Professional Documents
Culture Documents
Meaning and Circular Definitions (F. Orilia)
Meaning and Circular Definitions (F. Orilia)
Meaning and Circular Definitions (F. Orilia)
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 support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Springer is collaborating with JSTOR to digitize, preserve and extend access to Journal of
Philosophical Logic
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
FRANCESCO ORILIA
ABSTRACT. Gupta's and Belnap's Revision Theory of Truth defends the legitimacy of
circular definitions. Circularity, however, forces us to reconsider our conception of mean-
ing. A readjustment of some standard theses about meaning is here proposed, by relying
on a novel version of the sense-reference distinction.
1. INTRODUCTION
(1) a circle is a locus of point in the same plane equidistant from some
common point,
(2) knowledge is justified true belief.
The nature of the link between the analysandum and the analysans is a
matter of dispute. Castafieda (1980, p. 59) distinguishes a View (A) and a
View (B) of analysis, as follows:
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
156 FRANCESCO ORILIA
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 157
For the convenience of the reader, let us briefly review how RTD works,
and in particular how it can be used to introduce circular predicates (for
further details, cf. Gupta and Belnap, 1993).
Assume that (i) L is a first-order language not involving the members
of the set 9 of primitive predicate constants, and (ii) L+ is just like L
except that it also involves the members of s (L is called a base language
with respect to the extended language L+). For simplicity's sake, the only
singular terms of these two languages are individual variables and primi-
tive constants. We shall take any wff with n free variables to be an n-adic
predicate. Two wffs that are notational variants of each other (i.e., roughly,
that differ only for the choice of their variables) will count as the same
predicate. 'Fn', 'Gn', 'Hn ', pn, 'Qn' will range over members of 9. 'A',
't', 'x', possibly primed or with subscripts, will range, respectively, over
wffs, individual constants, individual variables. For any wff A, any use of
'A(xl, ..., xn)' will indicate that x, . .., xn are all the variables occurring
free in A; moreover, 'A(xl / tl, ... , xn, / tn)' will denote the wff resulting
from A by replacing each free occurrence of xi with ti, for 1 < i < n.
Consider now a set +D containing, for each member Gn of 9, a unique
definition of the form
where AG(x1, ... , Xn) is a formula of L+. RTD provides a strategy that
permits us to plausibly consider each member of D - in spite of its circu-
larity - as a definition based on the expressive resources of L. This works
as follows. We assume standard first-order models M, M1, M2, ... of L,
which, obviously, do not tell us anything about the extension of the predi-
cates of L+ that are not in L. (Let us understand that + (+1, +2, .. .) is the
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
158 FRANCESCO ORILIA
i+ (4+, 4-, ...) is the interpreted language resulting from the 'syntactic
part' L+ and the 'semantic part' M (M1, M2, .. .), as well as the definitions
in i; cf. Gupta and Belnap (1993, p. 145).) We then 'extend' each model
M by means of hypotheses regarding the extension of the definientia, so as
to get models M + h (M + hi, M + h2,...) of L+ where h (hi, h2, ... )
is any such hypothesis. M (+) is a ground model (language) with respect
the extended model (language) M + h (4h) (cf. Gupta and Belnap, 1993,
p. 146). The extended models are arbitrary in that no appropriate relation
between a definiendum in D and its corresponding definiens can be taken
for granted from their point of view. Nevertheless, each model M + h
informs us about what the extension of a definiens is, once the corre-
sponding definiendum has been assigned an arbitrary extension. At this
point, RTD provides us with instructions, i.e., a rule of revision, concern-
ing how to construct, for any ground model M, a sequence of extended
models M + h in an effort to overcome the arbitrariness of each individual
member of the sequence. This gives rise to an evaluation sequence G of
++ (Gupta and Belnap, 1993, p. 188). Each member of this sequence
specifies a truth-value for each pair (A(x1, ..., xn), (dl, ..., dn)), where
(d1, ...., d) is an n-tuple of M. (Let us call such pairs predicate/individual
pairs; if the predicate in question is a definiens or definiendum we shall
also say definiens/individual pair or definiendum/individual pair, as the
case may be.) In this sequence as a whole some predicate/individual pairs,
but not all, acquire from some point onward a stable evaluation, i.e., they
are stably x in 6, where 'x' ranges over the set of truth-values {t, f,...},
and 't' and 'f' represent truth and falsehood (see Gupta and Belnap, 1993,
p. 167, for a precise definition). The fact that some evaluations stabilize
is taken to indicate that the arbitrariness of the extended models has been
eliminated.
