Remarks on Definitions and the Concept of Truth

Author(s): Anil Gupta

Source: Proceedings of the Aristotelian Society, New Series, Vol. 89 (1988 - 1989), pp. 227-246
Published by: Wiley on behalf of The Aristotelian Society
Stable URL: http://www.jstor.org/stable/4545102
Accessed: 06-05-2015 15:50 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

The Aristotelian Society and Wiley are collaborating with JSTOR to digitize, preserve and extend access to Proceedings of the
Aristotelian Society.

http://www.jstor.org

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 XV*-REMARKS ON DEFINITIONSAND
THE CONCEPT OF TRUTH'
by Anil Gupta
I
I wish to present a view of concepts (and, in particular, of the
concept of truth) that Nuel Belnap and I have come, aftermuch
hesitation, to believe is true.2We were led to consider the view
by a striking parallel that we noticed between the behaviour of
the concept of truth and that of certain kinds of definitions. I
think I can best explain the view if I begin by sketching this
parallel.
The relevant behaviour of the concept of truth is well known:
while some (perhapsmost) of its usesare unproblematic (e.g., its
use in 'some things Russell said about Wittgenstein are true')
some others are not (e.g., its implicit use in Epimenides'
assertion of the Liar 'all Cretans are liars', and its use in the
Truth-Teller 'this very sentence is true').3We have no difficulty
in assigning a truth-value to the former, unproblematic,
sentences (assuming that we know the relevant facts) but with
the latter, pathological, ones we encounter a variety of
difficulties. Sometimes, as in the Liar example, every possible
assignment of truth-value seems to lead to contradictions.

*
Meeting of the Aristotelian Society held at 5/7 Tavistock Place, London WC1, on
19 June, 1989 at 6.00 p.m.
'This paper is based on talks I have given at StanfordUniversity, University of Texas
at Austin, University of Wisconsin at Milwaukee, University of Western Ontario, and
Indiana University. I would like to thank the participants in these talks for their helpful
comments and questions. I would also like to thank the National Endowment for the
Humanities for the support they provided through their Fellowships for University
Teachers programme.
In developing the views reported in this paper I have received invaluable help from
Professor Nuel Belnap. This paper has benefited enormously from the insightful
suggestions he made in the numerous conversations that we have had on the subject.
2 view has its roots in Hans Herzberger'sand our earlier work on the semantic
paradoxes. This work can be found in Journalof Philosophical Logic11, 1982, no. 1.
3 In my discussion of the Liar I shall make the usual
assumptions that 'liar' means
never utters a truth' and that all Cretan utterances,except perhaps the one mentioned
of Epimenides, are false. Further I shall assume for simplicity that the 'objects'of truth
are sentences. The view I shall sketch does not require,however, that this assumptionbe
made.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
228 ANIL GUPTA

Sometimes, as in the Truth-Teller example, no assignmentleads

to contradictions but each appears to be arbitrary. And there
are numerousother possibilities.There is much dispute amongst
philosophersand logicianson how to understandthe pathological
behaviour of the concept of truth, but the existence of the
phenomenon is denied by none.
Let us now look at the behaviour of some definitions.Consider
the following:
(1) x is G =D x is F or x is both H and non-G,
where F and H are some clear and well-understood predicates.
In this definition the definiens
x is F or x is both H and non-G
containsthe definiendum G. Hence the definitionis circularand, by
traditionaldoctrines,illegitimate.Beforewe dismissit altogether,
however, let us look more closely at its behaviour. Its most
striking feature, and a principal reason for its traditional
dismissal, is that it is creative:
in conjunction with the usual rules
of inference it implies substantive, and possibly absurd, claims
about F and H. For, suppose for reductio
(2) x is non-F and x is H.
Suppose further that
(3) x is G.
The traditional logical rules for definitions allow that, given a
definition of the form
(4) x is G X

one can infer the definiens

(5) x-
from the definiendum
(6) x is G;
and, conversely, one can infer (6) from (5). Let us call the first
4For simplicity, I will discuss only definitions of one-place predicates. The points
made below carry over in the most straightforwardway to definitions of many-place
predicates, names, and expressionsof other logical categories.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 229

rule Definiendum Elimination(DfE), and the second Definiendum

(Dfl). Using DIE, we deduce from (3) that
Introduction
x is F or x is both H and non-G.
This in conjunction with (2) implies by the rulesof propositional
logic that x is non-G. Thus from the hypothesis (3) we have
deduced its negation. Hence (2) implies that
(7) x is non-G.
(2) and (7) yield by propositional logic that
x is F or x is both H and non-G.
Applying Dfl to this gives

