Professional Documents
Culture Documents
QUINE
reservations. One reservation has to do with the fact that some statements
are closely linked to observation, by the process of language learning.
These statements are indeed separately susceptible to tests of observation;
and at the same time they do not stand free of theory, for they share
much of the vocabulary of the more remotely theoretical statements.
They are what link theory to observation, affording theory its empirical
content. Now the Duhem thesis still holds, in a somewhat literalistic way,
even for these observation statements. For the scientist does occasionally
revoke even an observation statement, when it conflicts with a well
attested body of theory and when he has tried in vain to reproduce the
experiment. Bat the Duhem thesis would be wrong if understood as im-
posing an equal status on all the statements in a scientific theory and thus
denying the strong presumption in favor of the observation statements.
It is this bias that makes science empirical.
Another reservation regarding the Duhem thesis has to do with breadth.
If it is only jointly as a theory that the scientific statements imply their
observable consequences, how inclusive does that theory have to be?
Does it have to be the whole of science, taken as a comprehensive theory
of the world?
We should note that the sciences do link up more systematically than
people are apt to realize who forget about logic and mathematics; for
logic is shared by all branches of science, and much of mathematics is
shared by many. People tend unduly to see the logical and mathematical
components of science as different in kind from the rest, and hence fail
to see these components as something common to all the branches.
Ironically, this very neutrality, this fact of being shared by all branches
of science, has encouraged people to think of the logical and mathematical
components as different in kind from the rest, and hence to fail to recog-
nize the unity that they confer. Thus I see science as a considerably inte-
grated system of the world even now, though the explicit reduction of
major branches to theoretical physics is incomplete.
But we can appreciate this degree of integration and still appreciate
how unrealistic it would be to extend a Duhemian holism to the whole
of science, taking all science as the unit that is responsible to observation.
Science is neither discontinuous nor monolithic. It is variously jointed,
and loose in the joints in varying degrees. In the face of a recalcitrant
observation we are free to choose what statements to revise and what
EMPIRICALLY E Q U I V A L E N T S Y S T E M S OF T H E W O R L D 315
ones to hold fast, and these alternatives will disrupt various stretches o f
scientific theory in various ways, varying in severity. Little is gained by
saying that the unit is in principle the whole of science, however defensible
this claim may be in a legalistic way.
This is the end of my digression on the thesis of holism. I shall be con-
cerned from here on with the empirical under-determination of natural
science. It is a doctrine which, as I said, is plausible insofar as it is intelli-
gible. M y purpose will be to examine the meaning of this thesis more
closely, and to consider its limits and its consequences.
A notion that is evidently central to the thesis is that of observation.
This notion is subject to a curious internal tension. Observation affords
the sensory evidence for scientific theory, and sensation is private. Yet
observation must be shared if it is to provide the common ground where
scientists can resolve their disagreements. The observation must be the
distillate, somehow, of what is publicly relevant in the private sensations
of present witnesses. This delicate process of distillation is already
accomplished, happily, in our most rudimentary learning of language.
One learns the word 'blue' from another speaker, in the presence o f
something blue. The other speaker has learned to associate the word with
whatever inscrutable sensation it may be that such an object induces in
him, and one now learns to associate the word with the sensation, same
or different, that the object induces in oneself. All agree in calling the
object blue, and even in calling their sensations blue.
We do well to recognize this crucial role of language, for in view of it
we can spare ourselves the finicky task of defining the elusive notion of
observation. We can speak rather of observation terms and observation
sentences, thus cleaving to linguistic forms: 'blue', 'This is blue'. Obser-
vational expressions can be roughly distinguished from others by a be-
havioral criterion, involving no probing of sensations. For this is charac-
teristic of them: witnesses will agree on the spot in applying an observa-
tion term, or in assenting to an observation sentence, if they are conver-
sant with the language. Their verdicts do not vary with variations in their
past experience.
It is sometimes objected that a specialist may recognize at a glance
what the untrained observer cannot. The above behavioral criterion
settles this discrepancy in favor of the untrained observer. Specialists rest
content with the level of evidence that commands their expert agreement
316 w . v . QUINE
but in principle they usually could reduce this recondite evidence to ob-
servation terms at the layman's level.