Let us now be more specific about revision rules. Essentially, a rule
of revision consists of (i) a definition, (ii) the successor rule and (iii) a
limit rule. The successor rule - to be used at successor ordinals - tells
us (roughly) that the evaluation of a definiendum/individual pair, accord-
ing to the (n + 1)th member of an evaluation sequence, must be equal
to the evaluation of the corresponding definiens/individual pair, accord-
ing to the nth member of the sequence in question. The limit rule is the
policy to be followed in assigning an evaluation at limit levels to those
definiendum/individual pairs that have not yet acquired a stable evaluation.
Here several choices are possible (cf. Gupta and Belnap, 1993, p. 168).
For present purposes, we can assume that the only limit rule we have to
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 159
deal with is the simplest one, namely, Belnap's (1982) vacuous limit rule,
according to which the evaluations in question are assigned arbitrarily.
From the point of view of RTD, the notion of logical truth can be char-
acterized as follows: a sentence (0-adic predicate) A of L+ is logically
true (in the sense of RTD) iff for every M, (A, ()) is stably t in every
evaluation sequence for M. The following is a noteworthy feature of RTD:
all the sentences of L+ that have the same logical form as a logical truth of
L, as standardly defined in terms of truth in every model, are also logical
truths of L+, in the sense of RTD. In other words,
(GROUND-LOGIC) RTD preserves the logic of the ground languages.
Therefore, in particular, if the ground languages + are two-valued classical
languages, all sentences of L+ that have the form of theorems of first-order
classical logic are logical truths in the sense of RTD. Let us make, for
simplicity's sake, the following Classical Logic Assumption:
(CLA) The ground languages we deal with are classical.
Thus, we assume that the set of truth-values is {t, f}.
Given a revision rule and a model of L, each n-adic predicate of L+
can be naturally classified, in relation to an n-tuple of M, in terms of
pathological vs. categorical (nonpathological). The former unfolds into
paradoxical, truth-teller-like, and mixed; while the latter unfolds into true
and false. Moreover, there is a classification in terms of stable, which sub-
sumes the categorical and the truth-teller-like. Let us call each of these
notions an RTD-category. (In certain formal contexts, I shall use capital
letters to refer to an RTD-category, so as to single it out more explicitly
('CATEGORICAL', 'STABLE', etc.).) These concepts are to be under-
stood as follows (where A is an n-adic predicate of L+ and (dl, ...., d,) is
any n-tuple of a model M of L):4
A is stable for (dl, ...., d,) in M iff for every evaluation sequence & of
++, there is a truth value x such that (A, (dl,..., d,)) is stably x in 8.
A is nonpathological (categorical) for (dl, ..., d,) in M iff there is a truth
value x such that (A, (dl, ...., d,)) is stably x in all evaluation sequences
of 0++
A is true for (dl ...., d,) in M iff (A, (dl ...., d,)) is stably t in all evalu-
ation sequences of ++.
A isfalse for (dl, ..., d,) in M iff (A, (dl,..., d,)) is stably f in all eval-
uation sequences of ++.
A is pathological for (dl, ..., d,) in M iff (A, (dl, .... d,)) is truth-teller-
like, paradoxical, or mixed for (dl, ..., d,) in M.
A is truth-teller-like for (dl, ..., d,) in M iff (A, (dl, ..., d,)) is stable in
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
160 FRANCESCO ORILIA
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 161
may not be logically true, in spite of (D). Clearly, this cannot be the case,
if (D) involves certain forms of self-contradictory circularity, as in the
definition
Let us say that two sentences are co-assertable iff they have the same truth
value. The following Co-assertability Principle is in practice universally
accepted and indeed seems hardly objectionable:
(C-A) Sentences with the same meaning are co-assertable.
The following Functionality Principle underlies the compositional view
of meaning that is commonly accepted at least since Frege:
(FUNC) The meaning of a complex expression is a function of the mean-
ings of its parts, so that the meaning of the complex is preserved
when one of its constituents is replaced by a synonym.
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
162 FRANCESCO ORILIA
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 163
5. A POSSIBLE APPROACH
(S-R1) two expressions with the same sense must have the same exten-
sion;
(S-R2) two expressions with the same extension may have different sen-
ses.
