intellectual descendant
of Frege's notion of concept. 2 1 In both philosophers, the precise formu
lation of the problem depends on other doctrines; but the problem itself is
a general one which depends on few doctrinal claims. Let me state it as
best I can.
Thoughts are articulated. To form them we must combine simpler
building-blocks. But thoughts are no mere lists of their components.
There must therefore be things which hold them together as unities.
Let us call these (with apologies to modern physics) gluons. Gluons are
not the same kind of thing as the components they glue, and hence not
the kind of thing one can express claims about. But anything can be an
object of thought; in particular, we can think about gluons. Thus they are
the same kind of thing as other constituents, and we can express claims
about them, z2
We have, here, a contradiction typical of the limit of thought. It is also
not difficult to see that it fits the Inclosure Schema, albeit the limit case
where 1jJ(x) is 'x 0' (see 9 . 5) . Let <p(y) be 'y is expressible (in some
=
language, say English)' , so that 0 is the set of all states of affairs expres
sible in English. Take any of the gluons that holds an English proposition
together. And let 8(0) be a situation concerning that gluon, maybe that it
is a gluon of a certain proposition. Then, by the arguments we have had,
8(0) both is (Closure) and is not (Transcendence) expressible.
There is a simple, if bold, way out of the problem I have just explained.
To see what it is, recall Wittgenstein' s argument for the existence of
simple objects ( 1 2.6). This can be reformulated in a very general way
that is quite independent of Tractarian considerations.
Propositions (thoughts, or whatever) are certainly compositional . Now
consider any component of a proposition and ask what determines its
sense (meaning). Possibly, it has the sense it does in virtue of some
relation it bears to something else in language (thought); but this just
raises the question of how this itself has its sense. If we chase back sense
in this way, there are two possibilities. In the first case, the sense of any
20
The history of the problem goes back well before Frege. See Hylton ( 1 984). Russell also
grapples with it in his comments on verbs in sections 52-5 of ( 1 903).
21
As Anscombe observes, ( 1 959), pp. 108f.
22
And if one is tempted by the thought that something can be an object and a gluon, but in
different propositions, just consider, for example, the true (as it happens) statement 'I am
thinking about the the gluon of this proposition (the way that this proposition is com
posed)' . It is essential to the truth of this, that whatever it is that holds the proposition
together is, at the same time, the object of the thought.
The unity of thought 213
Introduction
As I observed in the conclusion of the previous chapter, a contradiction
at the limit of expression arises naturally for any view that postulates
semantic correlates, entities that are meanings, but may be avoided if no
such entities are invoked. It is natural, therefore, to try to avoid the
problem by giving an account of meaning that avoids a postulation of
this kind. In this chapter and the next we will look at theories .that do this.
As we shall see, though they may avoid the contradiction in question,
they do so only at the expense of taking us into other contradictions at
the limit of thought.
The idea that semantic correlates are to be avoided in giving an
account of meaning has been endorsed by a number of philosophers of
language writing since the Tractatus. One of the most important of these
is the modern US philosopher Quine. In this chapter we will consider his
work. We will also look at the philosopher in whom the Quinean pro
gramme reaches its greatest sophistication, Davidson. But let us start
with Quine.
I A slightly different version of this is Quine ( 1 959). The part of the argument relevant to
our present concern is usefully summarised in the first part of ch . 1 of Quine (1 969) .
214
Translation, reference, and truth 215
The copy theory in its various forms [such as that of the Tractatus] stands closer
[than a naturalistic theory] to the main philosophical tradition, and to the attitude
of common sense today. Uncritical semantics is the myth of a museum in which
the exhibits are the meanings and the words are labels. To switch language is to
change labels. Now the naturalist's primary objection to this view is not an
objection to meanings on account of their being mental entities, though this
could be objection enough. The primary objection persists even if we take the
labelled exhibits not as mental ideas but as Platonic ideas or even as the denoted
concrete objects. Semantics is vitiated by a pernicious mentalism as long as we
regard a man's semantics as somehow determinate in his mind beyond what might
be implicit in his overt behaviour. It is the very facts about meaning, not the
entities meant, that must be construed in terms of behavior.
And there are such indeterminate cases: language is shot through with
such indeterminacy, as Quine goes on argue.
216 Language and its limits
utterances o f the language quite differently. A s he puts it, there are ( WO,
p. 73):
rival systems of analytic hypotheses [that] can conform to all speech disposi
tions . . . and yet dictate, in countless cases, utterly disparate translations; not
mere paraphrases, but translations each of which would be excluded by the o ther
systems of translation. Two such translations might even be patently contrary in
truth value provided there is no [sensory] stimulation that would encourage assent
to either.
2 For a less artificial example, the Special Theory of Relativity is, arguably, observationally
equivalent to a classical ether theory that incorporates the Lorentz-Fitzgerald spatio
temporal contractions .
218 Language and its limits
equations that are compatible with all possible observations (since any
observation is performed only to some finite number of decimal places).
In this case, and provided we are physical realists, we may suppose that
there is a determinate law relating a and b, and so rendering one of these
equations determinately true, whatever observational evidence there is.
Similarly, if we suppose that there are objectively existing semantic cor
relates for each word or linguistic expression, we may suppose that one
set of analytic hypotheses is uniquely right. But it is exactly this that is
rendered impossible by the rejection of the 'museum myth' .
3 Though Quine comes dangerously close to this for other reasons. In 'Two Dogmas of
Empiricism' ( 1 9 5 1 ) Quine claims, famously, that, whatever the sensory evidence, any
claim can be held on to if we are prepared to make suitable changes elsewhere in our
web or beliefs.
Translation, reference, and truth 219
4 The later Wittgenstein might be counted as one of this number, as we shall see in the next
chapter. Hylton ( 1 990-1 ) might also fall into the category. According to him, our
language does not mean anything, at least in any transcendental sense.
S See the discussion in I l . l .
6 See, for example, Priest ( 1 987), ch 2.
220 Language and its limits
7 OR contains, also, further arguments to the effect that reference is indeterminate (for
example, based on the Loewenheim-Skolem theorem), but these need not concern us
here.
Translation, reference, and truth 22 1
parts, or wot not. The Thesis generates a contradiction, but this time
concerning reference rather than sense. Objects in the world transcend
anything we can determinately refer to in speaking. Yet, patently, Quine
does refer to rabbits, rabbit parts and other objects in his ruminations on
reference. Even the skeptic about sense would be hard-pressed to deny
this. Indeed, even to claim that one cannot refer determinateiy to objects,
presupposes that we can refer to those objects (and not to undetached
object parts) to say what it is that we cannot refer to! 8 Thus we have a
contradiction at the limits of expression again. The Indeterminacy of
Reference cannot be expressed (Transcendence); yet Quine expresses it
(Closure).
Quine is alive to the fact that there are problems here. He puts what he
takes to be the central one thus (OR, p. 47):
We seem to be maneuvering ourselves into the absurd position that there is no
difference on any terms, interlinguistic or linguistic, objective or subjective,
between referring to rabbits and referring to rabbit parts or stages . . . Surely
this is absurd , for it would imply that there is no difference between the rabbit and
each of its parts or stages . . . Reference would seem now to become nonsense not
just in radical translation, but at home.
