June 16, 2004

The Paradoxes of Denotation

Graham Priest

1 Introduction
This talk is about those paradoxes of self-reference that deploy the
notion of denotation, reference, and related notions—paradoxes such
as Berry’s and Richard’s. These paradoxes are rarely singled out for
special treatment. It is often assumed that if one can provide a method
for solving other semantic paradoxes of self-reference, it will dispose
equally of these. This tendency has been accentuated in recent years
by logicians’ obsession with the liar paradox. Solve that, and the rest
will take care of themselves. This is, in fact, far from the case.
One important feature of the paradoxes of denotation is that they
employ descriptions, or something logically equivalent, in an essential
way. Of course, other paradoxes can be set up employing descriptions.
We might formulate the liar paradox in the following way, for example.
The fourth sentence of this paragraph is not true. But descriptions
can always be avoided in these contexts. This is not the case with the
paradoxes of denotation. The essential use of descriptions provides an
extra dimension along which a solution may be sought; but it also means
that considerations and constructions that are deployed in connection
with, e.g., the liar paradox are by no means guaranteed to transfer—at
least in a straightforward way.
Another important feature of the paradoxes of denotation is as fol-
lows. Most of the semantic paradoxes instantiate the naive bicondi-
tional that governs some semantic notion, such as truth or satisfaction,
to generate something of the form α ↔ ¬α. From this, by principles

Self-Reference.
Thomas Bolander, Vincent F. Hendricks,
and Stig Andur Pedersen (eds.).
Copyright c 2004, CSLI Publications.
111
June 16, 2004

112 / Graham Priest

such as the Law of Excluded Middle, α∧¬α follows. Thus, for example,
truth seems to be governed by the T -schema. If we write ‘x is true’ as
T x, and use h·i to indicate an appropriate naming device, we can write
this as T hαi ↔ α (where α is any closed sentence). We instantiate the
schema with a sentence, λ, of the form ¬T hλi to get T hλi ↔ ¬T hλi,
and so T hλi ∧ ¬T hλi.
Now, the notion of denotation seems to be governed by the D-
schema. If we write ‘y denotes x’ as Dyx, then we can write this as
Dhtix ↔ t = x (where t is any closed term). But though the paradoxes
of denotation may make use of the schema, they proceed by giving
independent argument for each arm of the contradictory conclusion,
α ∧ ¬α. The arguments do not go through an equivalence of the form
α ↔ ¬α.1
Differences of the kind just indicated have important consequences.
In what follows, we will explore these. It should be said, straight away,
that—with the exception of the material in the appendix to the paper—
there is nothing new in what follows: everything can be found some-
where in the literature. However, I think it worth drawing the points
together here, since their import is generally insufficiently appreciated.

2 Standard Examples and their General Features

Let us start with a brief reminder of the standard paradoxes of deno-
tation. The best known paradoxes of these are Berry’s, König’s, and
Richard’s. To these I add a fourth: Berkeley’s.2 (There are, of course,
others. We will note some in due course.)
Berry’s Paradox. Any language with a finite vocabulary, such as En-
glish, has a finite number of names of any pre-assigned length. (I use
‘name’ here to refer to any—non-indexical—designator, not just proper
names.) A fortiori, there is only a finite number of names that length
that refer to natural numbers. Hence there is a finite number of num-
1. You argue that there bers that can be referred to by name of less than, say, 100 words. There
is only a finite number must therefore be a least. By construction, this number cannot be re-
of numbers that can be ferred to by any name of less than 100 words. But ‘the least number
referred to by names that cannot be referred to by any name of less than 100 words’ refers
with a given upper to it. And this has only 16 words.
bound on their length, König’s Paradox. König’s paradox is similar to Berry’s, but since we
but ’100 words’ does have a non-denumerable infinitude of ordinals to play with, in some
not give such an upper ways it is easier. By a very standard cardinality argument, any language
bound. One could use
1 Most of the set-theoretic paradoxes work via an equivalence of the form α ↔ ¬α,
’100 symbols’ instead.
too. A notable exception is Burali-Forti’s paradox.
2 See Priest (1995a), 4.4-4.9, 9.3.
 June 16, 2004