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
164 FRANCESCO ORILIA
Clearly, G(x) and F(x) have exactly the same rule of revision and will be
equi-RTD-classified. On the other hand, H(x) and Q(x) have two different
revision rules (because they have two different definientia), but the same
RTD-classification - characterizing their pathological status - in every M.
In sum, revision rule and RTD-classification can be seen, respectively,
as sense and extension of predicates. If we wish to characterize the sense
of an expression as something that can somehow be 'in the mind', we can
alternatively say that some psychological counterpart of a rule of revi-
sion is the sense of a circularly defined predicate. We shall neglect this
complication in the following, since nothing hinges on it.
Clearly, the notion of sense is more general than the notion of rule of
revision. For instance, an expression such as 'red' can be assumed to have
a sense that is not a rule of revision. On the other hand, the notion of RTD-
classification is more general than the traditional notion of extension and
somehow encompasses it. For instance, traditionally, the extension of 'red'
can be viewed as EXTM* ('red'), i.e., the smallest set containing (i) every
((d), t) such that d is (represents) a red object and (ii) every ((d'), f) such
that d' is not (does not represent) a red object. Similarly, for any such d
and d', CM,* ('red') will contain ((d), TRUE) and ((d'), FALSE), but it
will also contain, e.g., ((d), STABLE). More interestingly, for example,
the notion of RTD-classification allows us to characterize the pathological
character of G(x) as defined by (CONTR). In fact, for any d, CM* (G(x))
will contain ((d), PATHOLOGICAL) and ((d), PARADOXICAL).
Once circularity is admitted, we are thus led to reformulate (S-R), (S-
R1) and (S-R2), as follows (for predicates):
(S-C2) Two predicates with the same RTD-classification may have differ-
ent senses.
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 165
Clearly, G(t) v -H(t) is paradoxical, yet the logical truth G(t) v -,G(t)
can be obtained from it by uniformly substituting H(x) with G(x).
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
166 FRANCESCO ORILIA
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 167
should be taken to express the same proposition. Yet (3) appears to express
a trivial tautology, whereas (1) appears to express a non-trivial informative
proposition. Furthermore, it makes sense to suppose that if Mary is not
very knowledgeable about geometry, both (M1) and (M2) below might be
true.
Yet, by appealing to (SP), it should follow from (1) and (M1) that
(M3) Mary believes that a circle is a locus of point in the same plane
equidistant from some common point.
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
168 FRANCESCO ORILIA
7. CONCLUSION
NOTES
1 This paper stems from conversations with Anil Gupta in September 1995, based on
an early version of Orilia and Varzi (forthcoming) and a two-page outline by Gupta on
meaning and circularity. This outline anticipated some of the ideas presented here (in
particular, the proposal to revise what I here call (FUNC)). Gupta was very generous in
permitting me to develop his material further on my own and in declining my offer of
joint autorship. He also helped me to improve a previous version of this work with several
important comments. Finally, I also wish to thank an anonymous referee for some useful
remarks.
2 Systems of assertions such as those giving rise to an inductive definition may be taken
to constitute a more complicated form of analysis.
3 For an implementation of the view that predication is circular, see Orilia (1995) and
(forthcoming).
4 The definitions below extend those of Gupta and Belnap (1993, p. 189) in order to
encompass n-adic predicates in general, and not just sentences.
5 (PE) obviously neglects the special problems engendered by intensional contexts,
which are not part of our concern at this juncture.
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms
MEANING AND CIRCULAR DEFINITIONS 169
6 For example, Bealer (1982, Ch. 3), also takes for granted that non-circularity is a
constraint on analysis.
REFERENCES
6. Gupta, Anil and Belnap, Nuel: The Revision Theory of Truth, The MIT Press,
Cambridge, Mass., 1993.
7. Orilia, Francesco: 'Knowledge Representation, Exemplification and the Gupta-
Belnap Theory of Circular Definitions', in Marco Gori and Giovanni Soda (eds.),
Topics in Artificial Intelligence, 4th Congress of the Italian Association for Artificial
Intelligence AI*IA 95, Springer-Verlag, Berlin, pp. 187-198, 1995.
8. Orilia, F.: 'Property Theory and the Revision Theory of Definitions', Journal of
Symbolic Logic (forthcoming).
9. Orilia, Francesco, and Varzi, Achille C.: 'A Note on Analysis and Circular Defini-
tions', Grazer Philosophische Studien (forthcoming).
This content downloaded from 157.253.50.50 on Sat, 01 Sep 2018 16:17:25 UTC
All use subject to https://about.jstor.org/terms