<< . .

( 29)

. . >>

properties be known or manifest to one, or that different people should

use the same properties of the substance in order to identify it. Any of
very numerous means of recognizing the substance may be applied.
Each of these methods will be fallible in principle. Identifying a sub-
stance and exploiting its possibilities is as fallible as any other practical
activity one engages in. One may always stumble and fall.


Although substance concepts, hence words for substances, can be used
for the purpose of classifying, the reverse does not hold in general.
There is a big difference between understanding something merely as a
class and understanding it as a substance. Conceptions used to classify
need only carve out some clear unequivocal extension within the do-
main to be classified. Conceptions that govern substance concepts must
locate genuine ontological grounds of induction. A substance concept is
distinguished by the role it is ready to play, accumulating additional
means of identification, and anticipating certain kinds of inductions as
likely to hold. A substance concept will be successful only if there really
is some substance out there it is hooked into. One reason it is an error
to place great value on operational definitions in science, for example,
is that operational definitions, as such, are merely classifiers, hence do not
necessarily correlate with substances. But it is substances rather than
classes that are of interest to science.
Because conceptions filling out substance concepts can sometimes be
used also as conceptions of classes, words for substances can vacillate be-
tween being understood as standing for substances and as standing only
for classes.When confidence is lost in the reality of a substance or in the
univocity of a substance term, it may begin to be used in a strictly clas-
sificatory way. For example, terms for many mental disorders have vac-
illated over the years between being understood as capturing substances,
naming single diseases for which single etiologies and therapies might
eventually be discovered, and as being merely classificatory, defining
useful groupings of symptoms for efficient transfer of information.
Nominals that are used only to express concepts of classes typically
are complexes built out of prior terms. They wear their analytical na-
tures on their sleeves. Their extensions are functions of unions and dis-
juncts of the extensions of the prior terms that compose them.Typically,
this sort of construction will be built up in the same way by all who
understand the syntactic forms of the language. Thus, although prior

conceptions attaching to the element terms in the complex may differ
from one speaker of the language to another, the mode of construction
of complex conceptions out of prior conceptions will be common.
These terms express analytical concepts, concepts only of classes. Sub-
stance concepts, on the other hand, are synthetical. A person™s concep-
tion of a substance may employ prior concepts used in the process of
identifying, but the substance concept is not equivalent to any mere
function of prior concepts.
But there are, of course, many exceptions to the rule that synthetical
concepts are expressed with simple nouns, analytical concepts with
compounds. Consider the term “red sulfur.” Red sulfur is not just sul-
fur that is red, but an allotrope of sulfur, a substance in its own right
with its own suite of properties different from other forms of sulfur. On
the other hand, red sulfur also happens to be the only substance that is
both red and (pure) sulfur. Does the nominal “red sulfur” correspond,
then, to a synthetical or an analytical concept?
Whether a word express an analytical or a synthetical concept may
sometimes depend on the user. For some people the concept for “red
sulfur” may be synthetical and for others analytical. A person who did
not understand that sulfur that is red happens to be a substance in its own
right would only have an analytical concept corresponding to the term
“red sulfur.” Accordingly, they would never recognize any part of the ex-
tension of “red sulfur” in any way other than by noting that it was sulfur
and also noting that it was red. And they would not attempt inductions
from samples of red sulfur to other samples of red sulfur that they would
not have attempted either from samples of red to other samples of red or
from samples of sulfur to other samples of sulfur. On the other hand, a
person might instead have a synthetical concept of “red sulfur.” That is
the kind chemists have, for example. More interesting, it would be pos-
sible to have a concept of this stuff, this substance, without knowing ei-
ther that it is red or that it is sulfur. One might recognize it as the sticky
so-smelling substance typically found in such-and-such context, and be
surprised to learn that it is always red, and that it is a kind of sulfur.
Similarly, should “Californian” correspond to a vague sort of sub-
stance, as suggested in Section 2.4, then although the “-ian” suffix sug-
gests an analytical concept, there will be at least two ways to have a
concept for “Californian,” one analytical and the other synthetical. It is
less plausible, of course, that one might have a reliable way of identifying
Californians that did not depend on first determining that they came
from California. Not every legitimate substance is reliably identifiable in