The Paradoxes of Denotation / 113

with a finite vocabulary, such as English, has a countable number of

names. Since the number of ordinals is not denumerable, there are
ordinals that are not referred to by any name. Hence there is a least.
By construction, this is not denoted by any name. But ‘the least ordinal
that is not referred to by any name’ refers to it.
Richard’s Paradox. Richard’s paradox is constructed employing diago-
nalisation. As for König’s paradox, the number of names in English is
countable. Order these in a definable way, say lexicographically. Let us
write the ith place in the decimal expansion of a real number, n, as ni .3
We now define the real number, r, by diagonalisation as follows. r is 2. Footnote 3: The
the number such that: (i) if the ith name in the enumeration does not single superscript dots
refer to a number, then ri is 5; (ii) if the ith name in the enumeration have been replaced by
does refer to a real number, s, then if si 6= 5, ri = 5 else, ri = 6. ellipses, since the dots
r is distinct from every number referred to on the list, since it differs where hardly visible.
in some decimal place. Hence it is not on the list. Yet the italicised Furthermore, ’infite’ has
description does refer to it and is one of the names on the list. been replaced by
Berkeley’s Paradox. There is an infinitude of objects in the world. You ’infinite’.
will never think about most of them, so there will be many objects that
you will never think about. Consider any one such—εx(you will never
think about x)—where this is an indefinite description. By definition,
you will never think about this. But if you have been following what I
have written, you just have.
It might be thought a mistake to classify Berkeley’s paradox with
the other three, since it does not make use of the notion of denotation
explicitly. But in fact, there are good reasons to group it with the better-
known paradoxes. To think of (conceive of) an object—at least in the
sense relevant to this paradox—is to bring before the mind something
that represents it. In particular, one may bring before the mind a name
which denotes the object. Thus if the representation is a description,
as it is in the case of Berkeley’s paradox, the notion of denotation is
deployed implicitly.
Secondly, all the standard paradoxes of self-reference—set-theoretic
and semantic—share a common form. All the paradoxes instantiate the
structure of an inclosure. This is constituted by a function, δ(X), and
predicates, ϕ(x) and ψ(X), such that:
1. Ω = {x : ϕ(x)} exists, and ψ(Ω)
2. If X ⊆ Ω and ψ(X) then:
3 The decimal expansion of a number may not be unique. Thus 0.3000 · · · is the

same as 0.2999 · · · . In case the expansion of a number is not unique, we always

mean the one with the infinite string of zeros.
June 16, 2004

114 / Graham Priest

(a) δ(X) ∈
/X (Transcendence)
(b) δ(X) ∈ Ω (Closure)
Contradiction arises for the case when X is Ω; for we then have δ(Ω) ∈
Ω ∧ δ(Ω) ∈ / Ω.4
Berkeley’s paradox is clearly in this family. Given its use of descrip-
tions, it is also in the same sub-family as the other denotation para-
doxes, as the following table illustrates (where µ is the least-number
operator, and, in Richard’s paradox, D(x, X) expresses the construc-
tion which diagonalises x out of a definable set, X).5

δ(X) ϕ(x) ψ(X)

x is a nat. no.
X is definable
Berry µx(x ∈
/ X) definable in
in < 90 words
< 100 words

x is a definable
König µx(x ∈
/ X) X is definable
ordinal

x is a definable
Richard ιxD(x, X) X is definable
real number

x is, was, or will

Berkeley εx(x ∈
/ X) X is conceivable
be thought about

3. Concerning table: All

symbols in first row are Before we proceed to look at the paradoxes themselves, let me make
now in bold face (rather one further comment arising from this classification. There is a general
than just some of the principle of adequacy on paradox-solutions, which can be called the
symbols). Vertical Principle of Uniform Solution (PUS ): same kind of paradox same kind
spacing of first row of solution. Putative solutions that apply to only some of a family
corrected. As before, of paradoxes cannot be getting to the heart of the matter. I will not
’words’ should probably discuss the Principle here,6 but simply note a corollary: we have a right
be replaced by 4 For full details concerning these and the other paradoxes of self-reference, see
’symbols’. Priest (1995a), chs. 9–11. Yablo’s paradox also fits the schema. See Priest (1997a).
5 One might argue that indefinite descriptions are not really denoting terms. I

think that this view is incorrect, but, if one does subscribe to it, Berkeley’s paradox
can be reformulated in various ways using only definite descriptions. Consider, for
example, the first thing thought about by Ceasar on the Ides of March that will
never be thought about by you—or if, by chance, everything thought about by
Caesar on that day is something you have thought or will think about, just choose
somebody else.
6 It is discussed at length in Priest (1995a), 11.5. See also 17.6 of the second
 June 16, 2004

The Paradoxes of Denotation / 115

to expect a uniform solution not only to the paradoxes of denotation,

but to all of the semantic paradoxes.7 Anything that does not apply to
all of them must be latching on to accidents, not essences.