8 As Ayer, I am told, once put the matter in a lecture: the claim that there is no fact of the
matter as to whether 'gavagai' refers to rabbits or rabbit parts, depends for its very force
on the fact that 'rabbits' and 'rabbit parts' do refer to distinct things .
222 Language and its limits
Thus, Quine's solution is the heroic one: give up the notion of reference
as meaningless. This was Wittgenstein's move in the Tractatus too (see
1 2.9). Does it work any better for Quine than it did for Wittgenstein? The
answer is 'no' for several reasons.
There is already a large irony in the quotation from Quine. Quine says
that we can make the sense of a query determinate by translating into a
background language. But it cannot do this if the statements of the
background language do not themselves have determinate sense. And,
of course, they do not, according to Quine. But we can exercise charity
here and take this to be a slip of the pen. 9 What Quine is talking about is
reference, not sense. What, then, is to be said about this?
The first thing is that the device of using a relative notion of reference
will not solve the problem that Quine flagged in the last quotation of the
previous section, of explaining how we manage to refer to one thing
rather than another. 1 O Relating the references of terms to those of a
background language can make them determinate only to the extent
that the references of the terms of the background language are determi
nate. And according to Quine they are not. We can attempt to render the
references of these terms determinate by a similar move. But this launches
a regress, and a vicious one. 1 1 We might attempt to stop the regress by, as
Quine suggests at one point, 'acquiescing in our mother tongue and
taking its words at face value' (OR, p. 49). But this hardly helps. Our
mother-tongue is no better off than any other. And the face values of its
terms are no better than the face values of a currency rendered worthless
by inflation. One way to see this is to ask what happens if we start by
asking about the references of the terms in our mother-tongue. The
regress is then a trivial one. 'Rabbit' means rabbit. But this is vacuous,
and no better than answering the question of where the origin of a frame
of reference is by saying that it is at the origin (to pursue Quine's analogy
between background language and frame of reference).
The second point is that we cannot, in any case, jettison an absolute
notion of reference. Our understanding of the way that language works
depends on such a notion. Consider a sentence such as 'Brisbane is in
Australia'. For this to be true (correctly applicable, warrantedly asserti
ble, acceptable, or wot not) is for the city referred to by 'Brisbane' (in the
context of utterance, etc.) to be located in the country referred to by
'Australia' . Thus, one cannot understand the sentence unless one pos
sesses, at least implicitly, the notion of reference. 1 2
It might, I suppose, be insisted that that notion of reference is the
relative one. But then the notion of truth, since it depends on it, must
also be a relative one. Truth, then, is relative to an idiolect, and so to a
person. The view therefore commits one to a relativism about truth of
exactly the Protagorean kind. I argued against this in 3 . 8 , and so shall not
pursue the matter further here. As a matter of fact, Quine himself, else
where, argues against such relativism on self-referential grounds. 1 3
The final and decisive point in the present context is that the relativis
ing move does not, in any case, avoid the problem at the limit of expres
sion that is our central concern. The trap is sprung by pointing out that
Quine himself, in his Thesis of the Indeterminacy of Reference, refers to
objects. If there is a legitimate relative notion of reference, this is beside
the point. And jettisoning an absolute notion of reference even makes
matters worse, since this undercuts Quine's own position. Both the Thesis
(that one cannot refer determinately to objects) and the arguments used
for it presuppose that the absolute notion of reference makes sense. 1 4 As,
indeed, it does.
It could be suggested that the notion of reference in the Thesis is a
relative one. This, I think, would be disingenuous: it is not what Quine
intends. But, in any case, it is not enough. For, in the claim that there is
no fact of the matter as to whether 'rabbit' refers to rabbits, both the
nouns 'rabbit' and '"rabbit' ' ' refer to objects (animals of a certain kind
and words). It might be suggested that these words, too, refer only
relative to a translation into a background language. But this does not
help either. For relative to a translation into a background language,
there is a fact of the matter as to whether or not 'rabbit' refers to
rabbits. Relative to a homophonic translation into English, for exam
ple, it is simply true that 'rabbit' refers to rabbits and not to rabbit
parts. Hence, Quine's Thesis would just be false.
12 The claim that our understanding of how language works depends on a notion of
reference is argued at greater length in Davidson ( 1 977). In that paper he also argues
that the notion of reference may be dispensed with in a certain sense: specifically, that one
should not expect an analysis of the notion of reference independent of its role in a theory
of meaning (truth). It remains true that a notion of reference is essential to such a theory,
if only as a 'theoretical construct'.
13 See Quine (1 975), pp. 327-8 .
14 Hylton ( 1 990-- 1 ) seems to miss this point, though he is vaguely aware that Quine cannot
express his own views. He notes (p. 288) that 'in one sense, there is nothing that Quine is
denying' . The predicament is not unlike that in which the later Wittgenstein finds himself,
as we will see in the next chapter.
224 Language and its limits
15 See Quine ( 1 98 1 ), p. 3 8 .
16
All the papers can be found toge ther in Davidson ( 1 984). Subsequent page references to
Davidson's views are to this. An excellent and simple account of his ideas can be found in
ch. 12 of Hacking ( 1 975). A more general but less readable account occurs in ch. 2. of
Platts ( 1 979).
Translation, reference , and truth 225
17 See, for example, Davidson ( 1 973), p. 1 39 . Although Davidson accepts the indeterminacy
of translation, he rejects, as I understand him, the indeterminacy of reference in the
required sense. Instead, he claims « 1 979), p. 239), we simply have an indeterminacy in
the matter of which language a speaker is speaking. If, however, there is no fact of the
matter as to which of several languages (having the same syntactic words) a speaker is
speaking, then there is, as far as I can see, no fact of the matter as to what a syntactic
word refers to; and so reference is indeterminate.
18 Strictly speaking, this should be 'utterance' rather than 'sentence', but I shall ignore
indexicals and the complications they pose here.
19 One might argue about these desiderata, but such issues will be of no relevance here.
20 Actually it is the Satisfaction Schema rather than the T-schema, but this is a simple
technical complication.
21 To formulate a truth theory for a natural language, the Tarskian techniques need to be
extended to deal with aspects of natural languages not found in the Tarskian paradigms;
this is all part of the Davidsonian programme.
226 Language and its limits
(or at least, a biconditional whose right-hand side means the same as 's is
not true' , and so from which this follows.) This is the instance of the T
schema that generates the Liar paradox. Hence the construction gener
ates a contradiction at the limits of cognition. Nor is this an accident: that
the theory should generate all T-sentences is part of its essence as a theory
of meaning?4 Thus the account of meaning generates another contra
diction at the limit of thought, fitting the Inclosure Schema, as we have
already seen ( 1 0.2).
Davidson is aware of the situation. What is his response? Whilst
claiming uncertainty, he says (pp. 28f.):
The semantic paradoxes arise when the range of the quantifiers i n the object
language is too generous in certain ways. But it is not really clear how unfair
to Urdu or Wendish it would be to view the range of the quantifiers as insufficient
to yield an explicit definition of 'true-in-Urdu' or 'true-in-Wendish' . Or, to put the
matter in another, if not more serious way, there may in the nature of the case
always be something we grasp in understanding the language of another (the
concept of truth) that we cannot communicate to him.
The thought about the quantifiers is not quite right: a Tarskian theory of
truth does not have to have the quantificational resources to turn the
recursive truth conditions into an explicit definition of truth. However,
the central point remains: the notion of truth for a language may always
transcend that language.
