Professional Documents
Culture Documents
Kemeny-Oppenheim1956 Article OnReduction
Kemeny-Oppenheim1956 Article OnReduction
On Reduction
by JOHN G. KEMENY, DXaTMOUTrICOLL~.CE,and
PAUL O P P E N H E I M
New Detinitions
As we see it, the essence of reduction cannot be understood by compar-
ing only the two theories; we must bring in the observations. It is not the
case that the vocabulary of T2 is in any simple way connected with the
vocabulary of T1, but only that T1 can fulfill the role T2 played, i.e., that
it can explain all that T2 can and normally more. This suggests the follow-
ing modification of the given definitions. W e take as fundamental the
concept of the reduction of a theory T2 by means of TI relative to observa-
tional data O. W e abbreviate this as Red(T1,T2,O).
Definition 3. Red(T1,T2,O) if:
(1) Voc(T2) contains terms not in Voc(T1). °
(2) Any part of O explainable by means of T2 is explainable by T~.1°
(3) T1 is at least as well systematized as T2.
W e must ask, however, what we can do about the undesired parameter O.
One possibility is to put tO for O, that is, relativize reduction with respect
to the present moment. Thus we get the concept that Tx today reduces T2.
Definition 4. Redt(T~,T2) if and only if Red(T1,T2,0t).
14 PHILOSOPHICAL STUDIES
And the corresponding internal reduction concept:
Definition 5. Intred t(Ti,T~) if and only if Redt(T1,T~) and Voc(T1) is
a proper subset of Voc(T~).
The above is a very restricted notion of reduction, but no doubt some
authors have this in mind. If we do not want to put a particular value of O
into Definition 3, we must eliminate the undesired variable by quantifica-
tion. And this seems to lead to a very fruitful approach.
Definition 6. Red(T1,T2) if for every O consistent with T~,
Red (T1,T2,O).ll
Definition 7. Intred(Ti,T~) if Red(T1,T2) and Voc(Ti) is a proper sub-
set of Voc(T~).
There is a natural way of basing the reduction of branches on these
concepts, namely by taking the theoretical body of the branch (at a given
time) as the T~ in question} 2
Delinition 8. Red(B1,B2,t) if and only if there is a theory T1 in B1 at t
such that Red (Ti,tTB~).
Definition 9. Intred(B1,B2,t) if and only if Red(Bi,B2,t) and B1 is a
branch of B~.
W e would like, of course, to make these definitions independent of t.
There is the obvious procedure of choosing t as "the present moment,"
which appears to be Nagers procedure. But there is no analogue of the
quantification of O. It is reasonable to require that the reducing theory
should serve in lieu of the reduced one no matter what the facts turn out
to be, but it is not reasonable to require that it should reduce every theory--
true or false. Hence in this case we must relativize to the present moment.
Definitions 6-9 represent our proposed definition of reduction.
Let us study definitions 6 and 7 (definitions 8 and 9 are just special
cases). Condition 1 in the former requires that Voc(T2) have certain
terms not in Voc(T1), and in the latter definition this is strengthened by
requiring that in addition Voc(T1) be a subset of Voc(T~). The other
two conditions are the same for both definitions. Condition 3 requires that
T1 be at least as well systematized as T~. W e could, if desired, strengthen
this to "better systematized." Condition 2 is the only one to which the
universal quantifier applies. Thus we have: For any observational state-
ment O, consistent with T2, if T2 can explain the same part.
This formulation allows a variety of interpretations of "explain." But
let us use the commonest one: TM T can explain a part of O if there are two
"non-overlapping" parts, O~ and O~, such that T - O 1 implies 02. Thus
the above condition becomes: If, for two given "non-overlapping" observa-
ON REDUCTION 15
tional statements (such that T2 is consistent with both) T2" O1 implies
02, then so does TI" O1. In particular, O1 may be empty (analytically
true). Thus a necessary, condition is that if T2 implies an observational
statement, then so does T1. But this is also suftieient; because if T2" O1
implies 02, then T2 implies O1 D O~, hence TI implies O1 D 02, hence
TI" O1 implies 02. This establishes the following theorem:
Theorem 1. Red(T1,T2) if and only if (1) Voc(T2) contains terms not
in Voe(T1), (2) every observational statement implied by T2 is also
implied by T1, and (3) T1 is at least as well systematized as 312. We have
Intred(T1,T2) if and only if the same three conditions hold, and in
addition Voe(T1) is a subset of Voe(T2).