3 Berry’s Paradox in More Detail

The precise structure of the paradoxical argument in the case of Berke-
ley’s paradox is simplicity itself. If we write Cx as ‘x is being thought
about (conceived)’, then it concerns a term εx¬Cx. Since, as a mat-
ter of fact, not everything is conceived of, ∃x¬Cx. By an appropriate
description principle, ∃xα(x) → α(εxα(x)), it follows that ¬Cεx¬Cx.
But because we can (and do) bring this description before the mind, we
have Cεx¬Cx. The structures of the arguments in the other paradoxes
enumerated are more complex, but it will be helpful for what follows
to spell out at least Berry’s paradox in a little more detail.
Let N (y, z) mean that the number of symbols in y is less than z.
Let β(x) be the formula ¬∃y(Dyx ∧ N (y, 100)) (‘x is not denoted by a
name with less than 100 symbols’). Considerations of size allow us to
establish that:
∃xβ(x)
Let d be the description:
µxβ(x)
Then by the appropriate description principle, ∃xα(x) → α(µxα(x)):
¬∃y(Dyd ∧ N (y, 100))
But by simple counting, it is easy to establish that N (hdi, 100). And
since d = d, the D-schema gives us: Dhdid. Putting these two things
together gives us: Dhdid ∧ N (hdi, 100). Thus:
∃y(Dyd ∧ N (y, 100))
Hence we have a contradiction.
Note that we have employed the D-schema and a description princi-
ple, but otherwise the logic we have employed is pretty minimal—just
the law of identity and conjunction introduction. Conceivably, it might
be thought that the law of identity fails if it concerns a non-denoting
term. But the very fact that ∃xβ(x) shows us that the term d denotes.
Hence, d = d does not fail on this count.8
edition.
7 Indeed, to all the inclosure paradoxes, semantic and set-theoretic.
8 A fuller formalisation, including the argument for the truth of ∃xβ(x), is given

in the appendix to ch. 1 of Priest (1987). This uses slightly more logic, but no
inferences that are invalid in, say, First Degree Entailment. In particular, the Law
of Excluded Middle is not used.
June 16, 2004

116 / Graham Priest

4 General Solutions
With this preliminary material out of the way, let us move on to look
at solutions to the denotation paradoxes. Do standard solutions to the
semantic paradoxes apply to the paradoxes of denotation? This is not
the place to review all standard approaches. We will look at just three
kinds of solution: Kripkean, Tarskian, and dialetheic (in that order).
One might object to all these solutions in their own terms. I shall not
be concerned with such objections here. My concern is whether these
solutions can be applied to the denotation paradoxes at all.
Kripkean Solutions. What characterises Kripkean solutions is the use of
a non-classical logic with truth value gaps. This may be supplemented
with a story about the conditions under which truth-value gaps arise.
(In Kripke’s case, this is his account of grounded and ungrounded sen-
tences.) The central feature of such accounts is the failure of the Law
of Excluded Middle: α ∨ ¬α. As I have already observed, in, say, the
Liar Paradox, the Law is applied to an equivalence of the form α ↔ ¬α
to give α ∧ ¬α.9
This solution cannot be applied uniformly to the paradoxes of de-
notation, for the simple reason that they do not all use the Law of
Excluded Middle. This is patent in the case of Berkeley’s paradox, but
as the formalisation of the previous section demonstrates, it is equally
4. Footnote 10: ’emply’ so of Berry’s.10
has been replaced by This conclusion might be thought puzzling. Doesn’t Kripke give us a
’employ’. construction which produces a consistent model of the T -schema; and
can’t this be applied in the same way to give us a consistent model of
the D-schema? So how can inconsistency arise? The construction cer-
tainly can be applied in a natural way. But, first, it does not validate
the T -schema or the D-schema, merely the rule-form of each. Secondly,
the trouble is that the denotation paradoxes employ not just the D-
schema, but also descriptions, and the most natural ways of treating
descriptions and denotation ensure that monotonicity, and so the iter-
ative construction that depends on this, fails.11
9 If one supervaluates over the gaps, one may retain the Law of Excluded Middle

itself. But then other pertinent principles of inference give way.