(8) x is G.
Contradictory conclusions (7) and (8) thus follow from (2).
Hence definition (1) allows us to prove a priorithat all Hs are Fs.
If we let H be 'identical to itself' and F be 'identical to the One',
we can prove a priori the doctrine of Monism.
The reader may have noticed that this argument bears a
striking resemblance to the argument in the Epimenides
paradox. In the latter, as in the former, we deduce on purely
logical grounds a contradiction from a contingent hypothesis.
(In the Epimenides the hypothesisis that Epimenides the Cretan
says 'all Cretans are liars' and all other Cretan utterances are
false.) The role played in the formerargument by the rules Dfl
and DfE is played in the latter by the rules TruthIntroduction (TI)
and TruthElimination(TE).
A 'A' is true
'A' is true TI A TE

There is another way of bringing out the similaritiesbetween

the two cases. Let us view the rules for truth, TI and TE, as
proceduresfor determining whether a sentence is true or not. To
determine the status of 'A is true' they direct us to determine the
status of A, and to determine the status of 'A is not true' they
direct us to not-A. These 'reductions' can be represented
diagrammatically as follows.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
230 ANIL GUPTA

'A' is true 'A' is not true

reduces reduces
to to

1 I

A not-A
The rules for definitions, Dfl and DfE, can similarly be viewed
as procedures for determining whether an object is G.
t is G t is not G
reduces reduces
to to

t not[ t ]
Suppose we try to determine of an object, say a, which is both
non-F and H whether it is G. Definition (1) directs us to
determine whether
a is F or a is both H and non-G.
Since a is known to be non-F and H, the problem reduces to
determining whether a is non-G. So:
a is G
reduces
to

Jp

a is non-G
But when we apply the procedureto 'a is non-G, we are directed
to determine whether
not [a is F or a is both H and non-G].

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 231

Again, since a is non-F and H, this holds if and only if a is G. So:

a is non-G
reduces
to

a is G
We thus enter an unending loop: 'a is G' reduces to 'a is non-G'
and this in turn reduces back to 'a is G. The same kind of
phenomenon occurswith the Epimenides:to deter-minewhether
it is true, the rules direct us to first determine whether it is not
true; and they reduce the second task in turn to the first.In both
cases we obtain the parallel patterns displayed below. (Here L
denotes the Epimenides sentence.)
a is G L is true

a is non-G L is not true

a is G L is true

a is non-G L is not true

Definition (1) does not result in any Truth-Teller-like

phenomenon, but it is easy to construct definitions that do. If
we change the clause 'x is non-U in the definiens of (1) to 'x is U
then 'a is U behaves like a Truth-Teller on all objects a that are

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
232 ANIL GUPTA

both non-F and H. To determine the status of 'a is G we are led

by the rules back to the very same starting point 'a is C. The
'reduction sequences' for 'a is G' and for 'TT is true' (here TT
is the Truth-Teller) are exactly parallel and look like this:
a is G TT is true

1~ 1
a is G TT is true

a is G TT is true

The new definition does not exhibit any Liar-likephenomenon.

Definitions that exhibit both kinds of pathologicality do,
however, exist. An example:
(9) x is G; either (x is F and H) or (x is F and non-H
and G) or (x is non-F and H and non-G).
By this definition 'a is C behaves like a Truth-Teller if a is both F
and non-H. It behaves like the Liar if a is non-F and H. These
examples and others like them lend supportto the thesisthat every
kindof pathologicalbehaviour that the conceptof truthexhibitscan be
mirrored in conceptswith circulardefinitions.
The similarities between concepts with circular definitions
and the concept of truth are not confined to the pathological.
Just as the rules for truth imply that 'true' is unproblematicover
a range of sentences, similarly the rules for definitions imply
that a circular concept G is unproblematic over a range of
objects. Suppose we want to determine whether an object b is G
in the sense of (9). Rules for definitions direct us to determine
whether
(10) either (b is F and H) or (b is Fand non-H and G) or
(b is non-F and H and non-G).