In many special eases (in particular when the universe of discourse is
finite) there will be a strongest observational consequence of T2, say T*~.
In these eases we may replace the second condition by the requirement
that T1 implies T*~.
Next we want to show that the Woodger and Nagel definitions are spe-
eial eases of ours. At least this is the ease if the definitions are modified by
changing C to C'. and since there is some doubt about the desirability of
requirement B2, we will prove our results irrespective of whether B2 is
included or not.
Theorem 2. If in Definition 1 condition C is replaced by ~V-C', then it
defines a special ease of Definition 7. And the same holds even if B2 is
omitted.
Proof: The only condition of Theorem 1 that is not obviously fulfilled is
the second condition. Let T'2 be the translation of T2 by means of the
bieonditionals. Take any interprettaion of Voc(T1) that makes T1 true.
Since T'~ follows from T1, it too must be true under this interpretation.
Now extend the interpretation to Voe(T2) by making the P in each
bieonditional synonymous with its M. This makes all the bieonditionals
true. And since according to these T2 and T'2 are equivalent, T2 must
also be true under the interpretation. Thus any observational statement
implied by T2 must be true. This shows that under any interpretation
making T1 true, an observational statement following from T2 must
also be true. Which proves that any observational statement following
from T~ must also follow from T1. (And B2 was not used.) Q.E.D.
Quite analogously we can prove:
Theorem 3. If in Definition 2 condition C is replaced by N-C', then it
defines a special ease of Definition 8. And the same holds even if B2 is
omitted.
The question arises of why a method of translation should be essential
16 PHILOSOPHICAL STUDIES
for Woodger and Nagel but not for us. The answer lies in the fact that
they attempt to establish a direct connection between two theories, while
our connection is indirect. If one theory is to follow from another, it must
be translatable into the vocabulary of the latter. 14 But it is entirely possible
that a theory should be able to explain all facts that another can, without
there being any method of translation. Of course, each set of theoretical
terms must be connected to observational terms, and hence to each other,
but this connection is normally much weaker than a full translation.
Naturally, we do not exclude the possibility that reduction may be ac-
complished by means of a translation. That is why the previous definitions
(at least in the modified form) are special cases of our corresponding defi-
nitions. But we maintain that they cover what is an extremely special case.
W e have offered four definitions for types of reductions, but it must not
be supposed that there is any sharp distinction between them. Generally
a case of reduction is classified as of one of the four types, but that does
not exclude that it could have been classified otherwise. Given a case of
Red(T1,T~), we could just as well have described it as the internal reduc-
tion Intred(T1,Tl"T2). Or we could have taken the branches BI and B2
which have T1 and T~_ as their theories (i.e., if we are willing to call them
branches), and speak of Red (B~,B2,t). Or we could take Ba to be the union
of B1 and B o and speak of Intred(B~,Ba,t). It all depends on what we wish
to recognize as a single theory (or body of theories) and what we call a
branch. It is rather that there is one process of reduction that is describ-
able from four different points of view.
Take as an example the reduction of pre-Newtonian mechanics (includ-
ing celestial mechanics) to Newton's theories. W e may think of each body
of theories as reduced to the newly created body, and then take this as an
example of Definition 6. Or we may think of the totality of previous
theories (which presumably used all of Newton's theoretical terms) as
reduced, in which case it is an example of Definition 7. And in this case
there is pretty nearly a method of translation, so this is as near as we ever
come to an example of Definition 1. But when we pass to the reduction of
Newtonian mechanics to relativity theory, there is nothing that even re-
motely resembles the translation by means of biconditionals.
Similarly for the reduction of branches. The more or less complete re-
duction of chemistry to physics has been used as a standard example. It is
optional whether we take this as an example of Definition 8 or whether
we speak of the reduction of physico-chemistry to physics, in which case
it is an example of Definition 9. And while some attempts have been
made to show that something like biconditional translation does exist, this
certainly is not the case if quantum mechanics is taken as the reducing
science.