multiple ways, if one counts as “identifying in the same way,” using de-
scriptions employing concepts of the same things in the same way.
(These prior concepts might each be governed by variable conceptions,
however.) Especially, as is true in this particular case, ways of identifying
that relate directly to the real ontological ground of a substance may be
uniquely reliable.Whether a substance concept is legitimately a one cri-
terion concept or not does not rest on something ephemeral called “the
rules of our language” (wherever they live) but lies in nature. If it is
necessary that a vixen be a female fox, the deep reason is that it is sets
of fox chromosomes that include two X chromosomes, copied from
prior sets of this sort (close enough), that are responsible for causing the
characteristic likenesses among vixens. Being a female fox is the real
essence of vixenhood, nor is this a purely a priori matter.
One last difference between identifying and classifying. Classifying re-
quires recognizing that a predicate applies to some definite subject. Sup-
pose, for example, that we tried to model the act of classifying individual
objects into the red ones and the green ones as, just, responding in one
way to the red individuals, in another to the green individuals. What
would determine that these responses constituted the classifying of indi-
viduals, rather than of time slices of individuals, or facing surfaces of indi-
viduals, or dye stuffs found on the surfaces of individuals, or pattens of am-
bient light impinging on retinas? You can™t classify without some grasp of
what you are classifying.You have to be able independently to think of the
object you want to classify before you can classify it. Identifying an object,
on the other hand, does not always require something conceptually prior.
Identifying a substance as animals do, merely for practical purposes, re-
quires only that behavioral responses the animal is disposed to learn by
employing the concept should be appropriate to that substance, that is,
they should be responses that are effective because of the properties or dis-
positions of that substance.True, identifying a substance for theoretical use
does require that you have some appropriate predicate concepts, ones that
you understand as applicable to, whether or not they are true of, the sub-
stance. I am not claiming that the only things we reidentify are sub-
stances.3 But whatever it is that one classifies, it is clear that the capacity to
think of members within the domain to be classified is more fundamen-
tal than the ability to classify. Identifying is a skill prior to classifying.

3 In Millikan (1984) Chapter 16, I talked quite a lot about reidentifying properties, and the
analysis of the act of reidentifying to be offered below in Chapters 9 through 12 applies
equally to concepts of substances and concepts of properties.

The traditional view among philosophers and psychologists has been
that the ability to apply a term for a kind or a stuff is an ability to clas-
sify.This view has often taken the form of assuming that kind terms and
stuff terms are descriptive, each corresponding to some sort of configu-
ration of properties. If the concepts corresponding to these terms are
concepts of configurations of properties, their extensions are naturally
determined analytically as a direct function of the extensions of those
properties. Both among contemporary psychologists and also in some
philosophical circles, this view is still much the most common form of
a more general position I will call “conceptionism.” Conceptionism is
the view that the extension of a concept or term is determined by
some aspect of the thinker™s conception of its extension, that is, by some
method that the thinker has of identifying it. I am fully in charge of the
extensions of my concepts; whatever I am, after due consideration, dis-
posed to apply them to is what they are concepts of.
One way of identifying a substance, of course, is by means of the
knowledge that it has certain properties or falls under a certain descrip-
tion peculiar to that substance.The classic form of conceptionism holds
that the conception that determines the extension of a substance term
is such a set of properties or such a description. We can call this classic
view “descriptionism.” Another form of conceptionism holds that the
extension of certain concepts or terms is determined by means of iden-
tification procedures not employing prior concepts of properties. Con-
cepts whose extensions are thought to be determined in this direct
manner are sometimes called “recognitional concepts.” Thus Fodor
remarks “if a concept is recognitional, then having certain kinds of ex-
perience would, in principle, show with the force of conceptual neces-
sity that the concept applies” (Fodor 1999). “Conceptionism” in either
form contrasts, of course, with “direct reference” theories of conceptual