10 Richard’s paradox does require the Law of Excluded Middle, since this makes

use of definition by cases. The business part of König’s paradox does not employ
it, but it is possible, as far as I know, that every proof of the uncountability of the
ordinals (from which it follows that there are undenoted ordinals) deploys the Law.
11 There are other ways, which do deliver monotonicity, as is demonstrated by

Kremer (1990) and Kroon (1991). But one has to insist that non-denoting inputs
give truth-valueless outputs. Thus if t is any non-denoting term and m is any de-
noting term, t = m is valueless, as is Dhtim. One cannot therefore say truly that
¬∃xDhtix, as one should be able to do.
 June 16, 2004

The Paradoxes of Denotation / 117

Tarskian Solutions. Let us move on to solutions of the Tarskian variety.

According to these, we must enforce a hierarchy of metalanguages, Li
(i > 0). If Ti is the truth predicate of Li , it is guaranteed to satisfy the
T -schema only for sentences of lower languages. If one can construct
a sentence, α, of the form ¬Ti hαi, this is a sentence of Li , and of no
lower language. Hence the argument to contradiction is blocked. We can
apply the same idea to the paradoxes of denotation. This time, each
Li —which now contains descriptions—contains a denotation predicate,
Di , which is guaranteed to work only for names of lower languages. De-
scriptions containing Di itself are not such names. Thus, in the for-
malisation of the Berry paradox for level i, if di is the description
µx¬∃y(Di yx ∧ N (y, 100)), though we may have di = di , we cannot
apply the D-schema of level i to infer Di hdi idi .
The solution cannot be employed uniformly, however. Berkeley’s
paradox stands out. If we try to apply the Tarskian strategy to this
paradox, the hierarchy of truth/denotation predicates gives rise to a hi-
erarchy of “thinking of” predicates, Cn —whether these are now thought
of as belonging to different languages is, in fact, irrelevant. We can
argue as before that, for any i, ∃x¬Ci x; hence, by an appropriate de-
scription principle, ¬Ci εx¬Ci x. But now the impossibility of using this
description in the appropriate schema metamorphoses into the claim
that one can think of an object specified by a level i description only
by a thought of a higher level; and thus if Cj εx¬Ci x then j > i. But
nothing would seem clearer than that one can have a thought whose
content is itself. I can, for example, think of this very thought. (I am
thinking of this thought; therefore I exist.) Even granted that thoughts
are stratified into levels, this must be a thought at some level. Suppose
that it is i. Then we have Ci n where n = εxCi x.
Dialetheic Solutions. Finally, and briefly, let us consider dialetheic so-
lutions. In such solutions, the proofs of contradiction are allowed to
stand; but the contradictions are quarantined by the use of an appro-
priate paraconsistent logic. Clearly, there is nothing in the denotation
paradoxes we have examined to undercut this solution. One may legit-
imately ask, however, if the effects of paradox are really quarantined.
The obvious arguments to triviality are broken, but are there other and
more subtle arguments? I will return to this question later.

5 Solutions Concerning Descriptions

Unlike the more standard paradoxes of self-reference, the paradoxes
of denotation use descriptions essentially. Perhaps the solution to the
paradoxes lies in the fact that these descriptions misbehave in some way.
June 16, 2004

118 / Graham Priest

For example, perhaps the descriptions in question fail to refer. Thus, for
example, the description employed in Berry’s paradox, d, is ‘the least
number not referred to by a name with less than 100 symbols’. If this
description fails to refer, then, one might argue, because of reference-
failure, d = d is untrue. Alternatively, existential generalisation may
fail for terms such as d. Either of these facts would break the paradox
argument. Let us, then, see whether the blame can be laid at the feet
of descriptions.