N o t always. There is expertise in tea tasting, in wine tasting, and in the
recognition of tones, chords, and timbre, that resists conversion to the
c o m m o n coin. 1 We should like to be able to reckon the esoteric terms in
these domains as observational for the experts, contrary to the proposed
behavioral criterion. And indeed even a c o m m o n observation term such
as 'blue' has its penumbra of vagueness, where witnesses m a y disagree
in their verdicts. The really distinctive trait of observation terms and
sentences is to be sought not in concurrence of witnesses but in ways of
learning. Observational expressions are expressions that can be learned
ostensively. They are actually learned ostensively in some cases and dis-
cursively in others, but each of them could be learned by sufficiently per-
sistent ostension. The behavioral manifestation of observationality, then,
namely, the ready concurrence of witnesses, serves merely as a rough
practical criterion.
Observation sentences are not incorrigible. A witness who has assented
to an observation sentence on the spot is permitted to reconsider his
verdict later in the face of conflicting theory. And of course the typical
observation terms are not subjective in reference, but objective. 'Blue'
was one; others are 'water', 'rabbit', 'ball', 'hard'. They recur in theoreti-
cal sentences.
N o w the doctrine that is up for clarification is that scientific theory is
under-determined by observable events. So we shall want to get clear on
the relation of theories to observations. Or, now that we have taken to
talking of observation sentences and terms, rather than of observations,
let us look to the relation of theories to observation sentences.
An observation sentence is an occasion sentence: it commands assent
on some occasions and not others, depending on what is happening where
and when the sentence is queried. On the other hand the sentences of
scientific theory are standing sentences. They are meant to be true or false
independently of the occasion of utterance. The observation sentences
cannot, as occasion sentences, be implied by theory; we must first change
them into standing sentences, by incorporating specifications of place-
times. Let us adopt, then, an arbitrary numerical system of spatio-tempo-
ral coordinates, and let us contemplate the infinite totality of what I shall
call pegged observation sentences. Each observation sentence expressible
E M P I R I C A L L Y E Q U I V A L E N T S Y S T E M S OF T H E W O R L D 317
within our own language, of standard logical form. But even within these
cozy limits I shall not want simply to identify a theory with the logical
Consequences of a theory formulation. The next consideration will show
why.
Take some theory formulation and select two of its terms, say 'electron'
and 'molecule'. I am supposing that these do not figure essentially in any
observation sentences; they are purely theoretical. Now let us transform
our theory formulation merely by switching these two terms throughout. 3
The new theory formulation will be logically incompatible with the old:
it will affirm things about so-called electrons that the other denies. Yet
their only difference, the man in the street would say, is terminological;
the one theory formulation uses the technical terms 'molecule' and
'electron' to name what the other formulation calls 'electron' and 'mole-
cule'. The two formulations express, he would say, the same theory.
Someone else might urge, however perversely, that they express very differ-
ent theories: both of them treat of molecules in the same sense but dis-
agree sharply regarding the behavior of molecules, and correpondingly
for electrons. Clearly, in any event, the two theory formulations are
e m p i r i c a l l y e q u i v a l e n t - that is, they imply the same observation condi-
tionals. I think, moreover, that we should individuate theories in such a
way as to agree with the man in the street: the two formulations formu-
late the same theory, despite their overt logical incompatibility. This is
why I do not want to identify a theory with the logical consequences of a
formulation. I do not want to require that two formulations of a theory
be logically equivalent, nor even logically compatible.
Certainly two formulations of a theory should be empirically equiva-
lent, in the sense just defined, even if not logically equivalent. Still em-
pirical equivalence must not be the only requirement, unless we are to
repudiate the doctrine of underdetermination out of hand; for that doc-
trine says that empirically equivalent theories can conflict. What is re-
quired of two formulations of a theory must be, in short, some relation
stronger than empirical equivalence and weaker than logical equivalence.