Conceptionist views “ the view that substance concepts are basically
classifiers, their extensions being determined by dispositions to apply

4 In Millikan (1998a and 1998b), I referred to both kinds of conceptionism as “description-
ism.” This resulted in understandable confusion on the part of several commentators. Here
I am shifting to what I hope is a more perspicuous terminology.

them “ underlie a surprising proportion of the masses of recent work
on “concepts” and “categories” in the psychological literature. It will be
worth spelling this out in some detail. I will try to show, indeed, that
throughout the changing variety of competing theories of concepts and
categorization developed by psychologists in the last half century, the
theoretical assumption of conceptionism, generally in the form of de-
scriptionism, has managed to go completely unchallenged. This is true,
despite the fact that Putnam™s and Kripke™s famous arguments (or at
least their conclusions) against descriptionism (Kripke 1972; Putnam
1975) have been rehearsed numerous times in the psychological litera-
ture, and despite a number of brave attempts to integrate these insights
into the psychological tradition (Gelman & Coley 1991; Keil 1989; Ko-
matsu 1992; Lakoff 1987; Markman 1989; Neisser 1987 Ch. 2).The dif-
ficulty, I believe, results from the fact that Putnam™s and Kripke™s insights
were almost entirely negative. They told us how the extensions of cer-
tain substance concepts are not determined, but they supplied no ade-
quate theories of how they are determined. Moreover, the tentative pos-
itive views that they offered focused more on the extensions of words in
a public language than on the nature of concepts, leaving obscure the
nature of the psychological states or processes that would constitute an
understanding of the meanings of the words discussed. Thus, they offered
little aid to psychologists. One aim of this book is to help remedy that.
The descriptionist holds that the referent or extent of a substance
term is determined by its falling under a description associated with the
term by the term user. Certain properties, relations, facts about origins,
facts about causes, similarities to prototypes, similarities to given exem-
plars, and so forth “ certain “information” about each portion of the
extent “ determines it to be a portion of the extent, and the thinker or
the thinker™s “mental representation” determines which information is
to play this role. In the psychological literature, this view is sometimes
found in caricature in the statement that concepts are features or prop-
erties, for example, “many properties are concepts themselves” (Barsalou
1987, p. 129).
Using the concept chair as his example, Komatsu (1992) expresses what
he claims is the most general question that psychological theories of con-
cepts have attempted to answer, thus: “. . . what information, very generally, is
represented by the concept chair, so that people are able to reason about chairs,
recognize instances of chairs, and understand complex concepts . . .”
(1992, p. 500, italics mine). Building on Medin and Smith (1981, 1984),
he applies this descriptionist formula to each of five accounts of concepts:

. . . the classical view (e.g., Katz 1972, Katz & Fodor 1963) . . . the family re-
semblance view (e.g., Rosch & Mervis 1975) . . . the exemplar view (e.g.,
Medin & Schaffer 1978) . . . the schema view [Komatsu later cites Bartlett
(1932), Minsky (1975), Piaget (1926), Rumelhardt (1980), Schank & Abelson
(1977), Winograd (1975), and Neisser (1975)] . . . the explanation-based view
(e.g., Johnson-Laird 1983, Lakoff 1987, Murphy & Medin 1985) [later he men-
tions the work of Gelman and Keil].