Davidson's proposed solution to the problem is essentially a Tarskian
one and, as such, problematic, as we have already seen ( 1 0.7). But, in the
present context, it is more interesting to note that the escape from the
contradiction at the limit of cognition generates a new one at the limit of
expression. For, according to Davidson's suggestion, the notion of truth
in some language, L, is not expressible in L. In particular, 'true-in
English' is not expressible in English (Transcendence). But, of course, it
is expressed in English by 'true-in-English' (Closure). 2 5 To say what
cannot be expressed, one has to express the very thing.
Davidson's suggestion that a speaker of one natural language may
possess a concept of truth that a speaker of another lacks, would also
appear to be beyond the limit of expression (though expressed), at least
according to Davidson himself. In 'On the Very Idea of a Conceptual
Scheme' ( 1 974b) he argues that the claim that one speaker may possess
concepts that another lacks - that is, has a different conceptual scheme -
makes, in the end, no sense (pp. 1 97f.): 26
The argument for this view is rather a long one but, if I understand it
right, turns on the rejection of semantic correlates and the consequent
indeterminacy of translation. If one were to think of concepts as did
Frege, then there would be no problem about the idea that speaker A
possesses a concept that speaker B does not: this is just a relationship of a
certain kind between a speaker and a semantic correlate. But this move is
open to neither Quine nor Davidson. The only way that either can make
25
See Priest and Crosthwaite ( 1 989), pp. 384ff.
26
Quine agrees. See ( 1 98 1 ), p. 42.
228 Language and its limits
sense of the idea is as the claim that (part of) the language of speaker A
cannot be translated into that of speaker B . But this situation can never
arise: it is always possible to translate. There may be a cost: translation
may require us to attribute some odd beliefs to speakers (p. 1 97). If, for
example, we translate 'gavagai' as 'book', and speakers tend to utter it as
rabbits go past, we may have to attribute to them some rather strange
metaphysical beliefs, for example that rabbits have writing in them (in
their entrails?). But strange cultures do have strange metaphysical beliefs;
so this is hardly outrageous . If there were a determinately correct transla
tion, our translation might be wrong. But the existence of a uniquely
correct translation is precisely ruled out by the indeterminacy of transla
tion.
Hence, both the statement that truth-in-English is not expressible in
English, and even the very claim that one speaker has a concept that
another lacks, appear to be contradictions at the limits of expression
(at least within the Quine/Davidson framework) .
Conclusion
We have seen in this chapter how a theory of meaning that jettisons the
notion of semantic correlates may yet engender contradictions at the limit
of thought. The Quine/Davidson theory gives rise to contradictions at the
limits of expression both because of the indeterminacy of sense and
reference, and in an attempt to avoid contradictions at the limits of
cognition.
Although Quine and Davidson both reject the existence of semantic
correlates, they both work very much in the Fregean truth-conditional
tradition. There are, however, approaches to meaning that come at the
notion from a quite different direction. One of these was proposed by the
later Wittgenstein himself. Another is proposed by Derrida, whose work
is inspired by Saussure, rather than Frege. In the next chapter we will
look at these accounts and see whether they fare any better at avoiding
contradictions at the limits of thought.
14 Consciousness, rules , and differance
Introduction
In the last chapter, we looked at the account of meaning of Quine and
Davidson. We saw that it rejected the existence of semantic correlates,
accepted the consequent indeterminacy of sense, and yet still produced
contradictions at the limits of thought. Their account of meaning is not
(necessarily) a behaviourist one, as I observed (in 1 3 . 3 and 1 3 . 6), but it is
certainly a third-person approach to meaning. It takes no account of
what it is like to use language from a first-person perspective; conscious
ness, if you like. Traditionally - at least in the ideationalist tradition of
Locke and others - this was thought to be of the essence of meaning. 1
Hence it may well be thought that we have so far missed an important
element of the story, and that this vitiates any conclusions we might draw
from our consideration of language. The purpose of this chapter is to
address this issue.
We will do so by looking at the views on meaning of two rather
different writers: the later Wittgenstein of the Philosophical
Investigations, and the modern French writer Derrida. These are very
different philosophers, in tradition, style of writing, and focus. Yet in
the present context they are strikingly similar. Both reject accounts of
meaning based on semantic correlates (though neither puts it that way);
both accept the consequent indeterminacy of sense; both take into
account the relationship between consciousness and meaning; and both
adopt a tantalising, but often frustrating, obliqueness to philosophical
issues. 2 In this chapter we will see all these things. We will also see
another property that they share - with each other and with all the
other writers on meaning we have been considering: their views all tres-
pass over the limits of the expressible, and so deliver contradictions at the
limits of thought.
3 I will prefix section references to the Investigations with 'PI'. Quotations are from the
third edition, 1 967.
4 Page references are to this.
5 For example, critiques of Kripke's interpretation can be found in Goldfarb ( 1 985), Tait
( 1 986), and Werhane ( 1 987). Kripke's book initiated a large literature, partly dealing with
his interpretation, partly dealing with his arguments, mainly mixed. For an excellent
discussion and survey (by no means exhaustive) of the literature, see Boghossian
( 1 989). I have resisted the temptation to take on the literature here: to do so properly
would require much more space than is appropriate in this context.
6 See also, for example, PI l O- 1 4.
Consciousness, rules, and differance 23 1
to one of the books on my shelf, and not to the motor bike which I will
ride home?
Well, suppose that a picture comes before your mind when you hear the word
'cube', say the drawing of a cube . In what sense can this picture fit or fail to fit a
use of the word 'cube'? - Perhaps you say: 'It's quite simple; - if that picture
occurs to me and I point to a triangular prism for instance, and say it is a
cube, then this use of the word doesn't fit the picture . ' - But doesn't it fit? I
have purposely chosen the example that it is quite easy to imagine a method of
projection according to which the picture does fit after all.
232 Language and its limits
The picture of the cube did indeed suggest a certain use to us, but it was possible
for me to use it differently . 7
This was our paradox: n o course o f action could be determined b y a rule, because
every course of action can be made to accord with the rule. The answer was: if
everything can be made out to accord with a rule, then it can also be made out to
conflict with it. And so there would be neither accord nor conflict here .
This, then, is the sceptical paradox. When I respond in o n e way rather than
another to such a problem as '68 + 57', I can have no justification for one
response rather than another. Since the sceptic who supposes that I meant quus
[by ' + '] cannot be answered, there is no fact about me that distinguishes between
my meaning plus and my meaning quus. Indeed, there is no fact about me that
distinguishes between my meaning a definite function by 'plus' (which determines
my response in new cases) and my meaning nothing at all .
7 P I l 4 1 goes on: ' Suppose, however, that not merely the picture of the cube, but also the
method of projection comes before our mind? - How am I to imagine this? - Perhaps I
see before me a schema shewing the method of projection: say a picture of two cubes
connected by lines of projection. - But does this really get me any further? Can't I now
imagine different applications of this schema too? - Well, yes. '
8 Some further candidates are discussed in c h . I of Kripke ( 1 982).
Consciousness, rules, and differance 233
Philosophy may in no way interfere with the actual use of language; it can in the
end only describe it . . . It leaves everything as it is. 1 O
But now we have a new problem. One of the language games we play is
about meaning. We say that some utterances have meaning and that
some do not, that 'book' and 'motor bike' mean different things, and
so on. In particular, and according to this, the view that there is no
determinate meaning, or that there is no determinately correct way to
apply the word 'book' is blatantly false. The price of Wittgenstein's
solution is, therefore, that it shows his analysis to be false; or, better, it
leaves him no language in which to express his claims about meaning and,
generally, rule-following, since he has just succeeded in justifying the
linguistic status quo . Kripke puts it nicely as follows (pp. 69f.):
10
Or as he puts it in a different context (PI98): 'every sentence of our language is "in order
as it is"'.