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 233

If b is F and H then (10) holds and we can conclude

unproblematically that b is G. On the other hand, if b is neither F
nor H then clearly (10) fails to hold and we can conclude that b is
not G. So by definition (10) G is unproblematicover objectsthat
are either both F and H or neither F nor H. (Note that by
definition (1) G is unproblematicover objectsthat are either For
non-H.)
Concepts with circular definitions, then, behave in ways that
are remarkablysimilar to the behaviour of the concept of truth.
They exhibit the same kinds of pathological behaviour as truth.
And like truth, they can be, and usually are, unproblematicover
a range of cases. These similarities suggest, first, that the
outright rejection of circular definitions in logic may be too
precipitous. For their behaviour is very much like that of a
concept that we do accept, and wantto accept. Perhaps a more
general logic of definitions is possible that will show us how to
make sense of, and work with, circular definitions. Second, the
similaritiessuggest that the perplexing behaviour of the concept
of truth might be explainable as arisingfromsome circularityin
its definition. Nuel Belnap and I have come to believe, after
much hesitation, that these suggestions are more than mere
possibilities, that they are close to actuality. I cannot discuss in
the space and time allotted to me all the details of the theory of
definitions that we are trying to develop. But I think I can
explain some of the fundamental ideas on which it is built and
some of the philosophically significant consequences t iat follow
from it.'
II
Let us see how one might construct a logical theory that would
do justice to both aspects of circular definitions, their occasional
pathologicality and their content. Such a theory would need to
if
specify
1. the meaning that definitions, circular ones included,
ascribe to their respective definienda;
2. the logical rules for working with definitions.

5Belnap and I are working on a book in which we will discuss the theory in greater
detail.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
234 ANIL GUPTA

Let us begin with the first of these tasks.

Let us accept the natural idea that a definition fixes
completely the meaning of its definiendum. Our problem is to
say how we should think of this meaning.As far as non-circular
definitions are concerned there is no difficulty if we accept the
traditional account. The meaning of a predicate by this account
is a rule that gives the extension of the predicate in all possible
situations. Or, equivalently, the meaning determines the
conditions for the predicate's applicability. For example, the
meaning of 'red', by the traditional account, is a rule that
determines the conditions under which an object counts as red.
Now a non-circular definition enables us to calculate the
extension of the definiendum once we are given the extensionsof
the terms in the definiens.6Hence by the traditional account of
meaning it explains the meaning of the definiendumon the basis
of the meanings of the terms in the definiens.There is a problem
with the traditionalaccount, however, if we want to preservethe
idea that a circular definition also fixes the meaning of its
definiendum. For, a circular definition does not in general
enable us to determine the extension of its definiendum.Like all
definitions, it does provide a rule for determiningthis extension,
once the extensions of all the terms in the definiens are given.
The problem is that as the definiendum occurs in the definiens,
to apply this rule we need to know the very thing we are trying to
determine, namely, the extension of the definiendum. To
capture the meaning that a circular definition ascribes to its
definiendum we need to think of meaning in a different way.
A circular definition, though it may not determine the
extension of the definiendum, does provide a rule that can be
used to calculate what the extension would be once we make a
hypothesisconcerningthe extension of the definiendum.This is the
key, in my view, to the problem of meaning before us.
The meaning a circular definition ascribes to its definiendum,
I wish to suggest, should be viewed as having a hypothetical
character. It does not determine the conditions of applicability
of the definiendum absolutely, but only hypothetically. We
cannot pick a set and say that it is the extension of the
6Here and below I assume that the definition is formulated in an extensional two-
valued language. Parallel things can be said for definitions formulated in other
languages.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 235

