22

Metaphor in science
THOMAS S. KUHN

If I had been preparing the main paper on the role of metaphor in science,
my point of departure would have been precisely the works chosen by
Boyd: Max Black's well-known paper on metaphor (Black, 1962b), to-
gether with recent essays by Kripke and Putnam on the causal theory of
reference (Kripke, 1972; Putnam, 1975a, 1975b). My reasons for those
choices would, furthermore, have been very nearly the same as his, for we
share numerous concerns and convictions. But, as I moved away from the
starting point that body of literature provides, I would quite early have
turned in a direction different from Boyd's, following a path that would
have brought me quickly to a central metaphorlike process in science, one
which he passes by. That path I shall have to sketch, if sense is to be made
of my reactions to Boyd's proposals, and my remarks will therefore take
the form of an excessively condensed epitome of parts of a position of my
own, comments on Boyd's paper emerging along the way. That format
seems all the more essential inasmuch as detailed analysis of individual
points presented by Boyd is not likely to make sense to an audience largely
ignorant of the causal theory of reference.
Boyd begins by accepting Black's "interaction" view of metaphor. How-
ever metaphor functions, it neither presupposes nor supplies a list of the
respects in which the subjects juxtaposed by metaphor are similar. On the
contrary, as both Black and Boyd suggest, it is sometimes (perhaps always)
revealing to view metaphor as creating or calling forth the similarities upon
which its function depends. With that position I very much agree and,
lacking time, I shall supply no arguments for it. In addition, and presently
more significant, I agree entirely with Boyd's assertion that the open-

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
534 THOMAS S. KUHN
endedness or inexplicitness of metaphor has an important (and I think
precise) parallel in the process by which scientific terms are introduced and
thereafter deployed. However scientists apply terms like "mass," "electric-
ity," "heat," "mixture," or "compound" to nature, it is not ordinarily by
acquiring a list of criteria necessary and sufficient to determine the refer-
ents of the corresponding terms.
With respect to reference, however, I would go one step further than
Boyd. In his chapter, the claims for a parallel to metaphor are usually
restricted to the theoretical terms of science. I suppose that they often hold
equally for what used to be called observation terms, for example "dis-
tance," "time," "sulphur," "bird," or "fish." The fact that the last of these
terms figures large in Boyd's examples suggests that he is unlikely to dis-
agree. He knows as well as I that recent developments in philosophy of
science have deprived the theoretical/observational distinction of anything
resembling its traditional cash value. Perhaps it can be preserved as a
distinction between antecedently available terms and new ones introduced
at particular times in response to new scientific discoveries or inventions.
But, if so, the parallel to metaphor will hold for both. Boyd makes less than
he might of the ambiguity of the word "introduced." Something with the
properties of metaphor is often called upon when a new term is introduced
into the vocabulary of science. But it is also called upon when such terms -
by now established in the common parlance of the profession - are intro-
duced to a new scientific generation by a generation that has already
learned their use. Just as reference must be established for each new ele-
ment in the vocabulary of science, so accepted patterns of reference must
be reestablished for each new cohort of recruits to the sciences. The tech-
niques involved in both modes of introduction are much the same, and they
therefore apply on both sides of the divide between what used to be called
"observational" and "theoretical" terms.
To establish and explore the parallels between metaphor and reference
fixing, Boyd resorts both to the Wittgensteinian notion of natural families
or kinds and to the causal theory of reference. I would do the same, but in a
significantly different way. It is at this point that our paths begin to diverge.
To see how they do so, look first at the causal theory of reference itself. As
Boyd notes, that theory originated and still functions best in application to
proper names like "Sir Walter Scott." Traditional empiricism suggested
that proper names refer by virtue of an associated definite description
chosen to provide a sort of definition of the name: for example, "Scott is
the author of Waverley." Difficulties immediately arose, because the choice
of the defining description seemed arbitrary. Why should being the author
of the novel Waverley be a criterion governing the applicability of the name
"Walter Scott" rather than a historical fact about the individual to whom
the name, by whatever techniques, does refer? Why should having written
Waverley be a necessary characteristic of Sir Walter Scott but having writ-