Consciousness, rules, and differance 235
None the less, the point remains: the conclusion that results from the
skeptical arguments, and that Wittgenstein wishes us to grasp, is beyond
expression (Transcendence). Yet it is possible to express it; I have just
done so and so does Kripke (Closure). Hence we have a contradiction at
the limit of expression.
PI 1 1 9 says:
The results of philosophy are the uncovering of one or another piece of plain
nonsense and of bumps that the understanding has got by running its head up
against the limits of language. These bumps make us see the value of the discov
ery.
It is just that when we make the discovery we cannot (but do) express it.
1 1 For a general introduction to Derrida's thought, a number of works can be consulted, for
example, Norris ( 1 987). The ground we will cover is well traversed in Culler ( 1 979).
12 Interestingly enough, one of the influences on Derrida, Nietzsche, employs this technique
for similar reasons, according to one commentator. Nehamas ( 1 985) argues that
Nietzsche espouses a view according to which there is no absolute truth: only relative
truths, 'perspectives' . The trouble is that to describe this view one has to put it forward as
a non-relative truth. We met a simlar problem for Protagoras in 3 . 8 .
1 3 But then, i f Derrida i s right, there is perhaps nothing determinate t o understand.
236 Language and its limits
14 This can be found in the early pages of Derrida ( 1 974), and is usefully summarised on pp.
1 9ff. of ( 1 9 8 1 ) .
15 Interestingly enough, similar thoughts can b e found i n the middle-period Wittgenstein.
See ( 1 975), p. 3 1 7. For a discussion of the similarities between Saussure and Wittgenstein,
see Harris ( 1 988).
Consciousness, rules, and differance 237
In the original context these were obviously meant to bring home the
insignificance of everyone other than Ozymandias; now, they bring home
the vanity of pretension. Derrida calls this phenomenon the iterability of
the text.
To find the concept we therefore have to look to the intentions (i.e.,
consciousness) of the utterer. This, therefore, is a kind of presence
(indeed, the paradigm case for Derrida since, then, meaning is present
in consciousness). But what would we find if we examined the intentions
of the utterer? They would, presumably, be something like: in uttering
such and such I intended to mean so and SO. 1 7 We do not, therefore, find
meanings; just more words (so and so) . And if we were to ask for the
meaning of these words we would merely find more words. Hence, we
never break out of the circle of words into a realm of semantic correlates.
Any word is referred to further words indefinitely, and this relationship
of referral, or, perhaps better, deferral, is just as constitutive of meaning
as Saussure's differences. As Derrida puts it (( 1 98 1 ), p. 26):
16
The notion of a text is much broader than that of language, and comprises other forms of
representation too, such as pictures, adverts, etc. This makes no essential difference to the
points being made here, so 1 shall ignore these other forms.
1 7 Or, in its Gricean variant ( 1 957), intend listeners to come to believe that (I believe that) so
and so, by means of recognising this intention.
238 Language and its limits
14.6 Deconstruction
Before we are finished with Derrida's basic notions, there is another we
need to discuss: deconstruction. Derrida, as I have said, denies presence.
He does not, as far as I am aware, provide a general argument against
presence. Rather, what he does do is take examples of texts which
endorse some notion of presence, either overtly or covertly, and show
that they are self-undercutting in some sense. This is called
'deconstruction'. Though the details of how to deconstruct depend on
the text itself, and so vary from case to case, there is a general frame.
Take a text that endorses some form of presence, 1r. By the Saussurian
semiotics, this draws its sense from its opposite, non-1r. And this means
that the pair (-rr, non-1T) provides, in some sense, an underpinning struc
ture for the text. (There may be other pairs providing other aspects of the
structure.) But the members of the pair are not on equal terms; since it is
1T that is supposed to be doing all the metaphysical work, 1T is the domi
nant member of the pair. In the jargon of deconstruction, it is privileged.
Deconstruction comes in two stages (or perhaps Hegelean moments,
since they may be carried out simultaneously). The first is to reverse the
privileging, to make non-1T the dominant member of the pair. This is done
by showing how, in contrast to the claims of the text, it implicitly shows
that non-1T is the more important and fundamental of the pair. This stage
is called, for obvious reasons, 'reversal' . The next stage, now that both
members of the pair are on an equal footing, is to examine the very
ground of the distinction between 1T and non-1T, and show that it is a
false antithesis. There is a space between them in which some new concept
lies, which both unites and differentiates the pair, but is not reducible to
either. This stage is called 'displacement' . In many respects it is reminis
cent of the third stage of Hegel's dialectic (7. 3) . 1 8 The new concepts in
question might be said to transcend (aufhebt) the relevant oppositions.
They are often called 'undecidables', by a rather shaky analogy with
Goedel's Incompleteness Theorem, and are expressed by words that
occur in the original texts, but which come to take on a whole new
18
Though see Derrida ( 1 98 1 ), p. 43.
Consciousness, rules, and differance 239
19 A summary is given on pp. 1 26--7 of Hart ( 1 989); another can be found in Johnson
( 1 98 1 ). Interestingly in the present context, Staten ( 1 984) (especially ch. 2) argues that
the Investigations is a text that deconstructs the Tractatus. Despite the fact that the
Wittgenstein of the Investigations is undoubtedly against what Derrida is against, I
find Staten's argument that the methodology of the Investigations is a deconstructive
one unpersuasive. It does not seem to me to fit the pattern I have sketched.
240 Language and its limits
20
In this, it is not unlike Sextus' strategy. See 3.4.
Consciousness, rules , and differance 24 1
14.8 DifJerance
We are not finished with the limit of the expressible yet, though. For
there is a more fundamental way in which Derrida's work transcends the
bounds of expression. It does so when it deals with difJerance. Difjerance
is a problematic notion, as numerous commentators have noted.
According to Derrida, it is the structure that gives rise to meaning; it is
the precondition of any meaning at all. It itself is beyond expression: it
cannot be described in any way? ' As he puts it (( 1 982), p. 26):
That being said - and on the o ther hand - to remain in this phase [of deconstruc
tion, viz . , reversal] is still to operate on the terrain of and from within the
deconstructed system. By means of this double, and precisely stratified, dislodged
and dislodging, writing, we must also mark the interval between inversion, which
brings low what was high, and the irruptive emergence of a new 'concept' , a
concept that can no longer be, and never could be, included in the previous
regime.
For the second observation, note that the totality of all linguistic
entities, that is, textuality itself, is structured by a certain binary opposi
tion. What all such entities have in common is their presupposition of
(the) metaphysics (of presence); crucially, the sign itself, of which all
linguistic expressions are composed, is the metaphysical notion par excel
lence. As Derrida puts it (( 1 978), pp. 280f.):
21
In ( 1 982), p . 6, Derrida likens the problem of speaking about differance to the problem
that negative theologians such as Cusanus (see \ . 8) have in talking about God.
