282 panu raatikainen

11
Nietzsche, F. 1885–89/2003. Writings from the Last Notebooks, ed. R. Bittner. Trans.
K. Sturge. Cambridge: Cambridge University Press.
Ramsey, F. 1925/1997. Universals. Repr. in Properties, eds. D. H. Mellor and A. Oliver.
Oxford: Oxford University Press.
Russell, B. 1919. Introduction to Mathematical Philosophy. London: George Allen and
Unwin.
Strawson, G. 1987. Realism and causation. Philosophical Quarterly 37: 253–77.
Strawson, G. 2003/2008. What is the relation between an experience, the subject of the
experience, and the content of the experience? Revised version in G. Strawson, Real
Materialism and Other Essays. Oxford: Clarendon Press.
Strawson, G. 2006. Panpsychism? In Consciousness and its Place in Nature, ed. A.
Freeman. Thorverton: Imprint Academic.
Weinberg, S. 1997. Before the Big Bang. The New York Review of Books 44/10.
Whitehead, A. N. 1927–8/1979. Process and Reality. New York: The Free Press.

On rules of inference and the meanings of

logical constants
Panu Raatikainen

1. Introduction
In the theory of meaning, it is common to contrast truth-conditional
theories of meaning with theories which identify the meaning of an expres-
sion with its use. One rather exact version of the somewhat vague use-
theoretic picture is the view that the standard rules of inference determine
the meanings of logical constants. Often this idea also functions as a
paradigm for more general use-theoretic approaches to meaning. In par-
ticular, the idea plays a key role in the anti-realist program of Dummett
and his followers.
In the theory of truth, a key distinction now is made between substantial
theories and minimalist or deflationist views. According to the former,
truth is a genuine substantial property of the truth-bearers, whereas
according to the latter, truth does not have any deeper essence, but all that
can be said about truth is contained in T-sentences (sentences having the
form: ‘P’ is true if and only if P).
There is no necessary analytic connection between the above theories of
meaning and truth, but they have nevertheless some connections. Realists
often favour some kind of truth-conditional theory of meaning and a
substantial theory of truth (in particular, the correspondence theory).
Minimalists and deflationists on truth characteristically advocate the use
Analysis 68.4, October 2008, pp. 282–87. © Panu Raatikainen
on rules of inference and the meanings of logical constants 283
02
theory of meaning (e.g. Horwich). Semantical anti-realism (e.g. Dummett,
Prawitz) forms an interesting middle case: its starting point is the use theory
of meaning, but it usually accepts a substantial view on truth, namely that
truth is to be equated with verifiability or warranted assertability. When
truth is so understood, it is also possible to accept the idea that meaning is
closely related to truth-conditions, and hence the conflict between use
theories and truth-conditional theories in a sense disappears in this view.

2. Carnap’s challenge
The view that the standard rules of inference (or, sometimes, just the
introduction rules) determine the meanings of logical constants seems to
be rather widely held:
(RIDMLC) The Rules of Inference Determine the Meanings of
Logical Constants.
The idea is certainly natural and may indeed appear in a sense quite
uncontroversial, given the completeness of all standard systems of rules of
inference for both propositional and (first-order) predicate logic.
But the popularity of such views notwithstanding, certain facts under-
mine them. Namely, although this is apparently very little known, Carnap
(1943) put forward considerations to the conclusion that in a definite
sense, it is not true that the standard rules of inference manage solely to
determine the meanings of logical constants. It can be shown that no
ordinary formalization of logic, and not the standard rules of inference (of
the natural deduction) in particular, is sufficient to ‘fully formalize’ all the
essential logical properties of logical constants. That is, they do not
exclude the possibility of interpreting logical constants in any other than
the ordinary way. As Carnap’s arguments seem to have been almost
entirely ignored,1 it is perhaps proper to remind the philosophical com-
munity of them.2

