Beruflich Dokumente
Kultur Dokumente
BALZER 1
T, and let IL(T)I be {o'1. . . . . ok, P,1 . . . . , R,,}. Further, let us assume
that the logic ~ employed by T (as given by the formation rules for
formulas, the logical axioms and the rules of derivation) is the same as
the one employed by T'.
With respect to reduction we will confine ourselves to the "bare
bones" used by Pearce. Again - although this concept of reduction
comprises only part of the structuralist concept - the restriction affects
my negative conclusions only inasfar as it will sharpen them. The
definitions of Sec. III can be easily emended to include the full
structuralist apparatus.
D1. If T = (Mp, M) and T' = (M~, M') are theories then p reduces T' to
Tiff
(1) p: Mp ~ M~, is a partial function, and onto
(2) for all x, y: if x ~ M and y = p(x) then y c M'.
I want to argue that the objection just stated essentially does not apply
in the domain of the empirical philosophy of science. By the empirical
philosophy of science I mean philosophy of science understood as the
empirical investigation of existing empirical theories.
Let me denote by I the domain of existing empirical theories, i.e., the
domain of intended applications of the empirical philosophy of science.
On the other hand, let ~ denote the class of all triples (T, T', p) for
which (1.1)-(1.4) above are satisfied. ~ may be called the domain of
application of the above argument. If v denotes the class of all sets then
I want to argue that ~t fq (I x I v) is "essentially" empty. In other
words: those triples (T, T', p) to which the argument can be applied
(which satisfy (1.1)-(1.4)) do not represent existing pairs of empirical
theories supplied with a reduction relation. I have added the
qualification "essentially" because there may be (and most likely there
are) some empirical theories yielding triples in ~ . But these are not
essential both in number and in importance, and I will neglect them in
the following. Also, I will not say much about (1.2). As it turns out (1.2)
typically will not be satisfied in the case of incommensurable theories.
But Feferman's results can be extended to cover non-disjoint vocabu-
laries too, under further specific assumptions. So my only remark in
connection with (1.2) is that "renaming" - which is freely performed in
formal logic - may not be used in an unrestricted way in the philosophy
of science.
INCOMMENSURABILITY 259
Intuitively, the reason why the argument does not apply to empirical
theories is that empirical theories make use of (and in this sense contain)
very strong mathematical theories which cannot be formalized in
logical systems in which the interpolation theorem (1.4) holds. There-
fore, the empirical theories themselves cannot be formalized by such
systems.
Actually, the situation is more delicate and requires more detailed
consideration. The mathematics used in empirical theories usually
contains classical analysis, or at least arithmetic or Euclidean geometry.
These mathematical theories can only be adequately formalized in
logical systems of at least second order because they involve axioms of
completeness (resp. induction) which cannot be fully expressed in
predicative systems. Therefore, existing empirical theories can only be
formalized in logical systems of higher order. But in such systems the
interpolation theorem does not hold any longer. Thus (1.4) fails and the
empirical theory under consideration cannot occur in a triple (T, T', p)
of premiss (P2).
To this conclusion there are three objections from the point of view
of formal logic. First, it may be pointed out that the claim that the
mathematical theories mentioned above can only be formalized in
higher-order systems is plainly false. We all know that Zermelo-
Fraenkel set-theory (ZF) can be formalized in first-order logic, and in
ZF we can define all the mathematics needed, including classical
analysis, geometry, and arithmetic.
Let us assume we make use of this possibility in the empirical
philosophy of science. The result will be a series of reconstructed
empirical theories such that the models of these theories are models of
ZF in which further special sets (like the set S of models of classical
particle mechanics relativized to a fixed s e t of particles) are dis-
tinguished.