242 Language and its limits
makes the opposition of presence and absence possible . Without the possibility of
differance, the desire of presence as such would not find its breathing space . This
means by the same token that this desire carries in itself the destiny of non
satisfaction. Differance produces what it forbids, makes possible the very thing
it makes impossible .
22 Derrida's view here is very similar to Kant's view concerning the Transcendental Illusion
(see 6 . 1 ) .
Consciousness, rules, and differance 243
But if the words are not to be taken at face value, what are they to be
taken to mean? In an interview with Ronse, Derrida takes up the matter
as follows (( 1 98 1 ), p. 1 4) :
I try t o write (in) the space i n which i s posed the question o f speech and meaning.
I try to write the question: (what is) meaning to say? Therefore it is necessary in
such a space, and guided by such a question, that writing literally mean nothing.
Not that it is absurd in the way that absurdity has always been in solidarity with
metaphysical meaning. It simply tempts itself, tenders itself, attempts to keep
itself at the point of the exhaustion of meaning.
When stripped of its hedging, what Derrida is saying is that his own
writing is meaningless. Hence, his reaction is exactly that of the
Wittgenstein of the Tractatus ( 1 2.9) and Quine ( 1 3 . 5). But, just as with
Wittgenstein and Quine, this is quite unsatisfactory. For Derrida is mak
ing himself understood, and so is not saying meaningless things . Indeed,
his statements about differ ance enter into the play of differances consti
tuted by discussions of texts (his own and others), and so are meaningful
by his own account, in exactly the same way that all other statements are.
Hence, any honest appraisal of the situation must admit that state
ments about differance are expressible. We therefore have a contradic
tion typical of a limit of thought. Claims about difference are not
expressible (Transcendence); yet they are expressed (Closure).
In fact, the contradiction fits the Inclosure Scheme in a simple way. Let
<p(y) be: y is a linguistic expression. Let 'IjJ(x) be: x is structured by some
binary opposition. Let 8(x) be some statement that concerns the notion
undecidable in terms of such an opposition. (Such statements would
typically occur in any text that deconstructs x.) As we have seen, if x is
a text structured by some binary opposition, 8(x) cannot be expressed in
244 Language and its limits
Conclusion
This completes our review of the later Wittgenstein and Derrida, and,
more generally, of modern theories of meaning that jettison semantic
determinacy. As we have seen, all such theories end in contradictions
at the limit of expression. There is also a further striking fact. All of
the theories of meaning that we have looked at in this part of the book
render some important states of affairs ineffable (whilst managing to
express them). And in all cases the states of affairs in question are ones
that are, in some sense, about notions that make expression in language
itself possible: the unity of the proposition (Frege, earlier Wittgenstein),
reference (Quine), truth (Davidson, albeit in an attempt to avoid contra
dictions at other limits), rule-following (later Wittgenstein), difJerance
(Derrida) . We have here, in all cases, an inclosure contradiction of the
kind we already noted in the Conclusion of chapter 1 2. <p(y) is 'y can be
expressed in language' (where y is some state of affairs), so that n is the
totality of things that can be expressed; 'lj;(x) is 'x n'; <5(n) is s, where
=
I have not tried to show that any half-way adequate theory of meaning
must end up in the same situation; I am not sure how one could go about
showing this . But the fact that each theory renders inexpressible facts
about what makes language possible - and that the theories are of such
different kinds - does suggest that there is something about the very
possibility of language that is like this. At the heart of language lies
what language cannot express (though each theorist may get a different
fix on it, as it were) . I think that there is probably much more to be said
about this; but at present, what this is, I do not know.
One could also, I am sure, take issue with many of the details of each of
the cases that I have discussed. However, in the present context, it is not
the details, but the over-all pattern that is both salient and important. It
is a pattern that we have witnessed at many points in the book. Assuming
the determinacy of sense, there are contradictions at the limits of expres
sion. There is a way out of the contradiction; notably, jettison the deter
minacy of sense. Yet ways of doing this merely take us into other
contradictions at the limit of expression. What to make of this pattern?
I will take this up in the Conclusion of the book.
particular, it is now the pertinent facts about the totality of languages of all discourses
that generate the contradiction. Formally, we parameterise the predicate cp(y) so that
cpj(Y) now reads 'y can be expressed in discourse i' (where the indices range over all
discourses) . If Sj is the state of affairs like s except that it relates to the language of
discourse i, we can now suppose that �CPj(Sj) but, quite consistently, that CPj/(Sj). But
now define cp*(y) as 3icpj(Y), and let s* be the corresponding state of affairs as it bears
on discourses in general. s* obviously expresses a proposition in some discourse, so
cp*(s*). But, by whatever argument we used before to establish Transcendence,
�cp*(s*). Hence we still have a contradiction at the limits of thought instantiating the
Inclosure Schema.
Conclusion
1 The pattern
In the chapters of this book we have traced notions of four limits of
thought: the limit of iteration, the limit of cognition, the limit of concep
tion (definition), and the limit of expression. We have tracked them
through Western philosophy, from its origins in ancient Greece to the
present. As we have seen, they have long been noted as the locus of
paradox and contradiction; and many who have discussed situations
that involve them have ended up (often unwillingly) in self-contradic
tion. This is hardly surprising if, as I argued in the Conclusion of chapter
1 1 , the limits themselves are contradictory. I take the contradictory
nature of these limits to be the best explanation of the continual histor
ical recurrence of this phenomenon; and therefore confirmed by it.
Further confirmation is provided by the fact that when people have
taken steps to avoid these contradictions the contradictions have reas
serted themselves elsewhere . In trying to avoid a contradiction at the limit
of the knowable, Sextus generated one at the limit of the expressible (3 .4);
Kant attempted to solve the Antinomies (contradictions at the limit of
the iterable) by invoking the distinction between phenomena and nou
mena (6.7), and in so doing generated contradictions at the limits of the
conceivable and expressible (5 . 5); Russell attempted to solve the para
doxes of self-reference (contradictions at all the limits), but merely suc
ceeded in generating another contradiction at the limit of expression
(9. 7). Theories of meaning that postulate semantic correlates generate
contradictions at the limit of expression (Conclusion of chapter 1 2);
but jettisoning semantic correlates just produces other contradictions at
the limits of thought (Conclusions of chapters 1 3 and 1 4). And several
solutions to the paradoxes of self-reference that we have looked at merely
reproduce the problem (9. 8 , 1 0.7, 1 1 .4).
Like the air in a partially inflated balloon, if the contradictions are
pressed down at one point, they come up at another. One might call this
the law of conservation of contradiction. If contradiction is inherent in
250 Conclusion
the very nature of the subject, this is hardly surprising: we should expect
contradictions to appear in any discussion of the topic which comes close
to doing justice to the facts.
2 Limitative theorems
The phenomenon of persistence is, in fact, already well recognised in
discussions of the semantic paradoxes: attempts to solve these paradoxes
always produce further paradoxes. Herzberger ( 1 980- 1 ), for example,
discusses the issue.