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
Metaphor in science 535
ten Ivanhoe a contingent one? Attempts to remove these difficulties by
using more elaborate definite descriptions, or by restricting the characteris-
tics on which definite descriptions may call, have uniformly failed. The
causal theory of reference cuts the Gordian knot by denying that proper
names have definitions or are associated with definite descriptions at all.
Instead, a name like "Walter Scott" is a tag or label. That it attaches to
one individual rather than to another or to no one at all is a product of
history. At some particular point in time a particular infant was baptized or
dubbed with the name "Walter Scott," which he bore thereafter through
whatever events he happened to experience or bring about (for example,
writing Waverley). To find the referent of a name like "Sir Walter Scott" or
"Professor Max Black," we ask someone who knows the individual about
whom we inquire to point him out to us. Or else we use some contingent
fact about him, like his authorship of Waverley or of the paper on meta-
phor, to locate the career line of the individual who happened to write that
work. If, for some reason, we doubt that we have correctly identified the
person to whom the name applies, we simply trace his life history or lifeline
backward in time to see whether it includes the appropriate act of baptism
or dubbing.
Like Boyd, I take this analysis of reference to be a great advance, and I
also share the intuition of its authors that a similar analysis should apply to
the naming of natural kinds; Wittgenstein's games, birds (or sparrows),
metals (or copper), heat, and electricity. There is something right about
Putnam's claim that the referent of "electric charge" is fixed by pointing to
the needle of a galvanometer and saying that "electric charge" is the name
of the physical magnitude responsible for its deflection. But, despite the
amount that Putnam and Kripke have written on the subject, it is by no
means clear just what is right about their intuition. My pointing to an
individual, Sir Walter Scott, can tell you how to use the corresponding
name correctly. But pointing to a galvanometer needle while supplying the
name of the cause of its deflection attaches the name only to the cause of
that particular deflection (or perhaps to an unspecified subset of galva-
nometer deflections). It supplies no information at all about the many
other sorts of events to which the name "electric charge" also unambigu-
ously refers. When one makes the transition from proper names to the
names of natural kinds, one loses access to the career line or lifeline which,
in the case of proper names, enables one to check the correctness of differ-
ent applications of the same term. The individuals which constitute natural
families do have lifelines, but the natural family itself does not.
It is in dealing with difficulties like this one that Boyd makes what I take to
be an unfortunate move. To get around them he introduces the notion of
"epistemic access," explicitly abandoning in the process all use of "dubbing"
or "baptism" and implicitly, so far as I can see, giving up recourse to
ostension as well. Using the concept of epistemic access, Boyd has a number

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
536 THOMAS S. KUHN
of cogent things to say both about what justifies the use of a particular
scientific language and about the relation of a later scientific language to the
earlier one from which it has evolved. To some of his points in this area I shall
be returning. But despite these virtues, something essential is lost, I think, in
the transition from "dubbing" to "epistemic access." However imperfectly
developed, "dubbing" was introduced in an attempt to understand how, in
the absence of definitions, the referents of individual terms could be estab-
lished at all. When dubbing is abandoned or shoved aside, the link it pro-
vided between language and the world disappears as well. If I understand
Boyd's chapter correctly - something I do not take for granted - the prob-
lems to which it is directed change abruptly when the notion of epistemic
access is introduced. Thereafter, Boyd seems simply to assume that the
adherents of a given theory somehow or other know to what their terms
refer. How they can do so ceases to concern him. Rather than extending the
causal theory of reference, he seems to have given it up.
Let me therefore attempt a different approach. Though ostension is basic
in establishing referents both for proper names and for natural kind terms,
the two differ not only in complexity but also in nature. In the case of
proper names, a single act of ostension suffices to fix reference. Those of
you who have seen Richard Boyd once will, if your memories are good, be
able to recognize him for some years. But, if I were to exhibit to you the
deflected needle of a galvanometer, telling you that the cause of the deflec-
tion was called "electric charge," you would need more than good memory
to apply the term correctly in a thunderstorm or to the cause of the heating
of your electric blanket. Where natural-kind terms are at issue, a number
of acts of ostension are required.
For terms like "electric charge," the role of multiple ostensions is diffi-
cult to make out, for laws and theories also enter into the establishment of
reference. But my point does emerge clearly in the case of terms that are
ordinarily applied by direct inspection. Wittgenstein's example, games, will
do as well as another. A person who has watched chess, bridge, darts,
tennis, and football, and who has also been told that each of them is a
game, will have no trouble in recognizing that both backgammon and
soccer are games as well. To establish reference in more puzzling cases -
prize fights or fencing matches, for example - exposure is required also to
members of neighboring families. Wars and gang rumbles, for example,
share prominent characteristics with many games (in particular, they have
sides and, potentially, a winner), but the term "game" does not apply to
them. Elsewhere I have suggested that exposure to swans and geese plays
an essential role in learning to recognize ducks (Kuhn, 1974). Galvanome-
ter needles may be deflected by gravity or a bar magnet as well as by
electric charge. In all these areas, establishing the referent of a natural-
kind term requires exposure not only to varied members of that kind but
also to members of others - to individuals, that is, to which the term might

