Professional Documents
Culture Documents
is a
proper extension of L, of course, but we should also notice that the portion
of the world relevant to the truth, falsity, or whatever of L
statements is
a proper extension of the portion of the world relevant to the truth, falsity,
etc., of statements of L. After all, we are assuming that our languages
contain the resources to self-refer so, as a result, L
. So L
since
we are assuming that all members of C are parts of S), should we expect
13
In fact, on this view, I am extending the language we speak every time I give a talk
on this topic, or write a new paper. Thanks to Stewart Shapiro for pointing this out.
What is a Truth Value. . . 195
all of the -relations that hold between C
and S
to be contained in R?
The answer, quite obviously, is no. If we are interested in relations of
type that hold between items in two domains, and we extend and thus
further complicate both domains, then in general we should expect that
new relations of type might arise in the more complex case that were not
present in the prior unextended case. The present thesis is merely that what
we should expect in general, with regard to situations with this structure, is
exactly what happens in the particular case under consideration here.
Thus, we can sum up the view being defended as follows: In extending
both the statements of our language and the portion of the world relevant to
semantic evaluation of those statements by adding the linguistic resources for
describing the semantic relations that hold between these two domains, we
have complicated the language/world structure in such a way as to introduce
new semantic relations into the picture.
Before moving on, it will be helpful to look at a particular example. As
before, let us take standard classical Peano arithmetic (PA) as our starting
point. Here, the language is just the standard one for Peano arithmetic, and
the relevant portion of the world is just the -sequence of natural numbers.
As a result, we nd that a classical, bivalent semantics suces for inter-
preting the language. In other words (and with the caveat from the end of
1 above in mind), there are only two relationships that can hold between
statements of PA and the structure of natural numbers: either what the
statement says is the case (it is true), or what the statement says fails to be
the case (it is false). Note that these truly are relations: What a statement
says will be the case if and only if some specic relation holds between that
statement and the relevant bit of the world, and what a statement says will
fail to be the case if and only if some other relation holds between that
statement and the relevant bit of the world.
But now consider what happens when we extend PA by adding a truth
predicate and a falsity predicate. Note that this in eect extends the portion
of the world relevant to evaluating the statements of our language, since our
language can now be used to make claims about not just numbers, but about
claims about numbers. We can now express the Liar sentence. And the Liar
sentence, on pain of contradiction, enters into neither the what is says is
the case relationship with the world, nor with the what it says fails to be
the case relationship. Instead, this sentence forces us to recognize a new
relationship that can hold between sentences of our extended language and
the relevant bits of the world, something like what it says is the case if
and only if what it says is not the case, which we have been abbreviating
196 Roy T. Cook
above as pathological.
14
Thus, we can add the following platitude to our
platitudes for truth and falsity:
Pathological : A sentence is pathological if and only if: what it says is the
case if and only if what it says fails to be the case.
15
Thus, we have the general picture of how truth values are generated by
semantic paradox and the revenge phenomenon. There are important con-
nections between this picture and one standard way of understanding the set
theoretic paradoxes and the development of Zermelo-Fraenkel set theory in
light of them, however, and it is to this topic that we turn in the next section.
3. Revenge and Indenite Extensibility
One way of describing the innite hierarchy of truth values discussed above
is to observe that the concept Truth Value is indenitely extensible. As a
result of this indeniteness, we can never be in a position to quantify over
14
Given the picture being sketched here, an additional complication arises when we
consider so-called Truth-Teller statements such as:
This statement is true.
Unlike the Liar, which can consistently be assigned neither true nor false, the Truth-
Teller can be consistently assigned either semantic value. Thus, it is in a sense semanti-
cally underdetermined, in the same sense that we might say that the Liar is semantically
overdetermined.
We can think of the relationship that holds between the truth-teller and the world as
something like:
What it says is the case if and only if what it says is the case.
One open question, on the present account, is whether this is the same relation as the
one that holds between the Liar statement and the world in other words, whether on
the present picture the extension of a classical language with a truth predicate and a
falsity predicate introduces one additional relation between language and world, or two.