definiendum. We can say only that it would be the extension if

such-and-such other set is supposed to be the extension. Let us
clarify this idea through an example.
Consider the definition:
(11) x is G Df (Xis F and everyobjecty that satisfiesthe
conditionyRx is G)or (x is non-Fand Hand
non-G).
Suppose the relevant facts are as follows. We will call this
collection of facts 'M'.
The domain of discourse (more briefly, D)
=A U lb, c, d}, where
A = 12t, al, a2i .. 5 i }5---
The extension or interpretation of F (more briefly, I(F))
-A U {c}
I(H) ={d}
I(R)-{<x, y>: (x=y=c) or (x=ai & y-aj & i<j)}
We can determine the extension of G on the basis of (11), if we
are given the extensions of all the terms in the definiens. The
difficulty, as we noted above, is that since G occurs in the
definiens, we need to know alreadythe extension of Gin order to
determine the extension of G. Suppose, however, that we make
an arbitraryhypothesisconcerning the extension of G. Suppose,
for example, that
(12) I(G)=0.
We can now apply definition (11) to determine which objects
fall in the extension of G, relative to our hypothesis. We need
only determine which objects satisfy the definiens:
(x is F and every objecty that satisfiesthe conditionyRx is
G) or (x is non-F and H and non-G).
It can be verified that these are ao and d. Hence, if we suppose
the extension of G to be 0 then the definition dictates that the
extension of G should be {ao,d}. Another example: if we suppose
that the extension of G is the domain D then the definition
dictates that everything except b and d should fall under the
extension of G. Similar calculations can be done for all possible
hypotheses. Definition (I 1) thus determines a function 6M that

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
236 ANIL GUPTA

takes as input a hypothetical extension X for G and yields as

output 6M(X), the set the definition rules as the resulting
extension of G. Observe that
d E X iff td M(X).
Thus, for all inputs X, 6M(X) ? X; i.e., no X is a fixed point of
6M.7 No matter what we hypothesize the extension of G to be, our
calculation reveals that it should be different.So definition (I 1)
does not enable us to pick one set as theextension of G. But it does
yield a function, 6M, that tells us 'extension-under-various
hypotheses'. This function, I want to suggest, captures the key
semantic information about G. It is the basis on which the
behaviour of G, both ordinary and pathological, can be
understood.
Even though a circular definition ascribesto its definiendum
a meaning with a hypothetical character, still this meaning can
enable us to make categoricaljudgments in some (perhaps not
all) cases.Thetransitionfrom themerelyhypotheticaltothecategorical is
to be madeby quantifyingoverall possiblehypotheses. If under all
possible hypothesesa definition 'yields' that an object falls under
the definiendum G then, clearly, it implies categorically that the
object is G. The main question is: how should 'yields' be
understood?
The suggestionI wish to make here is that to determine what a
definition 'yields' on a given hypothesis we should view the rule
8mprovided by a definition as a ruleof revision;
we should consider
the effects of its repeated applications. If we begin with the
hypothesis that the extension of G is 0 then definition ( 11) rules
that it should be 6M(0) = {ao, d}. But if we take the latter as the
extension, the definition revises it to
6M({ao, dl) = {ao al} [=6N1(6M(0))]-
Now we can apply the definition with {ao, a,} as the hypothetical
extension of G. The result is that the extension of G should be
6,M({ao, al}) {ao, a1, a2, d} [=61(6M(6N1(0)))].
And this process can be repeated ad infinitum.The stages of this
process, I am suggesting, are better and better approximations

'An object X is a fixed point of a function f iff f(x) = x.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 237

of what the definition says concerning the extension of G on the

original 0-hypothesis.8 Observe that some objects (namely the
Ws)eventually always fall in the resulting approximations. We
sav that these are positively stable on the 0-hypothesis. Some
(namely b and c) eventually always fall out of these approxi-
mations. These are negativelystable on the 0-hypothesis. On the
remaining objects (in our example, d) the pattern is forever
unstable; sometimes they are in, and sometimes out. We say that
these are unstableon the 0-hypothesis. On the first two types of
objects the definition 'yields' a definite verdict relative to the
original hypothesis; but on the unstables it does not.9
Observe next that the ais and b behave in the same way
regardlessof the hypothesiswith which we begin the revisionprocess.
Hence the rule of revision results in categorical judgments on
these objects: the ws are G but b is not G. The object d is unstable
on all hypotheses, reflecting its paradoxical character; c is stable
on all hypotheses, but it is positively stable on some, and
negatively on others. This corresponds to the intuition that it is
like the Truth-Teller.
I hope these brief remarks lend some plausibility to the idea
that we can make semantic sense of circular definitions. lfwe see
the meaning of the definiendum as having a hypothetical
character, we can begin to understand both how a circular
definition fixes this meaning and how the definiendum's
behaviour-ordinary and pathological-has its source in this
meaning.
III
Let us now turn to the second of our two tasks, to give the logical
rules governing definitions. It is clear that these rules should be
modifications of Dfl and DfE; the modifications being necessary
because, as we saw at the outset, Dfl and DfE yield contradictions
when applied to circular definitions. How, then, should these