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
Metaphor in science 537
otherwise have been mistakenly applied. Only through a multiplicity of
such exposures can the student acquire what other authors in this book (for
example, Cohen and Ortony) refer to as the feature space and the knowl-
edge of salience required to link language to the world.
If that much seems plausible (I cannot, in a presentation so brief, hope to
make it more), then the parallel to metaphor at which I have been aiming
may be apparent as well. Exposed to tennis and football as paradigms for
the term "game," the language learner is invited to examine the two (and
soon, others as well) in an effort to discover the characteristics with respect
to which they are alike, the features that render them similar, and which
are therefore relevant to the determination of reference. As in the case of
Black's interactive metaphors, the juxtaposition of examples calls forth the
similarities upon which the function of metaphor or the determination of
reference depend. As with metaphor, also, the end product of the interac-
tion between examples is nothing like a definition, a list of characteristics
shared by games and only games, or of the features common to both men
and wolves and to them alone. No lists of that sort exist (not all games
have either sides or a winner), but no loss of functional precision results.
Both natural-kind terms and metaphors do just what they should without
satisfying the criteria that a traditional empiricist would have required to
declare them meaningful.
My talk of natural-kind terms has not yet, of course, quite brought me to
metaphor. Juxtaposing a tennis match with a chess game may be part of
what is required to establish the referents of "game," but the two are not,
in any usual sense, metaphorically related. More to the point, until the
referents of "game" and of other terms which might be juxtaposed with it
in metaphor have been established, metaphor itself cannot begin. The
person who has not yet learned to apply the terms "game" and "war"
correctly can only be misled by the metaphor, "War is a game," or "Profes-
sional football is war." Nevertheless, I take metaphor to be essentially a
higher-level version of the process by which ostension enters into the estab-
lishment of reference for natural-kind terms. The actual juxtaposition of a
series of exemplary games highlights features which permit the term
"game" to be applied to nature. The metaphorical juxtaposition of the
terms "game" and "war" highlights other features, ones whose salience had
to be reached in order that actual games and wars could constitute separate
natural families. If Boyd is right that nature has "joints" which natural-kind
terms aim to locate, then metaphor reminds us that another language might
have located different joints, cut up the world in another way.
Those last two sentences raise problems about the very notion of joints
in nature, and I shall return to them briefly in my concluding remarks about
Boyd's view of theory change. But one last point needs first to be made
about metaphor in science. Because I take it to be both less obvious and
more fundamental than metaphor, I have so far emphasized the metaphor-

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
538 THOMAS S. KUHN
like process which plays an important role in finding the referents of scien-
tific terms. But, as Boyd quite rightly insists, genuine metaphors (or, more
properly analogies) are also fundamental to science, providing on occa-
sions "an irreplacable part of the linguistic machinery of a scientific
theory," playing a role that is "constitutive of the theories they express,
rather than merely exegetical." Those words are Boyd's, and the examples
which accompany them are good ones. I particularly admire his discussion
of the role of the metaphors which relate cognitive psychology to computer
science, information theory, and related disciplines. In this area, I can add
nothing useful to what he has said.
Before changing the subject, however, I would suggest that what Boyd
does say about these "constitutive" metaphors may well have a bearing
wider than he sees. He discusses not only "constitutive" but also what he
calls "exegetical or pedagogical" metaphors, for example those which de-
scribe atoms as "miniature solar systems." These, he suggests, are useful in
teaching or explaining theories, but their use is only heuristic, for they can
be replaced by nonmetaphorical techniques. "One can say," he points out,
''exactly in what respects Bohr thought atoms were like solar systems with-
out employing any metaphorical devices, and this was true when Bohr's
theory was proposed."
Once again, I agree with Boyd but would nevertheless draw attention to
the way in which metaphors like that relating atoms and solar systems are
replaced. Bohr and his contemporaries supplied a model in which electrons
and nucleus were represented by tiny bits of charged matter interacting
under the laws of mechanics and electromagnetic theory. That model re-
placed the solar system metaphor but not, by doing so, a metaphorlike
process. Bohr's atom model was intended to be taken only more-or-less
literally; electrons and nuclei were not thought to be exactly like small
billiard or Ping-Pong balls; only some of the laws of mechanics and electro-
magnetic theory were thought to apply to them; finding out which ones did
apply and where the similarities to billiard balls lay was a central task in the
development of the quantum theory. Furthermore, even when that process
of exploring potential similarities had gone as far as it could (it has never
been completed), the model remained essential to the theory. Without its
aid, one cannot even today write down the Schrodinger equation for a
complex atom or molecule, for it is to the model, not directly to nature,
that the various terms in that equation refer. Though not prepared here and
now to argue the point, I would hazard the guess that the same interactive,
similarity-creating process which Black has isolated in the functioning of
metaphor is vital also to the function of models in science. Models are not,
however, merely pedagogic or heuristic. They have been too much ne-
glected in recent philosophy of science.
I come now to the large part of Boyd's chapter that deals with theory
choice, and I shall have to devote disproportionately little time to my