The unsatisfying example of 'molecule' and 'electron' consisted, in-
tuitively speaking, in switching the meanings of the two words. When the
man in the street protested that the conflict between the two theory
formulations was merely terminological, his point was that it could be
resolved by treating the one formulation as not quite English, and tram-
320 w . v . QUINE
lating its respective words 'molecule' and 'electron' into the English
words 'electron' and 'molecule'. I want to preserve this insight while
avoiding problems raised by the terms 'meaning' and 'translation'. I can
do so by appealing to little more than a permutation of vocabulary; in
the present case, of course, a mere switching of the predicates 'molecule'
and 'electron' throughout one of the theory formulations. I propose that
we count two formulations as formulations of the same theory if, besides
being empirically equivalent, the two formulations can be rendered
identical by switching predicates in one of them.
This criterion needs a little broadening, in obvious respects. Since
logically equivalent formulations were in any event to count as formula-
tions of the same theory, we should not require that a switching of terms
render formulations identical; we should only require it to render them
logically equivalent. Further, we should not limit the permutation to a
switching of two predicates; we should allow permutations of many.
Finally, it would be arbitrary to require this transformation to carry
predicates always into simple one-word predicates. The intuitive notion,
after all, was a reconstruing of predicates; and the general way of re-
construing an n-place predicate is by supplying an open sentence in n
variables, not caring whether there happens to be a word in our language
with the same extension as that open sentence.
By a reconstrual of the predicates of our language, accordingly, let me
mean any mapping of our lexicon of predicates into our open sentences
(n-place predicates to n-variable sentences). Thus the predicate 'heavier
than' might be mapped to the open sentence 'x is heavier than y', an
identity mapping changing nothing, while the predicates 'molecule' and
'electron' might be mapped to the respective open sentences 'x is an
electron' and 'x is a molecule', producing our example.
So I propose to individuate theories thus: two formulations express
the same theory if they are empirically equivalent and there is a recon-
strual of predicates that transforms the one theory into a logical equiva-
lent of the other. 4
I was able to define reconstrual more simply than otherwise only by
assuming something about the form of our language that I now ought to
make explicit. I am assuming the standard logical form of language at its
most economical; there are just truth functions, quantification, and a
finite lexicon of predicates. I made no provision for names or for functors,
EMPIRICALLY EQUIVALENT S Y S T E M S OF T H E W O R L D 321
for there are well known ways of serving the purposes of these devices
on the more austere basis.
We have now settled the individuation of theories, within our parochial
confines. We have said when to count two formulations as expressing the
same theory. Given this equivalence relation, it is a routine matter to say
what a theory is. The method is artificial but familiar: theories are the
equivalence classes of that equivalence relation. The theory expressed by
a given formulation is the class of all the formulations that are empirically
equivalent to that formulation and can be transformed into logical equi-
valents of it or vice versa by reconstrual of predicates.
It is usual in the literature to require of a theory that it be deductively
closed. In our present terms, what this means is that if you change a
formulation of a theory merely by annexing some logical consequences
of that formulation, the result will still be a formulation of the same
theory. We have insured this by requiring only that the reconstrual of
predicates render the formulations logically equivalent, not identical.
Thanks to the circumstance that any formulation is equivalent to itself
plus any of its consequences, it is easily shown that theories as I have
defined them are deductively closed.
So defined, theories are classes of theory formulations. But at this point
we must liberalize the notion of a theory formulation, so as not to be
limited to the few formulations that are physically available on paper.
W h a t are wanted rather are linguistic sequences in the abstract sense, in
their infinite variety. Each single word or letter can still be explained as
the class of all its tokens, a class of actual inscriptions, since we are as-
sured that these classes all have members variously situated in space-time.
Sentences, however, and longer expressions, are to be taken rather as
mathematical sequences of their component words or letters. An expres-
sion in this sense is a function, or class of ordered pairs; the first word
or letter of the expression is paired with the number 1, the second
with 2, and so on. In this way we can assure the existence of all expressions
however long, all theory formulations as yet unconceived, all texts as yet
unwritten; all 'possible' expressions, as one might say. Theories, finally,
are classes of formulations, hence classes of expressions in this abstract
sense; classes o f functions.