ON REDUCTION 17
In conclusion, we hope to have shown that the two previous definitions
were too narrow in that they excluded most actual cases of reduction. W e
have presented reasons for the belief that narrowness is an unavoidable
fault of any definition trying to establish direct connections between
theories. Of course, we are open to correction on this point. But if this
is right, then our much wider alternative is the natural choice.
Independently of this we argued for the inclusion of an additional con-
dition, namely that the reducing theory be as well (or better) systematized
as the theory reduced. It is on these two points that our definition must
be judged.
It must be pointed out that our definition shares with the previous ones
a serious oversimplification. W e are ignoring the fact that the theory to
be reduced may be only approximately true, and only with certain restric-
tions at that. W e certainly would not want to require that an incorrect
prediction of T2 should be a consequence of T1. W e might suggest that
it is some modification T'2 of T2 that is actually reduced to T1.1~ But such
a T'2 is not usually formed, and it may be very difficult to formulate it. In
addition, we would still be ignoring the fact that T'2 holds only approxi-
mately.
Further Research
There are several related concepts which deserve consideration. The
authors hope to discuss them in a future paper. They are here listed in
the hope that others may be interested in some of these problems.
While Definitions 6 and 7 seem to answer all questions concerning the
reduction of one theory to another, there are some questions about the
reduction of branches not answered by Definitions 8 and 9. W e often hear
speculation about whether a branch B2, which is not today reduced by B1,
is in some sense "reducible" to it. Presumably this involves some hypothesis
about the future development of science. W e should also consider cases
in which, although B2 is not reduced to B1, some part of it is. This would
lead to a concept of "partial reduction" and possibly even a numerical
measure of the "reductive power" of a branch of science.
There is an especially important special case of reduction that deserves
independent treatment. This is reduction by means of a micro-theory.
There is also a clear connection to Feigl's "levels of explanation, ''~6 and
it would appear to be fruitful to carry his ideas further by means of the
definitions here offered.
Received ]uly 24, 1955
NOTES
In this example the economy is achieved by eliminating some biological terms.
This reduces the number of theoretical terms in science as a whole. But it is not neces-
sarily the case that the reducing branch has fewer terms than the branch reduced. It
18 PHILOSOPHICAL STUDIES
may be the case that physics has fewer terms than biology, but this question is never
considered in the reduction-literature.
2There is an excellent discussion of the reduction of theories of heat to statistical
mechanics by E. Nagel in his "The Meaning of Reduction in the Natural Sciences,"
in Science and Civilization, edited by R. C. Stauffer (Madison: University of Wisconsin
Press, 1949).
s While the terms of Voc(T~) now become logically superfluous, there may be good
extra-logical grounds for keeping them. The replaceable terms may be less abstract than
the terms of Voc(Tx), and hence more convenient to work with. It may also be the case
that Tx is less well confirmed than T~, and hence we hesitate to make a final replace-
ment--at least till further e~Sdence is available.
' C f . J. H. Woodger, Biology and Language (Cambridge: Cambridge University
Press, 1952), and E. Nagel, "Mechanistic Explanation and Organismic Biology," in
Philosophy and Phenomenological Research, vol. 11, no. 3 (March 1951). W e hope
that we have done no injustice in translating these definitions into the terminology of
the present paper.
Nagel makes use of the same type of argument as was used by Poincar6 to show
that science cannot give us value-judgments, since a value-statement cannot follow from
a factual premise: The argument is based on the supposed fact that a term missing from
the premises cannot occur in the conclusion of an argument, in any essential form. But
from the premise P(a) we can conclude P(a) v Q(a), showing that the argument
is not strictly correct. (This counter-example was shown to the authors by C. G. Hempel,
who is apparently the first one to have noticed the loophole.) Nevertheless, while the
argument is not correct, the conclusion is essentially all right. While predicates not
in the premises (and independent of terms in the premises) may occur in the conclu-
sion, they play an inessential role. For example, if such a predicate is replaced every-
where by its negation, the conclusion still follows.