My intuition is the latter, although I will not argue for that here.
There is of course a generalized version of the question: Given an arbitrary language L,
how many new truth values are introduced when we extend the language by adding the
resources for describing the semantics of L? The answer must be at least one, by the argu-
ments above, but narrowing it down further is a topic that will be postponed to later work.
15
Reasoning classically, this denition of pathological implies that no sentences are
pathological. Thus, it is important to remember that the reason for introducing this
semantic predicate in the rst place is that we have already extended our initial classi-
cal language to one that requires (at least) a three-valued logic. Viewed within such a
many-valued context, the denition does not collapse (see Cook [2] for details).
What is a Truth Value. . . 197
all truth values (and, as a result, we cannot quantify over all statements, or
speak a universal language).
The notion of a concept being indenitely extensible seems to have rst
appeared in the writings of Bertrand Russell on the paradox that bears
his name:
the contradictions seem to result from the fact that . . . there are what
we may call self-reproductive processes and classes. That is, there are
some properties such that, given any class of terms all having such a
property, we can always dene a new term also having the property
in question. Hence, we can never collect all of the terms having the
said property into a whole; because, whenever we hope we have them
all, the collection which we have immediately proceeds to generate a
new term also having the said property. ([13], p. 144.)
The idea that a number of central mathematical concepts, such as Cardi-
nal Number, Ordinal Number, and Set, are indenitely extensible, and
that the set-theoretic paradoxes associated with these concepts are caused
by illegitimately attempting to collect together and/or quantify over the ex-
tensions of indenitely extensible concepts, is now a largely unchallenged
part of philosophical folklore.
Michael Dummett formulated a slightly more precise characterization of
the notion of a concept being indenitely extensible in The Seas of Lan-
guage (the term indenite extensibility was coined by Dummett as well).
Dummett writes that:
The paradoxes both the set theoretic and the semantic paradoxes
result from our possessing indenitely extensible concepts. . . An
indenitely extensible concept is one for which, together with some
determinate range or ranges of objects falling under it, we are given
an intuitive principle whereby, if we have a suciently denite grasp
of any one such range of objects, we can form, in terms of it, a
conception of a more inclusive such range. . . By the nature of the case,
we can form no clear conception of the extension of an indenitely
extensible concept; any attempt to do so is liable to lead us into
contradiction. ([4], p. 454)
We can reformulate Dummetts criteria for a concept to be indenitely ex-
tensible as follows:
198 Roy T. Cook
Definition 3.1. A concept C is indenitely extensible if and only if there is
(we can identify?
16
) a function f mapping collections of objects to objects
such that, given any denite collection D of Cs, f(D) is a C and f(D) is
not in D.
In other words, a concept is indenitely extensible if, and only if, when
confronted by a denite collection of objects falling under the concept, we
can always nd another object not in that collection which also falls under
the concept.
The concept Ordinal is the prototypical case of indenite extensibility:
Given any set (i.e. denite collection) of ordinals, we can identify an ordinal
not in the set for example, the successor of the supremum of the set.
Similarly, we can nd operations that map any collection of sets onto a set
not contained in the collection (e.g., the powerset of the transitive closure
of the set), or that map any set of cardinals onto a cardinal not in that set
(e.g., the cardinality of the powerset of the union of that set of cardinals).
Thus, the concepts Set and Cardinal are also indenitely extensible.
Given the points raised in the previous section, however, the concept
Truth Value is indenitely extensible as well. Constructing the requisite
function is not dicult: Let F map any set of incompatible truth values S
to the truth value of the statement:
Rev: Rev has a truth value in S other than true.
A generalized version of the arguments given in the rst section of this paper
demonstrates that Rev must have a truth value not in S (in addition, the
truth value of Rev cannot be the true!). As a result, we can map any denite
collection of truth values onto a truth value not in that collection. So the
class of truth values is indenitely extensible.