To see how and why this arises, let us start, first, with the limitative
theorems of modern logic. 1 It is a commonplace to observe that the
paradoxes of self-reference underlie many of these. For example, if a
theory could express its own truth predicate we would have (other things
being equal) the Liar paradox. Hence, no theory can express its own truth
predicate (if consistency is to be maintained): Tarski's Theorem. If a
theory containing arithmetic were axiomatic it would be able to express
its own proof predicate; we would then have a version of the Knower
paradox (Goedel's paradox; see 1 0.2). Hence, the theory is not axiomatic
(if consistency is to be maintained): Goedel's first Incompleteness
Theorem. Moreover, any attempt to show a (suitably rich) theory to be
consistent must be able to express the provability predicate for that
theory. This cannot be expressed in the theory itself, or we would have
the same paradox. Hence, any attempt to show a theory to be consistent
must use resources that outstrip those of the theory in question (if it is
consistent): Goedel's second Incompleteness Theorem. Hence, in both
Tarski's Theorem and Goedel's second Incompleteness Theorem, the
limitative result is simply that a certain notion cannot be expressed in
the theory if contradictions at the limits of thought are to be avoided. 2
Now, attempts to solve the semantic paradoxes need to show that the
semantic notions of English (or some other natural language) are, despite
appearances, consistent. This requires us to formulate a theory about
how those notions behave, and establish its consistency, normally by
producing a semantic interpretation for it. By the limitative theorems,
this is going to require resources beyond those available in the theory. It
is unsurprising, then, that each attempt to solve the paradoxes generates
1 For an informal discussion of these see, for example, Fraenkel, Bar-Hillel, and Levy
(1 973), chapter 5, section 7. For technical details, see Boolos and Jeffrey ( 1 9 74), chapters
I S, 1 6.
2 If we allow that the relevant theories may be inconsistent, and so represent the contra
dictions, then the limitative results fail. See Priest ( 1 994b).
The persistence of inclosure 25 1
semantic machinery not available in the theory. But where is the theorist
to locate this machinery? It can usually be incorporated into the theory
itself in a natural way; but, in this case, paradoxes (often of the extended
variety) arise - the persistence of contradiction; or else the notions in
question can be relegated to a distinct metatheory. 3
As far as solving the paradoxes goes, the first move is obviously fatal.
But so, in its way, is the second: for the aim of the exercise was to show
that the semantic notions of English are consistent, but we now have
more semantic notions than can be expressed in our original theory,
and so our theory is wrong, or at least, incomplete. We might attempt
to avoid the problem by showing that the object theory plus meta theory is
consistent, but we are now off on a regress . Moreover, this fails. The
move is a form of parameterisation and merely regenerates the problem
at the limit, as we saw ( 1 0.6, 1 0. 7) - the persistence of contradiction.
This is inevitable: if we are trying to find a solution to the semantic
paradoxes of English then the metalanguage must be one for English.
Hence, if consistency is to be maintained, the relevant notions cannot be
expressed in English (Transcendence). But the notions are always
expressed in English (Closure). Hence, attempts to avoid these contra
dictions at the limits of thought ultimately end up in contradictions at the
limit of expression - the persistence of contradiction.
We have already seen an example of this with Davidson and 'true-in
English' ( 1 3 . 7). A second example is Herzberger himself in another of his
papers ( 1 970) . In this, he analyses a paradox he calls the paradox of
grounding (a semantic version of Mirimanoffs paradox; see 8 .4, 8 . 5).
His conclusion (in the form of a pertinent limitative theorem) is that
the notion of a predicate of a language being grounded cannot be
expressed in that language . The irony of this is that Herzberger has no
difficulty explaining what it is for a predicate of English to be grounded,
so showing that it can be expressed in English.
Here, he means the totalities of all truths, all knowable things, etc. (There
is no problem about collections of some of them.) His solution to the
problem is, therefore, of the most familiar kind.
In the Grim universe, then, there is no totality of all entities of the
pertinent kinds. But the Grim universe is grimmer than this. We saw
( 1 1 . 1 ) that solutions that deny the existence of such totalities run foul
of Cantor's Domain Principle. Grim concedes that quantification over
these totalities is also impossible. He even produces a nice variation of
Russell' s paradox to show that one cannot quantify over all propositions
(pp. 1 1 3ff.). Essentially, it goes as follows. Suppose that you can quantify
over all propositions, and consider the proposition that all propositions
are not about themselves. If it is not about itself, then, since it is about
those propositions that are not about themselves, it is about itself.
Conversely, if it is about itself then, since it is not about those proposi
tions that are about themselves, it is not about itself.
We can tighten up the argument at the same time as showing that the
contradiction here is an inclosure paradox (in fact, one that fits Russell's
schema), as follows. We will say that proposition p is about proposition q
(Apq) iff p is of the form Vxa(x), and a(q). Now let rp(y) be 'y is a
His resolution of the 'tension' is to insist that what his arguments show is
that (p. 1 22):
the very notion of all truths - or of all propositions . . . - is itself incoherent .
5 The situation is not unlike that in which Russell found himself. See 9.7.
254 Conclusion
6 One might say, I suppose, that the whole story just becomes one of Grim(m)'s fairy tales.
The persistence of inclosure 255
7 Section 1 .7 of Grim's book contains a short critique of the line on the semantic paradoxes
that I have advocated. A minor criticism given is that the line results in a 'bewildering
flurry' of contradictions (p. 27). A number of the contradictions he cites depend on the
contraposibility of the T-schema, which I endorsed in the postscript of Priest ( 1 979a), but
came to reject in the more systematic account of ( 1 987) (see especially 5 .4). Whether the
contradictions that remain are bewildering is a subjective matter, not an objective one. His
major criticism is that one ought not to accept a contradictory account whilst a consistent
account (namely his) is available. As we have seen, it is not.
256 Conclusion
Many people have the urge to push their powers as far as possible, and
so explore their limits. This is obviously true of the physical powers
exercised in sports and other tasks of endurance; but it is equally true
of our powers of thought, such as conceiving, describing, expressing, etc.
Exploring a limit of thought has a curious Looking-Glass property,
however. In exploring the limit, one perforce finds oneself on the other
side. Hegel put it thus: 8
great stress is laid on the limitations of thought, of reason, and so on, and it is
asserted that the limitation cannot be transcended. To make such an assertion is
to be unaware that the very fact that something is determined as a limitation
implies that the limitation is already transcended.
in order to be able to draw a limit to thought, we should have to find both sides of
the limit thinkable (i .e. we should have to be able to think what cannot be
thought) .
Quite so. Whereof one cannot speak, thereof one has just contradicted
oneself.
8 Miller ( 1 969),
p. 1 34.
9 Pears and McGuinness ( 1 9 6 1 ), p. 3.
Bibliography
Aczel P. ( 1 988), Non- Well-Founded Se ts, CLSI Lecture Notes, no . 14, Stanford .
AI-Azm S . J. ( 1 972), The Origins of Kan t 's A rguments in the An tinomies,
Clarendon Press.
Alston W . ( 1 956), 'Ineffability' , Philosophical Review 6 5 , 506-22.
Anderson A., Belnap N., and Dunn J . M. ( 1 992) , Entailment: the Logic of
Relevance and Necessity, vol. 2, Princeton University Press.
Anscombe G. E. ( 1 959), An In troduction to Wittgenstein 's Tractatus, Hutchinson
and Co.
Armstrong D . M . ( 1 965), Berkeley 's Philosophical Writings, Collier-Macmillan.