8
I mean 'better' in a weak sense:the possibilitythat a successorstage is exactly as good
as its predecessor is not excluded.
9 In order to develop a satisfactorytheory of definitionswe would need to extend the
revision process into the transfinite. I cannot discuss here how this problem is to be
solved. For some earlier proposalsfor its solution (within truth-theory)see the papersby
Belnap, Herzberger, and me in the journal cited in note 2. I am not fully satisfiedwith
these proposals, however.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
238 ANIL GUPTA

rules be modified? The semantical scheme sketched above

suggests a natural answer. We notice that if the definiens,
A(x, G),
holds of an object x at a revision stage i then the definiendum,
x is C,
holds at the revision stage i+ 1. Similarly, if the definiendum
holds at i + 1 then the definiens must hold at i. This suggests
that we should keep trackof stagesof revisionwhen applying Dfl
and DfE. This can be done by associating an integer index with
each step in a derivation. Application of a rule for definitions
will result in a conclusion with an index one above or one below
that of the premiss. The modified versions of DfI and DfE are:

A(t, ) l G(t)' DI
DflrDfErA(t, G)

Note that stages are indicated by integer superscripts on

formulas.
The logical rules for the connectives are classical with the
proviso that their application requires that the premissesand
the conclusion have the same indices. Thus to apply modus
ponens,the two premisses A and 'if A then B' must have the
same index i, and the conclusion deduced will also have the
index i. To prove 'ifA then B' with index i by ConditionalProofwe
need to prove X~from the hypothesisA'. The calculus has only
one special rule, IndexShift, that allows arbitraryshifting of the
index of a formula so long as there are no occurrences of the
definiendum in it. The acceptability of this rule, I hope, is plain.
Let C1 be the logical calculus whose rules are
Dflr + DfEr+ Index Shift + Classical Logic.

A sentence A is logicallydeduciblein C1from a definitionD iff a

derivation of A', for some index i, can be constructed in C1 from
D. Note that since the correctness of a derivation is preserved
under uniform shifting of indices, if A' can be deduced for one
index i then it can be deduced for any index i.
I will give one simple example of a derivation in C1. Consider
again definition (1).

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 239

(1) x is G l. x is F or x is both H and non-G.

We can prove from it that all Fs are Gs as follows.

(i) (x is F)0 [Supposition]

(ii) (x is F[ ' [Index Shift, (i)]
(iii) (x is F or x is both H and non-G)['
[Propositional Logic, (ii)]
(iv) (x is G)? [DfIr, (iii)]
(v) (ifx is F then it is G)
[Conditional Proof, (i)-(iv)]
(vi) (All Fs are Gs)'
[Universal Generalization (v)]
Observe that by the revisionscheme given above 'all Fs are Gs' is
categorically assertible under all conditions M; hence the
implication holds from the semantical viewpoint also. On the
other hand, since 'all Hs are Fs' is not so assertible,the definition
does not semantically imply 'all Hs are Fs'. Nor can this sentence
be derived in C1. Its earlierproofon the basisofDfl and DfE does
not go through because of the presenceof indices. (In particular,
if we hypothesize (2) with index 0 and (3) with index 1 then we
can prove the negation of (3) not with index 1 but with index 0.
As a result the derivation of (7) from (2) breaks down.)
Let us note some important propertiesof this calculus.First,it
can be shown to be sound with respectto the semanticssketched
above. A completeness claim can be established also, but the
semantical notions need to be defined with some care. Second,as
a consequence of the soundness of the calculus, circular
definitions are not creative in C1; and as a consequence of
completeness, they are not barreneither. The definitionsare not
creative in the sense that they do not enable us to prove
essentially new things, i.e., things that do not involve the
definienda. They are not barren in the sense that at least some
claims about the definienda can be established. Third, the
indices employed in the calculus are important only in the
context of hypothetical reasoning. Within categorical contexts,
they can be dispensed with; one can work with DfI and DfE
unmodified. This is a consequence of the fact that a uniform
shifting of indices preservesderivability. Fourth,the indices are
dispensable in all contexts if the definition in question is non-