The idea that the semantic paradoxes are caused by indenite extensi-
bility, however, although suggested by Dummett, has not been throroughly
explored. In addition, Dummetts own solution to the paradoxes that arise
as a result of indenite extensibility adopt intuitionistic logic and math-
ematics is not adequate for dealing with the Liar paradox and its vari-
ants, since the contradiction arising from the Liar sentence can be derived
16
The phrase we can identify is inserted to emphasize that there are at least two read-
ings of the notion of indenite extensibility a platonist one, which we are mobilizing
here, and a constructive one, where indenite extensibility is connected to our ability to
identify such a function. The latter reading is likely closer to what Dummett himself had
in mind, although the distinction does not make much dierence in the present context.
What is a Truth Value. . . 199
constructively.
17
The failure of Dummetts response to the indenite exten-
sibility of semantic concepts such as Truth Value does not alter the fact
that these concepts are, in fact, indenitely extensible, however.
The correct response, of course, is to deal with the indenite extensibil-
ity of the concept Truth Value in the same manner that we deal with the
indenite extensibility of other concepts such as Set, Ordinal, and Cardi-
nal. The full account can only be sketched here.
18
Any attempt to describe
the semantics of the language we speak automatically and immediately ex-
tends that language by adding expressive resources required to describe the
semantics, and in particular, the truth values, of the original language. As
a result, the potential to repeatedly carry out such extensions implies that
there is a hierarchy of richer and richer languages, and a corresponding hier-
archy of more and more extensive collections of truth values corresponding
to these languages. We can never reach a point, however, where we have
all of the truth values, or where extensions to our language are no longer
required in order to describe things that we might wish to describe (such as
the semantics of a particular language).
As a result, there is an innite hierarchy of truth values (in fact, a proper
class, since there will be at least as many truth values as ordinals
19
, and the
totality of truth values cannot be collected together into a denite collection
that is, into a set. The similarity between the structure of potential lan-
guages and their truth values and the structure of the set-theoretic universe
(as described by, say, ZFC) is not accidental
20
, and this way of dealing with
17
For a detailed discussion of Dummetts application of indenite extensibility to the
semantic paradoxes, as well as a clear examination of the various factors underlying the
failure of Dummetts strategy, the reader should consult Williamson [16].
18
The reader should consult Cook [2] for a detailed development of a formal semantics
corresponding to this informal sketch.
19
A full explication of the semantic picture sketched here requires a proper class of
truth values. The reason is simple: If our language is rich enough (and part of the point
of the present project is to avoid unmotivated restrictions on the expressive power of our
languages) we can formulate statements such as:
This sentence has one of the truth values that can be modeled
within our current set theory and which is not the true.
Statements like this one that involve quantication over large collections of truth values
(including truth values which might not be required to interpret the language in which such
statements are formulated) allow us to make large jumps up the hierarchy of languages.
See Cook [2] for details.
20
It is worth noting that this parallel between the commonly accepted diagnosis of
the set-theoretic paradoxes and the diagnosis of the semantic paradoxes is accidental, in
a certain sense: I have never found a priori arguments that various paradoxes such
200 Roy T. Cook
the semantic paradoxes has the added advantage of providing a uniform solu-
tion to the set-theoretic and the semantic paradoxes. In the end, both types
of paradox result from the same phenomenon: the indenite extensibility of
concepts used (in an illegitimate manner) in the derivation of the paradox.
As a nal note, we should say something about restrictions on express-
ibility in the present view. Unsurprisingly, there are some things that we
think that we can say (or, perhaps, that we think that we think that we can
say) which we cannot say. In particular, we cannot say all truth values,
or, more carefully, since we did just in fact say (or, actually, write/read) it,
we cannot say what we intend to say when we use this phrase. If this phrase
meant what intuitively one might think it does, then we could construct an
Ultimate Liar sentence:
Ult: Ult has one of the truth values (that is,
one of all of them), other than the true.