Because the question how to define a theory is interesting in itself, I
have pursued it farther than required for what I want to say about the
322 w.V. QUINE
only that such branching alternatives exist, but that they are inevitable.
Thus suppose we had an adequate theory of nature, and then we were to
add to it some gratuitous further sentences that had no effect on its
empirical content. By ringing changes on these excrescences we might get
alternative theories, logically incompatible, yet always empirically
equivalent. This gratuitous branching of theories would be of no interest
to the thesis of under-determination, since the adequate original theory was
itself logically compatible with each one of these gratuitous extensions;
they were incompatible only with one another. What the thesis of under-
determination calls for is unavoidable branching. 5 The adequate original
theory, in our imaginary example, would itself have had to be one of
several equivalent and logically incompatible theories if it was to illustrate
the thesis of under-determination.
In its full generality, the thesis of under-determination thus interpreted
is surely untenable. It must fail for weak theories, theories that imply no
rich store of observation conditionals. If the implied observation condi-
tionals (redundancies aside) are finite in number, we can simply take the
conjunction of them, a single sentence, as our theory formulation. It
contains its observation conditionals without remainder; they are all it is.
It is implied by every empirically equivalent theory, and can conflict with
none of them. Any that it conflicted with would have to be internally in-
consistent, and so not empirically equivalent.
So we see that the thesis of under-determination must fail where only
finitely many observation conditionals are implied. They, in conjunction,
are their own theory formulation. But much the same thing can happen
even where a theory irreducibly implies infinitely many observation con-
ditionals; for it may happen that these can all be encompassed by a single
universally quantified conditional, or by finitely many. Such a theory
formulation, again, affords a tight fit. No theory formulation that implies
just those same observation conditionals can conflict with it, unless indeed
it is to-inconsistent. And I think we may reject to-inconsistent theories.
The empirical content of a theory formulation is summed up in
the observation conditionals that the formulation implies. These are
material conditionals; each is a truth function of pegged observation
sentences. If we could check the truth values of all the pegged observa-
tion sentences, we could evaluate any observation conditional without
consulting the theory formulation; theory could be dispensed with. But we
324 w . v . QUINE
quences would be observation conditionals, but for Craig they can be any
sentences. Then Craig shows how to specify a second or Craig class of
sentences which are visibly equivalent, one by one, to the sentences of the
desired first class; and the remarkable thing about this second class is that
membership in it admits of a mechanical decision procedure.
In the cases that matter, these classes are infinite. Even so, the second
or Craig class evidently makes the original finite formulation dispensable,
by affording a different way of recognizing membership in the desired
first class. Instead of showing that a sentence belongs to it by deducing it
from the finite formulation, we show it by citing a visibly equivalent
sentence that belongs, testably, to the Craig class.
This result does not belie under-determination, since the Craig class
is not a finite formulation, but an infinite class of sentences. But it does
challenge the interest of under-determination, by suggesting that the
finite formulation is dispensable; and indeed the Craig class, for all its
infinitude, is an exact fit, being a class of visible equivalents of the desired
class. In fact I might say just how excessively visible these equivalences
are. Each sentence in the Craig class is simply a repetitive self-conjunc-
tion, 'ppp... p', of a sentence of the desired class.
Having said this much, I would do well to finish the Craig story. Why,
when the desired class itself is undecidable, should this Craig class of its
repetitive self-conjunctions be decidable? The trick is as follows. Each o f
the desired sentences (each of the desired observation conditionals, in our
case) is deducible from the original finite formulation. Its proof can be
coded numerically, G6del fashion. Let the number be n. Then the corre-
sponding sentence in the Craig class is the desired sentence repeated in
self-conjunction n times. The resulting Craig class is decidable. To decide
whether a given sentence belongs to it, count its internal repetitions;
decode the proof, if any, that this number encodes; and see whether it is a
p r o o f of the repeated part of the given sentence.