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
240 ANIL GUPTA

circular. This is a consequence of Index Shift, which allows the

following derivation to be constructed:
(i) GC [Supposition]
(ii) A(t, G)'1 [DfEr, (i)]
(iii) A(t, GV ' [Index Shift, (ii)]
(iv) Gi [Dflr, (iii)]
Hence indices on Ct can be shifted arbitrarily. It can be shown
by induction that the same holds of all formulas. So with non-
circular definitions the indices do no work; they 'can be
eliminated.
Traditional doctrine imposes two requirements on definitions,
the requirement of non-creativity and that of eliminability.
According to the first requirement, essentially new things should
not be provable through the use of definitions. According to the
second, the definienda should always be, in principle, eliminable:
for every statement containing the definienda one should be
able to produce an equivalent statement not containing them.
The two requirements, it seems to me, are not of equal status.
The requirement of non-creativity is clearly correct. It would be
strange indeed if one could prove the doctrine of Monism, say,
from a definition of some irrelevant predicate G. This require-
ment is respected by all definitions, circular and non-circular, if
they are used according to the rules sketched above. The
correctness of the second requirement is not so clear. Why
should all definitions satisfy it? Why must the definiendum
always be eliminable? As far as I know, no argument for this
requirement is to be found in the published logical literature. 10
It seems to me that the root motivation for the requirement lies
in two ideas: the traditional conception of meaning and the idea
that a definition should fix the meaning of its definiendum. (For
first-order languages one can show by Beth's Definability
Theorem that these two ideas entail the eliminability of the
definiendum.) If, however, one rejects, as we have, the
traditional conception of meaning then it is possible to abandon
eliminability and yet preserve the idea that a definition fixes the
meaning of its definiendum. This is precisely what happens in

'"Only in one place have I seen an attempt to justify this requirement: Belnap's
unpublished , Voles on Ihe Ar of Logic.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 241

the semantical theory sketched above: definitions fix the

meanings of their definienda but eliminability does not hold.
The theory of definitions put forward here entails a few
changes in our general conception of definitions, but it leaves
our ordinary ways of working with non-circular definitions as
they are. We can use Dfl and DfE unmodified in the context of
such definitions. And we can assign definite extensions to their
definienda." By accepting the logical theory proposed above,
we do not lose anything. We do, however, gain something: we
can make sense of definitions that previously had to be ruled
illegitimate.
The theory in its present formulation is admittedly much too
simple to be completely satisfactory. On two points especially
improvements are needed. First, we would want to attribute
greater content to circulardefinitionsthan that attributed in the
theory given above. Second, we would want to allow systemsof
mutually interdependent definitions, i.e., systems in which a
concept G, for example, is defined using F, H, . . . , and Fin turn
is defined using G, H, .... I believe that these improvements
can be made, though the resulting theory is necessarily more
complex.
IV
I hope that the simple theory sketched above, even though it is
not completely satisfactory,does succeed in showing that logical
and semantical frameworks are possible in which circular
definitions make sense. And once we recognize thispossibility,
we must recognize also that some of our ordinary notions may
best be analyzed in such a framework.It may be that some of our
ordinary notions are circular or, more generally, are parts of a
system of mutually interdependent concepts.
The reader will hardly be surprised to learn that the above
framework can be applied to the notion of truth. Actually,
Herzberger, Belnap, and I had investigated'2 a revision
semantics for truth long before we (or at least 1) understood its