In other words, unrestrictedly general quantication is not possible on the
present picture. This, however, is not as worrisome as are, for example, the
restrictions on negation imposed by Graham Priest [11], however, since there
already exist independent reasons for doubting that unrestrictedly general
quantication is possible (see, e.g., the various articles in [12]). Nevertheless,
there is, in a certain sense, nothing that we want to say that we cannot
say on the present picture. While there is no language that can express
everything, since for any language there is an extension of it containing
additional (semantic) vocabulary, there is, on the present view, nothing we
need to say that cannot be said in some language. And it is here where
the real advantages of the present view over traditional approaches, such
as those developed by Tarski [15] and Kripke [8], is most evident. Both
Kripkes account and Tarskis suer from severe limitations on what can be
expressed: Tarskis hierarchy bans self-referentiality (or, at the very least, it
bans self-referential statements involving semantic expressions, such as the
Liar sentence) and Kripkes construction does not allow the introduction of
as the semantic and set-theoretic ones should have a single uniform solution at all
convincing. Instead, such arguments have always struck me as expressing little more than
an aesthetic judgment of the arguer that is, these so-called arguments typically appear
to amount, in the end, to little more than the thought that the universe would be a
prettier place if all of these paradoxes has a single uniform solution. Lacking faith in the
prettiness of the universe, I was immensely surprised when, in developing this view, clear
connections between the set-theoretic and the semantic paradoxes began to emerge. To put
it simply: I take the existence of these connections to be a profound discovery regarding the
underlying shared nature of these puzzles. Discovering such a shared underlying nature,
however, was never, in my view, a desideratum on a successful account of the paradoxes.
What is a Truth Value. . . 201
semantic expressions or connectives necessary to describe the three-valued
construction itself (since such expressions and logical notions are non-mono-
tonic, as Kripke notes). The present approach, however, suers from neither
of these expressive defects, allowing use to say everything (or very nearly
everything) that we think that we think we can say.
Acknowledgements. The present paper has beneted from collegial feed-
back from the philosophy departments at The Ohio State University; St.
Cloud State University; The University of Minnesota, Twin Cities; The Uni-
versity of Minnesota, Duluth; The University of Nevada, Las Vegas; and
The University of St Andrews. Thanks are also due to Peter Hanks, Hannes
Leitgeb, ystein Linnebo, Kevin Scharp, Stewart Shapiro, Greg Taylor, Neil
Tennant, Giulia Terzian, James Woodbridge, and Crispin Wright for helpful
discussions on these issues.
References
[1] Beall, J. (ed.), Revenge of the Liar: New Essays on the Paradox, Oxford University
Press, 2007.
[2] Cook, R., Embracing Revenge: On the Indenite Extensibility of Language, in [1]:
3152, 2007.
[3] Cook, R., Clarifying the Logic of Clear Cases in [10]: 75104, 2005.
[4] Dummett, M., The Seas of Language, Oxford University Press.
[5] Eklund, M., What Vagueness Consists In, Philosophical Studies 125: 2760, 2005.
[6] G odel, K., On Formally Undecidable Propositions of Principia Mathematica and
Related Systems, New York: Dover, 1992.
[7] Greenough, P., Vagueness: A Minimal Theory, Mind 112: 235281, 2003.
[8] Kripke, S., Outline of a Theory of Truth, Journal of Philosophy 72: 690716, 1975.
[9] Lackey, D. (ed.), Essays in Analysis, Braziller, 1973.
[10] Moruzzi, S., and A. Sereni (eds.), Issues on Vagueness, Il Poligrafo, 2005.
[11] Priest, G., Doubt Truth to Be a Liar, Oxford: Oxford University Press, 2006.
[12] Rayo, A., and G. Uzquiano (eds.), Absolute Generality, Oxford Univ. Press, 2007.
[13] Russell, B., On Some Diculties in the Theory of Transnite Numbers and Order
Types, in [9]: 135164, 1906.
[14] Tarski, A., Logic, Semantics, Metamathematics: Papers from 1923 to 1938, Hackett
Publishing Company, 1983.
[15] Tarski, A., The Concept of Truth in Formalized Languages, in [14]: 152278, 1935.
[16] Williamson, T., Indenite Extensibility, Grazer Philosophische Studien 55: 124.
Roy T. Cook
Department of Philosophy
University of Minnesota
819 Heller Hall 271 19th Ave
Minneapolis, Minnesota, USA
cookx432@umn.edu