Ambiguity. It may be suggested that a name may have more than one
referent. ‘Aristotle’, for example, refers to more than one person. Nor-
mally, we resolve such ambiguity by fixing the context. But maybe
even when we have fixed the context, a name may have more than
one denotation. For example, why can I not simply christen every ob-
ject in the world, and a fortiori every number, with the name ‘Bruce’.
In this case, every natural number, real number, ordinal, or object in
general, is referable/conceivable, and the paradoxes lapse. (Richard’s
paradox lapses because the diagonal definition assumes that denotation
is unique.)
The solution does not stand up to inspection, however. As is so often
the case, it merely gives us the wherewithal to reformulate the para-
doxes. Consider Berry’s paradox as an example. Call a name univocal if
it has a single denotation. The number of univocal names with less than
100 symbols must be no greater than the number of names with less
than 100 symbols. Hence there must be numbers that are not referred
to by them. Now consider the least number that cannot be referred to
by a univocal name with less than 100 symbols. The name just used
refers to it; and being a definite description, it is univocal. Hence it
is referred to by a univocal name. Similar reformulations apply to the
other paradoxes.
In reply, it might be suggested that perhaps definite descriptions
are not univocal. A different theory of how they function is required.
Whilst any evaluation of such a view would have to await the theory
in question, the suggestion would seem to fail. Even if ordinary definite
descriptions may have more than one denotation, it would seem to be
within our power to create univocal names of the required kind. Thus,
let us define an operator, ι∗ , that works as follows: if there is a unique
thing satisfying a(x) let ι∗ xa(x) refer to it and only it. Otherwise. . . (it
refers to nothing, or a unique non-existent object, or whatever). Now,
taking Berry as an example again, consider ι∗ x(x is the least number
not referred to by a univocal name with less than 100 symbols). This
univocally refers to it.
 June 16, 2004

The Paradoxes of Denotation / 119

Contexts. In the previous suggestion context made a brief appearance.

Maybe it can be deployed in a different way. It is clear that the deno-
tations of some terms depend on context. (Consider, e.g., ‘the present
president of the USA’). Perhaps, in these paradoxes, the crucial number
is not definable in one context, but definable in another.12
The suggestion is something of an act of desperation. The descrip-
tions employed in the denotation paradoxes do not contain indexicals
of any obvious kind, and hence do not appear to change their deno-
tation from context to context. It must be insisted that ‘definable’ is
itself context-sensitive. In fact, more than this, it must be insisted not
only that it is context-sensitive, but that the context changes in the
course of the argument—something for which there is, in general, no
clear rationale.13 5. Footnote 13: ’100
But in any case, there would appear to be formulations of the para- words’ should probably
dox that circumvent the issue of context. Call a number Definable (with be ’100 symbols’.
a capital ‘D’) in i words if it can be defined in some context or other by
a name of i words. We can simply run, e.g., Berry’s paradox employing
the notion of Definability. We consider the least number not Definable
by less than 100 words. It is defined by that description in this par- 6. ’100 words’ should
ticular context, and so in some context. Hence, it is Definable in less probably be ’100
than 100 words. But now, it will probably be asked, how do we know symbols’. Three
that there are numbers that are not Definable in less than 100 words? occurrences.
Maybe every number can be so Defined in some context. And in that
case, ∃x¬∃y(Dyx ∧ N (y, 100)) is false, and µx¬∃y(Dyx ∧ N (y, 100))
is an empty term. Now, it may be the case that every natural num-
ber can be referred to by some description with less than 100 words in
some context or other, but it seems implausible to suppose that one
might avoid the version of König’s paradox which employs the notion
of Definability in the same way. The number of ordinals is so large that
it would seem to be impossible that every one can be referred to in
some context. A context is fixed by a bunch of parameters: time, place,
speaker, audience, etc. If this is the case, the number of contexts has

12 A solution along these lines is offered by Simmons (1994).

13 One way to bring this home is simply to make it explicit that the context is
fixed. The paradoxes then reappear. Thus, take Berry’s paradox for example. The
place is here; the time is now; I am the speaker; you are the audience; the topic of
discourse is Berry’s paradox, and I am giving a version of it. Call this context c. The
version of the paradox I give is as follows. There is only a finite number of names
with less than 100 words. A fortiori, the number of numbers that I can refer to in
this context, c, is finite. Consider the least number that I cannot refer to (in this
context). By construction, I cannot refer to it (in c). But I have just referred to it
by ‘the least number that I cannot refer to in this context’. Similar reformulations
apply to the other paradoxes.
June 16, 2004