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
Metaphor in science 539
discussion of it. That may, however, be less of a drawback than it seems, for
attention to theory choice will add nothing to our central topic, metaphor.
In any case, with respect to the problem of theory change, there is a great
deal about which Boyd and I agree. And in the remaining area, where we
clearly differ, I have great difficulty articulating just what we disagree
about. Both of us are unregenerate realists. Our differences have to do
with the commitments that adherence to a realist's position implies. But
neither of us has yet developed an account of those commitments. Boyd's
are embodied in metaphors which seem to me misleading. When it comes
to replacing them, however, I simply waffle. Under these circumstances, I
shall attempt only a rough sketch of the areas in which our views coincide
and in which they appear to diverge. For the sake of brevity in that at-
tempt, furthermore, I shall henceforth drop the distinction on which I have
previously insisted between metaphor itself and metaphorlike processes. In
these concluding remarks, "metaphor" refers to all those processes in
which the juxtaposition either of terms or of concrete examples calls forth a
network of similarities which help to determine the way in which language
attaches to the world.
Presupposing what has already been said, let me summarize those por-
tions of my own position with which I believe Boyd largely agrees. Meta-
phor plays an essential role in establishing links between scientific language
and the world. Those links are not, however, given once and for all. Theory
change, in particular, is accompanied by a change in some of the relevant
metaphors and in the corresponding parts of the network of similarities
through which terms attach to nature. The earth was like Mars (and was
thus a planet) after Copernicus, but the two were in different natural
families before. Salt-in-water belonged to the family of chemical com-
pounds before Dalton, to that of physical mixtures afterwards. And so on. I
believe, too, though Boyd may not, that changes like these in the similarity
network sometimes occur also in response to new discoveries, without any
change in what would ordinarily be referred to as a scientific theory. Fi-
nally, these alterations in the way scientific terms attach to nature are not -
logical empiricism to the contrary - purely formal or purely linguistic. On
the contrary, they come about in response to pressures generated by obser-
vation or experiment, and they result in more effective ways of dealing with
some aspects of some natural phenomena. They are thus substantive or
cognitive.
These aspects of Boyd's and my agreement should occasion no surprise.
Another one may, though it ought not. Boyd repeatedly emphasizes that
the causal theory of reference or the concept of epistemic access makes it
possible to compare successive scientific theories with each other. The
opposing view, that scientific theories are incomparable, has repeatedly
been attributed to me, and Boyd himself may believe I hold it. But the
book on which this interpretation is imposed includes many explicit exam-

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
540 THOMAS S. KUHN
pies of comparisons between successive theories. I have never doubted
either that they were possible or that they were essential at times of theory
choice. Instead, I have tried to make two rather different points. First,
comparisons of successive theories with each other and with the world are
never sufficient to dictate theory choice. During the period when actual
choices are made, two people fully committed to the values and methods of
science, and sharing also what both concede to be data, may nevertheless
legitimately differ in their choice of theory. Second, successive theories are
incommensurable (which is not the same as incomparable) in the sense that
the referents of some of the terms which occur in both are a function of the
theory within which those terms appear. There is no neutral language into
which both of the theories as well as the relevant data may be translated for
purposes of comparison.
With all of this I believe, perhaps mistakenly, that Boyd agrees. If so,
then our agreement extends one step further still. Both of us see in the
causal theory of reference a significant technique for tracing the continu-
ities between successive theories and, simultaneously, for revealing the
nature of the differences between them. Let me provide an excessively
cryptic and simplistic example of what I, at least, have in mind. The tech-
niques of dubbing and of tracing lifelines permit astronomical individuals -
say, the earth and moon, Mars and Venus - to be traced through episodes
of theory change, in this case the one due to Copernicus. The lifelines of
these four individuals were continuous during the passage from heliocentric
to geocentric theory, but the four were differently distributed among natu-
ral families as a result of that change. The moon belonged to the family of
planets before Copernicus, not afterwards; the earth to the family of plan-
ets afterwards, but not before. Eliminating the moon and adding the earth
to the list of individuals that could be juxtaposed as paradigms for the term
"planet" changed the list of features salient to determining the referents of
that term. Removing the moon to a contrasting family increased the effect.
That sort of redistribution of individuals among natural families or kinds,
with its consequent alteration of the features salient to reference, is, I now
feel, a central (perhaps the central) feature of the episodes I have previ-
ously labeled scientific revolutions.
Finally, I shall turn very briefly to the area in which Boyd's metaphors
suggest that our paths diverge. One of those metaphors, reiterated through-
out his chapter, is that scientific terms "cut [or can cut] nature at its joints."
That metaphor and Field's notion of quasi-reference figure large in Boyd's
discussion of the development of scientific terminology over time. Older
languages succeeded, he believes, in cutting the world at, or close to, some
of its joints. But they also often committed what he calls ureal errors in
classification of natural phenomena," many of which have since been cor-
rected by "more sophisticated accounts of those joints." The older lan-
guage may, for example, "have classified together certain things which