Obviously there is no place for any of this in practice, as Craig was the
first to emphasize. Just counting the repetitions and then decoding the
p r o o f from the G6del number would require astronomical time, and
each sentence in the Craig class would require astronomical space if it
were to be written. And, after all that, the old original finite formulation
still has to be consulted in checking the proof. Craig's point is of course
strictly theoretical, and as such it is important.
326 w . v . QUINE
As I said before, Craig's result does not refute the thesis of under-
determination, since the Craig class, for all its tightness of fit, is not a
finite formulation. However, this technicality is rather a frail reed at which
to grasp. After all, one could reasonably extend the notion of theory
formulations to apply not just to an expression but to a recursive set of
expressions. So the thesis of under-determination would seem to be
demoted to the status, at best, of a thesis affirming a certain contrast
between expressions and recursive sets of expressions.
However, I see the importance of the thesis of under-determination as
lying elsewhere. The more closely we examine the thesis, the less we seem
to be able to claim for it as a theoretical thesis; but it retains significance
in terms of what is practically feasible. A tempered version, the most
favorable available, might run as follows. We, humanly, are capable of
encompassing more true observation conditionals in a loose theory for-
mulation than in any tight system that we might discover and formulate
independently of any such loose formulation. And then the thesis would
go on to say, as before, that for each such formulation there will be
others, empirically equivalent but logically incompatible with it and
incapable of being rendered logically equivalent to it by any reconstrual
of predicates.
Even in this form the thesis is moot. It no longer stands to reason, as
it seemed at first to do. The question now is whether we are underesti-
mating the power of reconstrual of predicates. It does still stand to reason,
overwhelmingly, that any theory formulation we may hope to devise as
an adequate system of the world will be a loose one; that there will be
others empirically equivalent to it and logically incompatible with it.
This much is illustrated by the very trivial example where the words
'electron' and 'molecule' were switched, and by the half-trivial example
from Poincar6; but these incompatibilities were reconciled by reconstrual
of predicates that preserved empirical equivalence. What is moot is
whether there are also bound to be cases not thus reconcilable.
The easy way to recognize empirical equivalence of two theory formu-
lations is by seeing a reconstrual of predicates that will carry the one into
the other. So it was with the examples just mentioned. But surely this is
not the only way. We might study two incompatible theory formulations,
trying in vain to imagine an observation that could decide between them,
and we might conclude that they are empirically equivalent; we might
EMPIRICALLY E Q U I V A L E N T S Y S T E M S OF T H E W O R L D 327
Harvard University
Cambridge, Mass.
NOTES
x I am indebted to Joseph Cowan here. For valuable criticism of the paper in general
I am indebted to Burton Dreben.
To dispel a misunderstanding in Harold Morick, 'Observation and Subjectivity in
Quine', Canadian Journal o f Philosophy 1 (1974), pp. 109-127, I must stress that a
pegged observation sentence is not an observation sentence. It is a non-observational
sentence obtained by pegging an observation sentence.
8 There is substantially this idea in B. M. Humphries, 'Indeterminacy of Translation
and Theory', Journal of Philosophy 67 (1970), pp. 167-178, particularly pp. 169f.
4 Avishai Margalit has suggested to me that this amounts to equating theories that
can be formulated by the same Ramsey sentence. (F. P. Ramsey, The Foundations of
Mathematics, ed. by R. B. Braithwaite, Routledge and Kegan Pad, London, 1931,
chapter ix(A), 'Theories'.)
5 This requirement evidently disqualifies, for our purposes, an example of empirically
equivalent and logically irreconcilable theories that is offered by Clark Glymour in
his important paper 'Theoretical Realism and Theoretical Equivalence', Boston
Studies in the Philosophy o f Science, Vol. IIIV (1971), pp. 275-288. In his example the
empirical evidence is covered by the statement that there are infinitely many objects.
This content can be organized indifferently in a theory of dense order and a theory of
discrete order, and these two theories are irreconcilable, sharing, as he says, no com-
mon model. But this is a case of avoidable branching.
6 William Craig, 'Replacement of Auxiliary Expressions', Philosophical Review 65
(1956), pp. 38-55.