" This is so because the conditions of applicability of a definiendum G in a non-

circular definition do not depend on any prior hypothetical extension for G. Hence, the
rule of revision is a constant function and entitles one to speak of theextension of the
definiendum.
2 See the papers in the journal cited in note 2.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
242 ANIL GUPTA

general significance and rationale. We studied first-order

languages in which the truth-predicate was interpreted via a
rule of revision, and we found that the revision rule had very
attractive properties indeed. It yielded intuitively correct
categorical judgments on ordinary unproblematic sentences;
these always stabilized at the same value no matter what the
initial hypothesis of revision. The intuitively paradoxical
sentences always oscillated in the revision process; they never
stabilized.Truth-Tellersand other sentenceslike them stabilized
on all hypotheses, but sometimes they stabilized as true,
sometimes as false. Other kinds of pathological sentences
exhibited their own distinctive behaviour in the revision
process.
I do not want to give the impressionthat there are absolutely
no problemswith the revision theory of truth. There are worries
about the proper way of extending the revision processinto the
transfinite.And there is room for disagreementover whether the
theory's pronouncements on some particular cases correspond
to the observed behaviour of truth. However, I do believe that
the theory yields an attractive account of truth, an account that
is superior in some ways to that offered by other theories.'3
One attractive feature of the revision theory is that it enables
us to accommodate and make sense of the fundamental
intuition concerning truth, the intuition that is enshrined in
Tarski's Convention T. Many philosophersand logicians, from
Aristotle down to Tarski, have noted that the equivalences
captured by the rules TI and TE, i.e., the equivalences of the
form,
'A' is true if and only if A,
are constitutive of the notion of truth; that they define the
conditions under which a sentence is true. Observe, however,
that these equivalences yield truth-conditionsthat are circular.
For example, the equivalence

"$I have argued in 'Truth and Paradox' (Journalof PhilosophicalLogic 11, 1982,
pp. 1-60) that the revision approach is better in some respects than the Fixed-Point
approach of Saul Kripke, Robert. L. Martin, and Peter Woodruff.Their papers can be
found in Martin (ed.), RecentEssaysonTruthandtheLiarParadox,Oxford University Press,
1984. Herzberger's and my papers are also reprinted in this collection.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 243

'Everything Daniel says is true' is true if and only if

everything Daniel says is true,
defines the truth-conditionsfor the sentence 'everything Daniel
says is true' in terms of the notion of truth itself. The logical
problem of truth can be viewed as arising from a tension
between the fundamental intuition (which implies a circularity
in truth) and standard logic and semantics (which rule out all
circularity).The strategythat is generallyadoptedin constructing
theories of truth is to work within standardlogic and semantics.
The inevitable consequence has been that the fundamental
intuition has been denied in one way or another. The approach
via circular definitions offersa differentstrategy.It enables us to
take the fundamental intuition at face value. We can see the
equivalences as giving circular conditions for truth and we can
use them to define a rule of revision for it. When we do, we find
that the rule of revision they yield is preciselythe one whose
virtues (at least some of them) I enumerated two paragraphs
ago.
When I first thought about the problem of truth my attitude
was that the fundamental intuition, though largely correct, was
not completely valid and, like other theorists, I took this to be
the basic lesson of the paradoxes. (This attitude is present in my
'Truth and Paradox'.) Now I believe I see things more clearly:
The fundamental intuition is completely correct. The equival-
ences areconstitutiveof the notion of truth. But what this showsis
that truth is a circularconcept and we need a logical and seman-
tical frameworkto make sense of this circularity. The theory of
truth I advocate is obtained by combining the general theory of
definitions with the fundamental intuition. This theory explains
the behaviour of truth, both ordinary and pathological, as
arising from the fundamental intuition. In particular,it is because
truth is a circular concept that it exhibits the paradoxical
behaviour that philosophers and logicians have found so
perplexing. As I see it, the fundamental intuition and the
existence of the paradoxes are not only nota problem for the
revision theory of truth, they are confirmingevidence for the
theory.
One final observation concerning truth: the fact that
paradoxical sentences involving truth are invariably self-