1
There are, however, some important exceptions: Carnap’s results are taken into
account at least in Smiley and Shoesmith 1978, Belnap and Massey 1990, Smiley
1996, Rumfitt 1997, Rumfitt 2000, and Humberstone 2000. These works certainly
deserve to be read in this connection, though the aims and emphasis of them are quite
different from mine here. Unfortunately, they have not changed the overall situation
to a large extent; it still seems that Carnap’s challenge is not sufficiently known – in
particular among those who should be most concerned with it, i.e. the advocates of
RIDMLC. More generally, the large literature on the meanings of logical constants is
basically quiet about it.
2
I’ll consider only the case of propositional logic. Carnap also gives examples of
non-normal interpretations for quantification theory. But propositional logic suffices
for making the point.
284 panu raatikainen
02
Let us consider the following principles that seem to be essential for
negation and disjunction. They follow, in classical logic, from the seman-
tical definitions of connectives (e.g. in terms of truth tables), but they also
hold in intuitionistic logic, where truth is identified with provability
(Carnap only considered explicitly classical logic.) A and B are arbitrary
sentences:
(N1) A is true fi ¬ A is false.
(N2) A is false fi ¬ A is true.
(D1) A is true and B is true fi (A ⁄ B) is true.
(D2) A is true and B is false fi (A ⁄ B) is true.
(D3) A is false and B is true fi (A ⁄ B) is true.
(D4) A is false and B is false fi (A ⁄ B) is false.
Yet the standard formalizations of logic (rules of inference) do not rule out
non-normal interpretations which violate these principles. There are (in
the context of propositional logic) two different kinds of non-normal
interpretations of connectives allowed by the ordinary rules of inference:
there may either be a violation of (N1) or a simultaneous violation of (N2)
and (D4). (As it happens, the rules of inference guarantee that (D1)–(D3)
cannot be violated.) In the first kind of non-normal interpretation, where
(N1) is violated, for any sentence A, both A and ¬ A are true, and
moreover, every sentence is true. In the second kind of non-normal inter-
pretation, where both (N2) and (D4) are violated, both A and ¬ A are
false, but (A ⁄ ¬ A) is true; in that case, there are infinitely many true and
infinitely many false sentences (see Carnap 1943: Chapter C).
In sum, although the standard rules of inference completely formalize
logical truth and logical consequence, they do not fully represent all logical
properties of the connectives. Thus, for instance, it arguably belongs to the
logical properties of disjunction that a sentence of disjunction with two
false components is false. Nevertheless, this property is in no way repre-
sented in ordinary formalizations of logic. Similarly, it is certainly a part of
the intended meaning of negation that a sentence and its negation cannot
both be true, and further that they cannot both be false either.3 But, again,
these properties are not in any way represented in the usual rules of
inference.
The forgotten facts above present a serious challenge for RIDMLC, the
view that rules of inference determine the meaning of logical constants – a
problem one to which should at least give serious consideration.

3 Carnap (1943: 100) calls the former, quite naturally, the principle of (excluded)
contradiction and the latter, less happily, the principle of excluded middle.
on rules of inference and the meanings of logical constants 285
02
3. Rules and truth
What then is the moral of all the above? This depends on various other
general philosophical background assumptions. There are several different
alternatives here:
(a) A radical formalist may just deny the very meaningfulness, or at
least relevance, of the notions of truth and falsehood (and consequently, of
(N1)–(D4)) and insist that his use-theoretical approach is a genuine alter-
native to the truth-conditional approach and that it would beg the ques-
tion to appeal to (N1)–(D4) against it. With respect to this sort of radical
view, the above considerations have no force. But such a view is in itself
very controversial and problematic. I do not believe that any contempo-
rary adherent of RIDMLC accepts such a radical formalism, certainly not
intuitionists such as Dummett, Prawitz and their followers. Consequently,
I shall not consider this extreme view any further.
(b) One might introduce the minimalist or deflationist notion of truth
and falsehood, and try to argue that they suffice for avoiding Carnap’s
problem and to save RIDMLC – more on this below.
(c) One may alternatively assume that there is some sort of match
between the proof-theoretical meaning-giving rules of inference and
semantical notions of truth and falsity (possibly understood, as by intu-
itionists, in terms of provability), and that the former simultaneously
determines a suitable notion of truth (possibly a non-realist, epistemic or
verificationist one, which equates truth with provability). In particular,
something like this seems to be the view of Dummett, Prawitz and their
followers. For this kind of view, Carnap’s problem seems to pose a real
challenge.