120 / Graham Priest

some determinate cardinality, as, therefore, does the number of things

that can be referred to in a context. But the number of ordinals is
greater than this. It would seem that this conclusion can be avoided
only by, again, playing fast and loose with the notion of context. Con-
siderations of the kind we have just been considering apply similarly to
Berkeley’s paradox.
There may well, of course, be more to be said about each of the
possibilities I have just discussed. The subject being what it is, I am
sure there is. Equally, there may be different stories as to why something
else to do with descriptions invalidates the arguments. Such possibilities
have to be treated on their merits. But let me conclude with one final
observation. If one solves the paradoxes of denotation by appealing
to some doctrine specifically about naming—perhaps something about
how descriptions work—then, since the other paradoxes of self-reference
do not employ this notion, the solution will not be applicable to them.
Such solutions will therefore fly in the face of the Principle of Uniform
Solution.

6 A Paradox of Hilbert and Bernays

The discussion of the paradoxes of denotation has taken us quite a
long way. But we have not finished yet. This is because there is a
paradox of denotation that I have not mentioned so far; and in some
ways this paradox is more virulent than the ones I have discussed. It was
formulated originally (as far as I am aware) by Hilbert and Bernays.14
In a nutshell, it goes as follows. Consider the name ‘1 + the denotation
of this name’. Suppose it denotes n. Then it also denotes 1 + n. Since
denotation is unique, n = n + 1. And so, if you like, 0 = 1.15
To make matters more precise, let me give a semi-formalisation of
the paradox. By standard techniques of self-reference, one can construct
a term, t, such that t has the form 1 + µxDhtix. Now, t = t, and hence:
t = 1 + µxDhtix
But by the D-schema, Dhtit. Hence:
∃xDhtix
By the properties of the least number operator, DhtiµxDhtix. So by
the D-schema once more, t = µxDhtix. Substituting back in (1) gives:
t = t + 1.
14 For
references and discussion, see Priest (1997b).
15 A
variation of Hilbert and Bernay’s paradox is given by Simmons (1994). Es-
sentially, he considers a term, t, of the form: the sum of all the numbers denoted by
terms in the set {h1i, hti}. If hti denotes n, then n = n + 1. If hti fails to denote,
hti simply denotes 1. Contradiction in either case.
 June 16, 2004

The Paradoxes of Denotation / 121

The result is classically intolerable. It might be thought that one

could tolerate the result if one uses a paraconsistent logic. So what if
∃x(x = x + 1)? After all, we still have ¬∃x(x = x + 1). Indeed, there are
even formal inconsistent arithmetics where both hold.16 The trouble is
that the above argument is just an example of Hilbert and Bernays’
paradox. Instead of 1 + x, we could use any arithmetic function, f (x),
to infer, in exactly the same way, that existence of a term, t, such that 7. Should ’that
t = f (t). Now let f be the function such that: existence’ maybe be ’the
if x > 0 then f (x) = 0, else f (x) = 1 existence’ ?

If t = 0, then 0 = f (0) = 1; if t > 0 then t = f (t) = 0 and we are back

with 0 = 1 again. So the result cannot be accepted even by a dialetheist
about numbers (unless they are of a very extreme kind!).
What to say about this paradox? There is, in fact, a quite obvious,
and as far as it goes, adequate, response. (Another dialetheic response
is discussed in the appendix to this paper.) This is simply to deny
that the term t denotes. After all, if it did denote, it would have to
denote both a number and its successor. And if t doesn’t denote, we
can fault the argument in a couple of places. We can deny that t = t.
Alternatively, we can deny the instance of existential generalisation.
Unfortunately, we can formulate a version of the paradox that takes
the possibility of non-denotation into account. We simply use a notion
of description that guarantees denotation. Specifically, let us define the
description operator µ∗ as follows:
µ∗ xα(x) = µx (∃yα(y) ∧ x = µxα(x)) ∨ (¬∃yα(y) ∧ x = 0)

Using the Law of Excluded Middle, it is easily shown that:

∃x (∃yα(y) ∧ x = µxα(x)) ∨ (¬∃yα(y) ∧ x = 0) (1)
∗
Hence, all µ -terms denote. Moreover, by the appropriate description
principle:
(∃yα(y) ∧ µ∗ xα(x) = µxα(x)) ∨ (¬∃yα(y) ∧ µ∗ xα(x) = 0)
Hence, if ∃yα(y) it follows that:
µ∗ xα = µxα(x). (2)
∗ ∗
We now run the argument using µ -terms. We construct a term, t , of
the form 1 + µ∗ xDht∗ ix. Since all µ∗ -terms denote, we have t∗ = t∗ ,
and hence:
t∗ = 1 + µ∗ xDht∗ ix
16 See, e.g., Priest (1997c). In such arithmetics, although ∀x∀y(x + 1 = y + 1 ⊃

x = y) holds, we cannot use ∃x(x = x + 1) to infer that 0 = 1 since the disjunctive

syllogism fails.
June 16, 2004

122 / Graham Priest

But by the D-schema, Dht∗ it∗ . Hence:

∃xDht∗ ix
By the properties of the least number operator, Dht∗ iµxDht∗ ix. So by
the D-schema once more, t∗ = µxDht∗ ix. By an application of (2),
8. Footnote 17: Doesn’t t∗ = µ∗ xDht∗ ix, and so t∗ = t∗ + 1, as before.17
the Tarskian approach Appealing to truth-value gaps does solve this version of the paradox.
always consist in If there are such gaps, we may take it that the Law of Excluded Middle
building a well-founded fails. Hence, we can no longer show (1), and that all µ∗ -terms denote.
hierarchy of In particular, to show that µ∗ xDht∗ ix denotes, we would have to show
meta-languages that ∃xDht∗ ix ∨ ¬∃xDht∗ ix. And there is no way of establishing this.
L1 , L2 , L3 , . . . where Hence, appealing to truth value gaps, and specifically, assuming that
each Li is a the sentence ∃xDht∗ ix is neither true nor false, will solve the problem.
meta-language of Li−1 ? Unfortunately, as we have already seen, it will not solve the ordinary
Here you are talking Berry paradox.
about a non-wellfounded Fortunately, then, dialetheism will solve this version of the paradox
hierarchy of as well—though not by simply accepting the contradiction as true. (As
meta-languages we have seen, this is to no avail.) Given that we now have the Law of
. . . , L−3 , L−2 , L−1 , L0 , Excluded middle, we can establish that every µ∗ -term denotes. But to
which I guess was never establish (2) requires the disjunctive syllogism α ∨ β, ¬α ` β. (We have
Tarski’s intention. to rule out one disjunct to infer the other.) Now the syllogism is invalid
in paraconsistent logics; it fails to be truth preserving if α is both true
9. A non-matched right and false. The α in question here is ¬∃xDht∗ ix ∧ µ∗ xDht∗ ix = 0; and
parenthesis has been this may be both true and false if ∃xDht∗ ix is. Hence, we may solve
removed from the the paradox by supposing this sentence to be a dialetheia.
formula α and a missing Of course, this does not show that the paradox cannot be recon-
’∗ ’ and ’x’ have been structed in some other way. But what can be shown is that assuming
added. that descriptions may fail to denote, one can construct an inconsistent
model of the D-schema, principles concerning descriptions and self-
reference, in which, e.g., 0 = 1 is not true.18 These principles do not,
therefore, engender triviality. (The model has to be inconsistent since
it validates the argument of Berry’s paradox.)
17 It is also possible to give a Yabloesque version of this paradox (as was pointed

out by Simmons in a talk at the conference Heaps and Liars, University of Connecti-
cut, October 2002). A simple version is as follows. Consider a sequence of terms, t i ,
i ≤ 0, such that ti = h1 + δ(ti−1 )i, where δ(t) is the denotation of t if it has one, and
0 otherwise. All the terms in the sequence therefore denote. Suppose that δ(t 0 ) = k.
Either k = 0 or k > 0. In the latter case, k = δ(t0 ) = 1 + δ(t−1 ). So δ(t−1 ) = k − 1.
Either k−1 = 0 or k−1 > 0. In the latter case, carry on in the same way. Eventually,
we must find an i such that δ(ti ) = 0. But then 0 = δ(ti ) = 1 + δ(ti−1 ). Contradic-
tion. (Note that the sequence of terms can be accommodated in an infinite regress
of Tarski metalanguages, Li , i ≤ 0. Hence, a Tarskian approach cannot solve this
paradox.)
18 For details, see Priest (1999).
 June 16, 2004

The Paradoxes of Denotation / 123

7 Conclusion
In this paper, I have reviewed the paradoxes of denotation and some of
the lessons to be learned from them. Most importantly, these paradoxes
have distinctive features that make the applicability of a number of
standard consistent solutions to the other semantic paradoxes highly
problematic when applied to them. Nor do these features appear to give
other satisfactory avenues for consistent solutions. A dialetheic solution
is the only simple and uniform solution to all the paradoxes in question.