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
244 ANIL GUPTA

referential'4is to be explained by the fact that it has a particular

kind of definition, namely, one that is based on the equivalences
mentioned above. It is not an essential feature of paradoxical
sentences, however, that they involve self-reference.Consider,
for example, the following definition:
(13) x is G=Dl X is not G.
(This definition is admittedly strangeand definesa notion that is
probably completely useless. Still, the general theory of
definitions recognizes the notion as legitimate.) Observe that by
this definition the sentence,
(14) Socrates is G,
is paradoxical, even though it is not self-referential:(14) is not
about itself even indirectly; it is about Socrates.
I hope the peculiarity of definition (13) does not obscure the
main point here, which is that 'self-reference', though it is
essential to paradoxes involving truth, is not essential for
paradoxes in general. I make this observationbecause it suggests
that the frameworkof circular definitions may be applicable to
problems that appear to be disparate from those under
discussion. It is clear that the frameworkwill apply to notions
that result in Liar-like pathologicality. Under this category fall
semantic notions such as 'reference' and 'satisfaction', and
intensional notions such as 'property', 'necessity', and 'knowl-
edge' (cf. the Knower's Paradox of Kaplan and Montague).'5
But I think that the approach may be more widely applicable.
Other philosophical puzzles may be amenable to the same kind
of treatment as that given for the Liar above-many more of our
concepts may belong to systemsof interdependentconcepts than
may appear at first sight. It is not going to be easy to provethis
claim, for outside of logic and semantics definitional links
between concepts are hard to establish. However, I do find
encouragement in the fact that for many concepts there are
'4 I mean 'self-reference'in a general sense:a sentence .4 is self-referentialif it is about

itselfor is about a sentence B that is about 4 or . . . . I shall not try to make the notion of
'aboutness' precise.
'5See N. Asher and H. Kamp, 'The Knower's Paradox and Representational
Theories of Attitudes' inJ. Halpern (ed.), Theoretical.4spects of Reasoning about Knowlkledge
(Los Altos: Morgan Kaufman, 1986), pp. 131-147; and R. Turner, 'A Theory of
Properties', J7ournalof Symbolic Logic 52, 1987, pp. 445-472.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
 REMARKS ON DEFINITIONS AND THE CONCEPT OF TRUTH 245

persistent puzzles, and that philosophers in their analyses often

go around in circles.
I would like to conclude this paper with some general
observations. We have been taught to judge philosophical
analyses and definitions in terms of two criteria: these are, in
Tarski's terminology, the criterion offormal correctness and that of
materialadequacy.The first requires us tojudge the analysis from a
formal, logical point of view; it is the one cited in the rejection of
circularity. The second requires us to judge the faithfulness of
the analysis to the ordinary notion being analyzed. We have
seen above that sometimes the two criteria can come into
conflict. The analysis that is materially adequate sometimes
violates the demand of formal correctness.'6 This conflict, I am
suggesting, should be resolved in favour of the second requirement
(material adequacy). I would like to go further in this direction
and propose that in philosophical theorizing we should abandon
entirely the requirement of formal correctness: a definition
should be evaluated only by how well it captures the material
aspects of a notion. Logic, in my view, imposes no a priori
requirements on definitions at all.
Under this liberalization trivial definitions such as
(15) x is a person =- x is a person
are admitted as logically proper, but they are usually materially
inadequate. Thus, (15) fails to be an adequate definition of
'person' since it implies that 'x is a person' is like the Truth-
Teller for all objects x, but an adequate definition should yield,
for example, that Socrates is a person and that Chicago is not.
The burden that has traditionally been borne by the requirement
of formal correctness is better carried, in my view, by the
requirement of material adequacy.
The philosophical moral most often drawn from the paradoxes
is that the domain of the meaningful is less extensive than it
appears to be, that certain seemingly meaningful concepts are in
fact meaningless. Russell's Vicious Circle Principle is the classic

'6JerroldLevinson has written in his review of George Dickie's book TheArtCircle,'A

circular definition, however segmented, no more clarifies anything than it informsor
instructs-one simply cannot elucidatethe content of a concept by usingand presupposingit
in the course of the elucidation [Philosophical 96, 1987, p. 145]'. I am in complete
Revziewu
disagreement with this claim.

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions
246 ANIL GUPTA

attempt to demarcate the meaningful from the meaningless. In

contrast, the moral I believe we should draw fromthe paradoxes
is that the domain of the meaningful is moreextensive than it
appears to be, that certain seemingly meaningless concepts are
in fact meaningful. There are more concepts in Platonic Heaven
than are dreamt of in classical ontologies.

Departmentof Philosophy
Universityof Illinois
P.O. Box 4348
ChicagoIL 60680

This content downloaded from 146.50.96.170 on Wed, 06 May 2015 15:50:41 UTC
All use subject to JSTOR Terms and Conditions