Bailey C . ( 1 970), 'Continuity and Infinity in Space ' , pp . 32-7 of M. Capek (ed. ) ,
The Concepts of Space and Time, Boston Studies i n the Philosophy o f
Science, vol. 2 2 , Reidel .
Barnes J . ( 1 979), The Presocratic Philosophers, Routledge and Kegan Paul.
(ed .) ( 1 984) , The Complete Works of A risto tle, Princeton University Press.
Barwise J. and Etchemendy J. ( 1 987), The Liar, Oxford University Press.
Beck L. W. (trans.) ( 1 950), Prolegomena to any Fu ture Metaphysics, Bobbs
Merrill.
( 1 966), The Second Analogy and the Principle of Indeterminacy' , Kant-Studien
57, 1 99-205 ; reprinted in T. Penelhum and J. MacIntosh (eds.), The First
Critique, Wadsworth, 1 969.
Bell J . and Machover M . ( 1 977), A Course in Mathematical Logic, North Holland
Publishing Co .
Bennett J. ( 1 966) , Kan t 's Analytic, Cambridge University Press.
( 1 974) , Kant 's Dialectic, Cambridge University Press.
Black M. ( 1 964) , A Companion to Wittgenstein 's Tractatus, Cambridge University
Press.
( 1 968), 'Frege on Functions', in E. Klemke (ed .), Essays on Frege, University of
Illinois Press .
Boghossian P. ( 1 989), The Rule-Following Considerations' , Mind 9 8 , 507-49.
Bolzano B . ( 1 950), Paradoxes of the Infinite, Routledge and Kegan Paul.
Boolos G. and Jeffrey R. ( 1 974) , Computability and Logic, Cambridge University
Press.
Bostock D. ( 1 973-4) , 'Aristotle, Zeno and the Potential Infinite', Proceedings of
the A risto telian Society 74, 37-5 3 .
( 1 988), Plato 's Theaetetus, Clarendon Press.
Brandom R. ( 1 983), 'Asserting' , Noils 1 7 , 637-50.
258 Bibliography
Kaplan D. and Montague R. ( 1 960), 'A Paradox Regained', Notre Dame Journal
of Formal Logic 1 , 79-90; reprinted as ch. 8 of Montague's Formal
Philosophy, Yale University Press, 1 974.
Kemp Smith N. ( 1 923), Commentary to Kan t 's Critique of Pure Reason, second
edition, Macmillan .
(trans.) ( 1 93 3 ) , Immanuel Kan t 's Critique of Pure Reason, second edition,
Macmillan .
Kenny A. ( 1 969), The Five Ways, Routledge and Kegan Paul.
Kerford G. B. ( 1 967), 'Protagoras of Abdera', pp. 505-6 of vol. 6 of Edwards
( 1 967) .
King H. R. ( 1 956), 'Aristotle Without Prima Materia ', Journal of the History of
Ideas 1 7 , 370-8 9.
Kneale W. and Kneale M . ( 1 962) , The Development of Logic, Oxford University
Press.
Koenig J . ( 1 905) , 'On the Foundations of Set Theory and the Continuum
Problem', in van Heijenoort ( 1 967) .
Kretzmann N. (ed .) ( 1 982), Infinity and Continuity in Ancien t and Medieval
Though t, Cornell University Press.
Kripke S . ( 1 975), ' Outline of a Theory of Truth', Journal of Philosophy 72, 690-
7 1 6; reprinted in Martin ( 1 984) .
( 1 982), Wittgenstein on Rules and Private Languages, Basil Blackwell.
Latta R. ( 1 898), Leibniz: The Monadology and o ther Philosophical Writings,
Oxford University Press.
Lear J . ( 1 977), ' Sets and Semantics' , Journal of Philosophy 74, 8 6- 1 02 .
( 1 979-80), 'Aristotelian Infinity' , Proceedings of the A risto telian Society 8 0 ,
1 8 7-2 1 0 .
Lehrer K. ( 1 990), The Theory of Knowledge, Routledge and Kegan Paul.
Leisenring A. C . ( 1 969), Mathematical Logic and Hilbert 's c-Symbol, MacDonald
and Co.
Lewis D. ( 1 99 1 ) , Parts of Classes, Basil Blackwell.
Lukasiewicz J . ( 1 953), 'The Principle of Individuation ' , Proceedings of the
A risto telian Society, Supplementary Volume 27, 69-8 2 .
( 1 97 1 ), ' O n the Principle of Contradiction i n Aristotle ' , Review of Metaphysics
24, 48 5-509.
Mackie J. ( 1 982), The Miracle of Theism, Clarendon Press.
Maddy P. ( 1 988), 'Proper Classes', Journal of Symbolic Logic 48, 1 1 3-39.
Martin G. ( 1 955), Kan t 's Metaphysics and Theory of Science, Manchester
University Press.
Martin R. L. (ed. ) ( 1 984), Recen t Essays on Truth and the Liar Paradox, Oxford
University Press.
Mayberry J . ( 1 977), 'On the Consistency Problem for Set Theory: an Essay on the
Cantorian Foundations of Classical Mathematics (I) ' , British Journal for the
Philosophy of Science 2 8 , 1 -34.
McDowell J . ( 1 973), Plato; Theaetetus, Clarendon Press .
( 1 976), 'Truth Conditions, Bivalence and Verificationism', ch. 3 of G. Evans
and J. McDowell (eds.), Truth and Meaning, Cambridge University Press.
Meyer R. K. ( 1 973), 'Intuitionism, Entailment, Negation ' , ch. 10 of H. Leblanc
(ed .), Truth, Syntax and Modality, North Holland Publishing Co.
262 Bibliography
( 1 976), ' Relevant Arithmetic' , Bulletin of the Section of Logic, Polish A cademy
of Sciences 5 , 1 3 3-7 .
( 1 987), 'God Exists! ' , Noils 2 1 , 345-6 1 .
Meyer R . K . and Mortensen C . ( 1 984), 'Inconsistent Models o f Relevant
Arithmetic', Journal of Symbolic Logic 49, 9 1 7-29 .
Miller A. V . (trans.) ( 1 969), Hegel 's Science of Logic, Allen and Unwin.
Moore A. W . ( 1 985), ' Set Theory, Skolem's Paradox and the Tractatus, Analysis
45 , 1 3-20 .
( 1 990), The Infinite, Routledge .
Murdoch 1. E. ( 1 968), 'On " Equality" of lnfinities in the Middle Ages' , A ctes du
Xle Congres In ternational d'Histoire des Sciences, Ossolineum.
( 1 98 1 ) , ' Henry of Harclay and the Infinite' , in A. Maieru and A. Paravicini
Bagliani (eds.), Studi sui XIV seculo in memoria di Anneliese Maier, Rome.
( 1 982), 'Infinity and Continuity' in N. Kretzmann, A. Kenny, and 1 . Pin borg
(eds .), The Cambridge History of Later Medieval Philosophy, Cambridge
University Press.
Nagel T. ( 1 986), The View from Nowhere, Oxford University Press.
Nehamas A. ( 1 985), Nietzsche: Life as Literature, Harvard University Press.
von Neumann 1 . ( 1 925), 'Eine A xiomatisierung der Mengenlehre' , Journal fiir die
reine und angewandte Mathematik 1 54, 2 1 9-40; reprinted in van Heijenoort
( 1 967).