6 While the concept of the simplicity of a theory is still in need of considerable study,
some partial results toward its explication are available. In some cases there will be com-
plete agreement among scientists as to which of two theories is simpler; for example, if
one can be stated in a single line, while the other--using the same vocabulary--requires
several pages to state, then there would be little room for argument. The first two steps
toward an explication were taken by Karl Popper in Log/k der Forschung (Vienna:
Julius Springer, 1935) sections 39-464, and by John G. Kemeny in "The Use of Sim-
plicity in Induction," PhilosophicaIReview, vol. 625, no. 3 (July 1953). This concept
of simplicity is a purely syntactic one applicable to theories as a whole. It must be
distinguished from certain nonsyntactic concepts, as well as from Nelson Goodman's
measure of the simplicity of sets of predicates.
Since in a translation atomic sentences are, in general, replaced by molecular ones,
there is little doubt as to the general validity of this claim.
s The authors are indebted to C. G. Hempel for clarifying their thinking on this point.
9 That this condition occurs in all definitions of reduction is due to the historic usage
of 'reduction.'
1oThe usage of 'explain' will be discussed below. There is a definite connection be-
tween this condition and the concept of "systematic power." The present condition im-
plies that sp(Ta,O) > or --~ sp(T~,O), but it is stronger than the latter. Cf. "Systematic
Power," by the present authors, in Philosophy of Science, vol. 22, no. 1 (January 1955).
"Every O" means, of course, every conceivable set of observational data, which in
turn means that O can be any observational statement. W e require that O be consistent
with T~, since otherwise we get queer explanations in which the "initial conditions" are
inconsistent wtih the theory, and hence anything follows from the two together. There
is no harm, however, in allowing O to be inconsistent with T1. Similarly, it does no
harm to apply the definitions to a self-contradictory T1. The resulting "reductions" are
merely uninteresting.
"Branch" is here taken in a syntactic, rather than pragmatic, sense. B1 is a branch
of B~ if its theoretical vocabulary is a proper subset of the theoretical vocabulary of B~.
FURTHER REMARKS ON DEFINITION AND ANALYSIS 19
Cf. C. G. Hempel and Paul Oppenheim, "The Logic of Explanation," Philosophy
of Science, vol. 15 (1948). While this definition is adequate for many purposes, it is
oversimplified in some ways.Theorem 1 depends on the form of this definition, but our
general approach is consistent with any explieatum of 'explain.'
1~Cf. footnote 6.
i~For example, T'~ may state that T~ holds under specifiedconditions.
le Cf. H. Feigl's contribution to the symposium on operationism in the Psychological
Review, vol. 52, no. 5 (September 1945).
A NUMBERof acute criticisms have been directed against some of the views
presented in my essay "Analytical Philosophy and Analytical Propositions"
(this journal, December 1953). I welcome them as providing stimulation
and opportunity for penetrating more deeply into the issues involved.
The central notion in my discussion is that of a theoretical de/init/on.
A theoretical definition of a term is intended to formulate a theoretically
adequate characterization of the objects to which that term applies. One
gives a theoretical definition of a term to attach to the term, as intension,
t h a t property which in the context of a given theory is most useful for
understanding or predicting the behavior of those entities which comprise
the (usual) extension of that term.
In his "Definitions in Analytical Philosophy" (this journal, April 1954)
Michael Scriven has correctly observed that a given theoretical definition
is compatible with a wide range of different theories (p. 38). Several dif-
ferent theories may have the same vocabulary (of terms or of concepts), or
they may have parts of their vocabularies in common. And that common
area of vocabulary may be all that is used in the definition of a given
term. With respect to a theoretical definition, therefore, Mr. Scriven is
right in saying that it is incorrect to talk of the theory which underlies it.
I am therefore happy to accept his suggestion that my remarks on this
point should be modified. I would say now that accepting a theoretical
definition involves accepting as correct some theory in whose terminology
the definition is formulated.
It should be clear, however, that a theoretical definition may be disputed
for either (or both) of two different reasons. One may dispute a theoretical
definition because of disagreement with the theory one believes to under-