4. Relation to bivalence and LEM

A central aim of the anti-realist program of Dummett, Prawitz and others
is to rebut the Law of the Excluded Middle (LEM) by undermining the
Principle of Bivalence. However, their fundamental equivalence,
Bivalence holds ¤ LEM is true,
only holds under two default assumptions:
(i) A is true or B is true fi (A ⁄ B) is true (which is stronger than our
D1 above, and entails it).
(ii) A is false fi ¬A is true. (i.e. our (N2) above)
But it is unclear at least how an adherent of RIDMLC (such as Dummett,
Prawitz and co.) is allowed to appeal to these principles. The rules of
inference do not suffice to entail them, as Carnap’s results show, so from
where does one get them, without begging the question?
286 panu raatikainen
02
5. Minimalism and deflationism
If the use theory of meaning (and RIDMLC in particular) is understood
as a competitor of the truth-conditional theory of meaning, it would
be somewhat problematic if an adherent of the former would appeal to
a substantial theory of truth in order to solve Carnap’s problem.
However, one may think that at least an appeal to the minimalist or
deflationist view on truth is unproblematic. According to deflationism,
the meaning of truth and falsity is exhausted by the following self-
evident equivalences:
(T1) A is true ¤ A,
(T2) A is false ¤ ¬A.
Alternatively, one might take falsity as a defined notion, in which case (T1)
would suffice, e.g. as follows:
A is false ¤def (¬A) is true.
With the aid of not much logic, these schemes seem to give us what we
want (i.e. principles (N1)–(D4)):
(N1) A is true fi A fi ¬¬A fi ¬A is false;
(N2) A is false fi ¬A fi ¬A is true;
(D1)–(D3) A is true or B is true fi A ⁄ B fi (A ⁄ B) is true;
(D4) A is false and B is false fi ¬A ∧ ¬B fi ¬(A ⁄ B) fi
(A ⁄ B) is false.
However, a closer look shows that this does not really help, but that
the problem has just moved to the next level. Namely, nothing here
rules out e.g. the possibility that a sentence A, or every sentence, is both
true and false. In other words, the problem now occurs on the level of
truth as the problem is that true and false are certainly assumed to be
mutually exclusive properties, but nothing here guarantees that. Every-
thing assumed so far allows that true and false overlap or are even
coextensive.
At this point, someone might suggest that we just add to our theory the
exclusion assumptions:
(E1) A is true ¤ A is not false
(E2) A is false ¤ A is not true.
At first sight, this would appear to solve the problem. But in fact, this does
not help either. More formally, these read simply as:
A is true ¤ ¬(A is false)
A is false ¤ ¬(A is true).
on rules of inference and the meanings of logical constants 287
02
And because we just have not been able to fix the meaning of the negation
in the intended way, any such principles, which essentially use negation,
cannot help either. This becomes especially evident when, using
T-equivalences, one replaces ‘A is true’ by A, and ‘A is false’ by ¬A; we
then get:
A ¤ ¬¬A
¬A ¤ ¬A.
These are, however, already provable tautologies of (classical) proposi-
tional logic (i.e. derivable with the help of the standard rules of inference)
and cannot possibly help in solving Carnap’s problem.

6. Conclusion
Carnap’s forgotten result poses at least prima facie a deep problem for
RIDMLC. Now I do not wish to go as far as to claim that RIDMLC
stands conclusively refuted. But certainly the above-considered facts
deserve attention, and those who build their philosophical theories on
RIDMLC owe us at least an explanation of how such problems could be
circumvented.4

P.O. Box 9, FIN-00014 University of Helsinki, Finland

[email protected]

References
Belnap, N. and G. Massey. 1990. Semantic holism. Studia Logica 48: 67–82.
Carnap, R. 1943. Formalization of Logic. Cambridge, MA: Harvard University Press.
Humberstone, L. 2000. The revival of rejective negation. Journal of Philosophical Logic
29: 331–81.
Rumfitt, I. 1997. The categoricity problem and truth-value gaps. Analysis 57: 223–35.
Rumfitt, I. 2000. Yes and no. Mind 109: 781–823.
Smiley, T. 1996. Rejection. Analysis 56: 1–9.
Shoesmith, D. J. and T. J. Smiley 1978. Multiple-conclusion Logic. Cambridge: Cam-
bridge University Press.

4 I am very grateful to Peter Milne for helpful discussions and correspondence on the
contents of this paper. In particular, §5 on the possible deflationist response owes
much to him, and the derivations of (D1)–(N4) from (T1)–(T2) there are solely due
to him.