I have two problems with such a program. (A) I would hesitate to call
a reconstruction of some empirical theory along these lines an
adequate reconstruction. The criterion of adequacy I have in mind here
is that in an adequate reconstruction only those axioms should be made
explicit which are treated explicitly in the corresponding informal
theory itself. Of course, this cannot be understood very strictly: often
one aim of reconstruction is to point to logical deficiencies, "gaps" in
the standard formulation of a theory, and to show how such gaps may be
260 W. BALZER
closed. But what I want to exclude by the criterion is that a whole new
theory is added in the course of reconstruction of a given theory. This
would be the case if, say, a reconstruction of particle mechanics
incorparates a system of ZF. Formal set theory does not play any role in
the introduction of physical theories, and even in the introduction of
those mathematical theories which are used in physics. I do not deny
that instances of the axioms of ZF may be used in informal, "naive" set
theoretic reasoning in connection with physical theories, and I do not
want to deny that in this sense set theory is "presupposed". But the
same holds for large parts of ordinary language, too, and nobody would
get the idea of incorporating a formalization of ordinary language in a
reconstruction of mechanics (even if this was possible).
Formal set theory basically has foundational character. I agree with
Putnam 6 that mathematics usually proceeds without foundations and in
a sense does not need foundations, and this extends a forteriori to
empirical theories. Therefore the incorporation of ZF into the recon-
struction of an empirical theory adds some new body of theory, and this
definitely leads to a theory very different from the one to be recon-
structed. In this sense such a reconstruction is inadequate.
What has just been said remains valid if ZF is replaced by some
weaker system of set theory which is still sufficient to express the
mathematics needed.
(B) Reconstructions incorporating ZF in a certain sense will beg the
question of reduction and incommensurability. For if all theories consist
of ZF plus further stipulations then there is always the possibility of a
trivial reduction (identity), and it is difficult to see in what sense such
theories would be incommensurable. The meaning of all the "interes-
ting", empirical concepts would be completely determined by the
meaning of the basic set theoretic notions because the former are
explicitly defined in terms of the latter. But the latter are the same in
both theories and so no meaning variance can occur, or so it seems.
Therefore, I do not regard reference to ZF as a refutation of what I
said above, namely that most empirical theories can be adequately
formalized (reconstructed) only in systems of higher-order logic.
A second objection against the latter statement might consist of
pointing out that the crucial mathematical theories (analysis, geometry,
arithmetic) do not really deserve the definite article. To speak of, say,
the theory of real numbers, it might be said, is to assume clarity where
there is none. For there are various formalizations of different logical
INCOMMENSURABILITY 261
T' are incommensurable and have some terms in c o m m o n then for all
reduction relations p and all translations F there must be some term
so that F is not an P,-preserving potential translation or F does not
render T and T' P,-p-commensurable. Of course, this explication is
dependent on the given formulations of the theories in question. It
certainly would be interesting to know whether the idea comprised in
(D5) can be stated invariantly for "equivalent" formulations of the
theories involved. In the light of various ideas about " e q u i v a l e n c e " of
empirical theories already to be found in the literature 1~ such an
invariant explication of incommensurability seems feasible.
By confronting (D3)-(D5) with the examples mentioned above there
seems to be one feature of the examples which does not yet fit nicely
with the general definitions. T h e (or some of the) crucial terms which
are c o m m o n to both theories in standard expositions of the theories are
not used as basic concepts. So in the usual treatments of relativity
" c l o c k " is no basic concept, and the same applies to "freely moving
b o d y " and " m o t i o n " in their respective theories. But it is easy to
enlarge the scope of (D3)-(D5) by including defined terms in addition
to primitives. If we replace "P," in (D3)-(D5) by some symbol which is
either primitive, or introduced by an extension by definitions then all
the intuitive considerations remain valid, and all relevant examples -
including those mentioned above - can be covered.
T o summarize, from the argument stated above I would not conclude
that Stegmiiller's hypothesis concerning incommensurability and re-
duction has to be rejected. I rather would conclude that incom-
mensurability should not be confused with non-translatability in the
weak sense of "potential translation" of (D2). If incommensurability is
understood in the way proposed in (D3)-(D5) as being linked with
certain forms of non-preserving potential translations then both the
concept of incommensurability and of translation become more rich
and interesting. The possibility of rendering incommensurable theories
comparable (though not with full meaning preservation), and thus the
possibility of scientific progress is maintained.
NOTES
t This paper was written during a fellowship at the Netherland Institute for Advanced
Studies (NIAS), 1982/83. I am indebted to T. Kuhn and W. Stegmiillerfor illuminating
discussions of the topic, and to G. Berger for correcting my English.
INCOMMENSURABILITY 267
REFERENCES