Appendix: Another Solution to the Paradox of Hilbert

and Bernays.
10. Numbering of
In this appendix, I will discuss another possible solution to the Hilbert appendix removed.
and Bernays Paradox. This is to the effect that the crucial term has—
not fewer than one denotation, but—more than one denotation. In-
tuitively, this is just as plausible. After all, if the naughty term, t,
denotes some number, it would appear to denote its successor too. But
this second approach will be more acceptable to a Meinongian, who is
committed to the view that all terms denote (though the denotations
might be non-existent).
For a start, what shape does logic have if terms may have multiple
denotations? The most crucial question here is how to define the truth
conditions of atomic sentences with such terms. If we suppose that a
predicate, P , has an extension, P + , and an anti-extension, P − , and
that any term has a set of denotations den(t), then the natural truth
conditions are as follows:
P t1 · · · tn is true iff ∃x1 ∈ den(t1 ) · · · ∃xn ∈ den(tn )hx1 , . . . , xn i ∈ P +
P t1 · · · tn is false iff ∃x1 ∈ den(t1 ) · · · ∃xn ∈ den(tn )hx1 , . . . , xn i ∈ P −
Note that even if P is a classical predicate (so that its extension and
anti-extension are complements), a sentence may still be both true and
false. This approach still, therefore, requires dialetheism.19
Does this approach solve the paradox? In fact, it validates all the
inferences employed in the argument except the substitutivity of iden-
tity. To see why, just consider a simple case of this, the transitivity
of identity. The extension of identity is, as usual, {hx, xi : x ∈ D},
where D is the domain of quantification. Now consider the inference
t1 = t2 , t2 = t3 ` t1 = t3 . Suppose that den(t1 ) = {a, b}, den(t2 ) = {b, c},
den(t3 ) = {c} (where a, b, and c are distinct objects). Then as is easy
to check, t1 = t2 and t2 = t3 are both true, but t1 = t3 is not. Hence,
these semantics break the argument. In fact, it is possible to construct
19 For a full discussion of multiple denotation, see Priest (1995b).
June 16, 2004

124 / Graham Priest

an inconsistent but non-trivial model of the D-schema, the appropriate

description principles and self-reference in a logic of multiple denota-
tion.20 This approach therefore solves the problem too.

References
Kremer, M. 1990. Paradox and reference. In J. M. Dunn and A. Gupta,
eds., Truth or Consequences, pages 33–47. Dordrecht: Kluwer Aca-
11. Replacement: demic Publishers.
’Dordrect’ → Kroon, F. 1991. Denotation and description in free logic. Theoria 57:17–41.
’Dordrecht’. Priest, G. 1987. In Contradiction. The Hague: Kluwer Academic Publishers.
Priest, G. 1995a. Beyond the Limits of Thought. Cambridge: Cambridge
University Press. Second (extended) edition, Oxford: Oxford Univer-
sity Press, 2002.
Priest, G. 1995b. Multiple denotation, ambiguity and the strange case of
the missing amoeba. Logique et Analyse 38:361–73.
Priest, G. 1997a. Inconsistent models of arithmetic, i: the finite case. Journal
of Philosophical Logic 26:223–35.
Priest, G. 1997b. On a paradox of Hilbert and Bernays. Journal of Philo-
sophical Logic 26:45–56.
Priest, G. 1997c. Yablo’s paradox. Analysis 57:236–42.
Priest, G. 1999. Semantic closure, descriptions and triviality. Journal of
Philosophical Logic 28:549–58.
Priest, G. 200a. If all terms denote something then some terms denote more
than one thing. To appear.
Simmons, K. 1994. Paradoxes of denotation. Philosophical Studies 76:71–
106.

20 See Priest (200a).

June 16, 2004