Norris C. ( 1 987), Derrida, Fontana Press.
Orenstein A. ( 1 977), Williard van Orman Quine, Twayne Publishers (G. K. Hall
and Co . ) .
Parkinson G. H . ( 1 973), Leibniz: Philosophical Writings, Dent and Sons.
Parsons C . ( 1 9 7 1 ) , 'Ontology and Mathematics', Philosophical Review 80, 1 5 1-76;
reprinted as ch. 1 of ( 1 98 3 ) .
( 1 974a), 'Informal Axiomatization, Formalization, and the Concept of Truth' ,
Syn these 27, 27-47; reprinted as ch. 3 of ( 1 98 3 ) .
( 1 974b), 'Sets and Classes', Noils 8 , 1 - 1 2 ; reprinted as c h . 8 of ( 1 98 3 ) .
( 1 974c), 'The Liar Paradox', Journal of Philosophical Logic 3 , 3 8 1-4 1 2 ; re
printed with an appendix in ( 1 98 3 ) and Martin ( 1 984).
( 1 977), 'What is the Iterative Conception of Set?' in R. E . Butts and 1 . Hintikka
(eds. ) , Logic, Foundations of Mathematics and Computability Theory, Reidel;
reprinted as ch. 10 of ( 1 98 3 ) .
( 1 983), Mathematics in Philosophy, Cornell University Press.
Passmore 1 . ( 1 970), Philosophical Reasoning (second edition), Duckworth.
Peano G. ( 1 906), 'Addition E ' , Rivista di Mathematica 8, 1 43-57.
Pears D . F. and McGuinness B . F. (trans . ) ( 1 9 6 1 ) , Tractatus Logico-
Philosophicus, Routledge and Kegan Paul.
Pitcher G. ( 1 977), Berkeley, Routledge and Kegan Paul .
Platts M. ( 1 979), Ways of Meaning, Routledge and Kegan Paul.
Pollock 1. ( 1 970), 'On Logicism', ch. 24 of E. D. Klemke (ed . ) , Essays on Bertrand
Russell, University of Illinois Press.
( 1 986), Con temporary Theories of Knowledge, Rowman and Littlefield .
Popkin R. ( 1 967), ' Skepticism', pp. 449-6 1 , vol . 7 of Edwards ( 1 967) .
Prawitz D. ( 1 965), Natural Deduction: a Proof- Theoretic Study, Almqvist and
Wiksell.
Bibliography 263
References are to pages, sections, and chapters. The table of contents can
also be consulted.
266
Index 267
Frege, G., 1 30, 1 62, 1 97, 202, 204, Hazen, A., 169
2 1 1 , 2 1 2, 2 1 4, 220, 225, 227, 228, Hegel, G. W. F . , 5, 7, 25, 77, 79, 8 1 ,
244 90, 1 1 2, ch. 7 , 1 2 5 , 1 8 6, 2 3 8 , 239,
on concepts, 1 2 . 1 - 1 2 . 2 , 2 1 2, 2 1 3 25 5f.
Frow, J . , xiv on the infinite, see True infinite,
Furth, M . , 1 9 8 False infinite
-'s, dialectic, 7 . 3 , 1 1 9, 238
Galileo, G., 1 2 1 Heidegger, M . , 243
Gallois, A . , xv Heijenoort, J. van, 1 3 5 , 1 36, 1 7 3 , 1 74,
Garver, N., 229 1 80
Gavagai, 2 1 6, 220, 22 1 , 223, 228 Heraclitus, 1 2 , 1 6
Geach, P., 1 9 8 , 208 Heron, J . , 2 3
Generated infinity, see Infinity Herzberger, H . , 1 67, 250, 25 1
Generator, see Infinity generator Heterological paradox, 1 56, 1 59, 1 60,
Genesereth, M . , 48 1 6 3 , 165, 1 67
Gluon, 2 1 2 Hick, J . , 36, 37, 40
God, 1 1 , 23ff. , 4 3 , 44, 7 1 , 8 2 , 9 1 , 92, Hilbert, D . , 69, 70, 72, 7 3
93, 95, 98, 1 07, 1 09, 128, 1 34, Hopkins, J . , 6 2
2 1 7, 2 1 8 , 24 1 Hossack, K . , xv
cosmological argument for the Hughes, G. E . , 1 67
existence of, 27, 2 . 7-2 . 8 , 1 02, Hume, D . , 5 1 , 1 1 3 , 2 1 7
105 Hylton, P . , 2 1 2, 2 1 9, 223
ontological argument for the
existence of, 6 1 , 4. 1-4 . 3 , 1 09 Illl - 0 bjects, 1 79ff.
Goedel, K . , 1 50, 1 56 Idealism, 6 1 , 65, 7 1 , 77, 8 3 , 9 3
-'s incompleteness theorems, 2 3 8 , Inaccessible cardinal, 1 76f.
250 Inclosure, 1 47 , 1 48 , 1 49, 1 50, 1 5 1 ,
-'s paradox, 1 59, 2 5 0 1 54, 1 57, 1 5 8 , 1 6 1 , 1 62, 1 72, 1 73 ,
Goldfarb, W., 230 1 1 . 6, 1 84, 1 8 6 , 1 87, 1 8 8-94, 1 97,
Goldstein, L . , xiv 243, 244, 249, 252, 253, 25 5
Gregory of Rimini, 3 5 Inclosure schema, 3, 9 .4, 9 . 5 , 149,
Grene, M . , 229 1 54, 1 5 5 , 1 57 , 1 58 , 1 60, 1 6 1 , 1 62,
Grewe, R., 1 8 2 1 6 8 , 1 70-3 , 1 84, 1 86, 1 8 8-94,
Grice, P . , 2 3 7 2 1 2, 226, 243 , 244
Griffin, J . , 205 Inconceivability, see Conceive
Grim, P., 1 62, 2 5 1 ff. Indefinability, see Definability
Ground, see Dependence (causal) Indefinitely extensible totalities, 1 1 . 3
Grue, 1 49 Indeterminacy, see Determinacy
Griinbaum, A . , 8 8 Ineffability, 23f. , 6 1 , 20 1 , 203, 2 1 l f.
Gupta, A., 1 67 2 1 2, 244
Guthrie, W. C. K . , 1 7 , 1 8 , 2 1 Inexpressibility, see Expressibility
Infinity, 3, 9, 2 3 , 24, 25, ch. 2, 43, 47,
Hacking, 1 . , 224 229 72, 75, 84, 97, 98, 99, 1 0 1 , 1 02,
Hallett, M . , 127, 1 2 8 , 129, 1 30, 1 32, 1 0 8 , 1 1 0- 1 5 , 1 2 1 , 1 4 1 , 148, 1 52,
1 34, 1 36, 1 37, 1 38 , 1 39, 180 175, 1 7 8 , 206, 207, 2 1 3 , 2 1 7, 255
Halmos, P., 33 absolute, ch. 8 . , 141, 1 44, 1 45, 1 47,
Harris, R . , 236 1 74, 1 75, 1 76, 1 77ff. , 1 8 3f. , 255
Hart, K . , xiv, 239, 244 actual (completed, categorematic),
Hart W., 1 37 1 1 , 2 1 , 2 . 2-2 .6, 37, 97f. , 1 07,
Hawking, S . , 98 1 1 0, 7.5, 8.7, 8 . 8
270 Index