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
Metaphor in science 541
have no important similarity, or [may] have failed to classify together,
things which are, in fact, fundamentally similar" (italics added). This way
of talking is, however, only a rephrased version of the classical empiricists'
position that successive scientific theories provide successively closer ap-
proximations to nature. Boyd's whole chapter presupposes that nature has
one and only one set of joints to which the evolving terminology of science
comes closer and closer with time. At least, I can see no other way to make
sense of what he says in the absence of some theory-independent way of
distinguishing fundamental or important similarities from those that are
superficial or unimportant.l
To describe the successive-approximation view of theory change as a
presupposition does not, of course, make it wrong, but it does point to the
need for arguments missing from Boyd's paper. One form such arguments
might take is the empirical examination of a succession of scientific theo-
ries. No pair of theories will do, for the more recent could, by definition, be
declared the better approximation. But, given a succession of three or
more theories directed to more-or-less the same aspects of nature, it should
be possible, if Boyd is right, to display some process of bracketing and
zeroing in on nature's real joints. The arguments which would be required
are both complex and subtle. I am content to leave open the question to
which they are directed. But my strong impression is that they will not
succeed. Conceived as a set of instruments for solving technical puzzles in
selected areas, science clearly gains in precision and scope with the passage
of time. As an instrument, science undoubtedly does progress. But Boyd's
claims are not about the instrumental effectiveness of science but rather
about its ontology, about what really exists in nature, about the world's real
joints. And in this area I see no historical evidence for a process of zeroing
in. As I have suggested elsewhere, the ontology of relativistic physics is, in
significant respects, more like that of Aristotelian than that of Newtonian
physics. That example must here stand for many.
Boyd's metaphor of nature's joints relates closely to another, the last I
shall attempt to discuss. Again and again, he speaks of the process of
theory change as one which involves "the accommodation of language to
the world." As before, the thrust of his metaphor is ontological; the world
to which Boyd refers is the one real world, still unknown but toward which
science proceeds by successive approximation. Reasons for being uneasy
with that point of view have already been described, but this way of express-
ing the viewpoint enables me to phrase my reservations in a different way.
What is the world, I ask, if it does not include most of the sorts of things to
which the actual language spoken at a given time refers? Was the earth
really a planet in the world of pre-Copernican astronomers who spoke a
language in which the features salient to the referent of the term "planet"
excluded its attachment to the earth? Does it obviously make better sense
to speak of accommodating language to the world than of accommodating

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024
542 THOMAS S. KUHN
the world to language? Or is the way of talking which creates that distinc-
tion itself illusory? Is what we refer to as "the world" perhaps a product of a
mutual accommodation between experience and language?
I shall close with a metaphor of my own. Boyd's world with its joints
seems to me, like Kant's "things in themselves," in principle unknowable.
The view toward which I grope would also be Kantian but without "things
in themselves" and with categories of the mind which could change with
time as the accommodation of language and experience proceeded. A view
of that sort need not, I think, make the world less real.

NOTE
1 In revising the manuscript to which this paragraph and those following are
addressed, Boyd has pointed out that both natural kinds and nature's joints may
be context- or discipline- or interest-relative. But, as note 2 to his paper will
indicate, that concession does not presently bring our positions closer together.
It may do so in the future, however, for the same root note undermines the
position it defends. Boyd concedes (mistakenly, I think) that a kind is "un-
'objective' " to the extent that it is context- or discipline-dependent. But that
construal of "objective" requires that context-independent bounds be specified
for context-dependence. If any two objects could, in principle, be rendered
similar by choice of an appropriate context, then objectivity, in Boyd's sense,
would not exist. The problem is the same as the one suggested by the sentence to
which this footnote is attached.

Downloaded from https://www.cambridge.org/core. University of Sussex Library, on 25 Jul 2018 at 18:32:59, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139173865.024