Descriptionism is most obviously compatible with nominalism, the
view that the members of the kinds that words name are grouped to-
gether either conventionally, according to the dictates of culture, or
according to patterns natural to human perception and thought. For
example, heavily sprinkled throughout the literature we find refer-
ences to “learning about people™s categorization decisions.” On this
view, the descriptions that govern concepts have their source either in
the conventions of society, or in peculiarities of the human perceptual
and cognitive systems, that is, in ways it is natural to us to generalize.
For example, in classical studies of concept learning, subjects were
typically set the task of learning imaginary categories defined by ar-
bitrarily chosen sets of properties, and many studies exploring family
resemblance or prototype or exemplar views of categorization have
also set arbitrary tasks. The view that the human mind has its own
ways of imposing various groupings of things into kinds, ways that
languages must respect in order to be learnable, has been evident es-
pecially since Rosch™s work on color categories (e.g., Rosch 1973,
1975). In this tradition, the psychological problem concerning cate-
gorization is understood to be that of ferreting out exactly what these
psychologically imposed principles are “ those principles in accor-
dance with which children or adults “prefer to sort” (Markman
1989). Thus, for example, Lakoff subtitles his 1987 book, “What Cat-
egories Reveal about the Mind.”
But descriptionism need not be allied with nominalism or conven-
tionalism. It also has been combined with realism about human cate-
gories.The realist holds that many of our categories correspond to kinds
that are grouped together by nature independently of mind. As we ac-
quire categories, we learn not merely, say, how to communicate with
others, but how to grasp structures that were already there in nature.
The view of substances that I am advocating is, of course, a variety of
realist view. It might seem that there is an incompatibility between re-
alism and descriptionism. If the extent of a category is determined by
nature, then it cannot be determined by fitting a certain description as-

sociated with a word. But in fact there are a number of ways in which
realism and descriptionism have been combined.
The simplest way is to take the extent of a substance term to be
fixed by one, or a set, of definite descriptions of the substance. Whether
it is supposed that the description is used rigidly or nonrigidly makes
no difference in this context. In either case, the thinker entertains a
prior description that determines the extent of his word or category.
Thus the classical twentieth-century view was that Aristotle himself was
a natural unit in nature, and that to have a concept of Aristotle was to
capture him in thought under a description such as “the teacher of
Alexander,” or under a suitable disjunct of descriptions. Similarly, there
has been a tendency in the psychological literature to interpret Kripke™s
(1972) and Putnam™s (1975) apparently antidescriptionist views on the
meaning of proper names and natural kind terms as invoking definite
descriptions on a metalevel. Kripke is thought to have claimed that the
referent of a proper name N is fixed in the user™s mind by the descrip-
tion “whoever was originally baptized as N,” and Putnam is thought to
have claimed that the extent of a natural kind term is fixed for laymen
by the description “whatever natural kind the experts have in mind
when they use term T.” This is what Fumerton calls “Russelling,” a the-
ory of direct reference (Fumerton 1989). It transforms it, of course, into
a descriptionist theory instead.
Theories that language categories are organized “probablistically”
(Medin 1989) by family resemblance or by reference to prototypes of-
ten combine realism with descriptionism. Families and prototypes are
usually taken to center over highly correlated properties, and these cor-
relations are taken to be empirically discovered. Thus, prototype theory
is naturally compatible with the view that many concepts end up paired
with real kinds. But probablistic theories are regularly interpreted as ex-
plaining only how the learner™s experience generates the category. Then
the actual extension of the category is taken to be determined, not by
the real extent of a kind, but by how the learner is inclined to classify
new examples. The same is true of exemplar theories and for variations
on these two views.Thus, Billman suggests that we should compare and
test psychological models of structure and processing of concepts by ex-
amining the function from “learning instances plus the target items to
categorize” to “the set of possible category judgments” (Billman 1992,
p. 415, emphasis mine) and Ward and Becker state that “category struc-
ture” can mean “the set of items that the learner considers to be mem-
bers of the category in question (i.e., the category extension)” (1992,

p. 454). Made explicit, the idea here seems to be that experience with a
natural kind may inspire the category, but the category extent is deter-
mined by the thinker™s potential decisions on exemplars. When all goes
well, our psychologically determined kinds may contain the same mem-
bers as the natural ones, that is all. Similarly, the realists Gelman and
Byrnes tell us, explicitly making reference to Chomsky™s theory of in-
nate grammar, that “[w]e can determine how languages and conceptual
systems are constrained by examining the forms and meanings that chil-
dren construct, and which errors they fail to make” (1991, p. 3).That is,
it is the child™s inclinations that constrain the concepts.
Most explicitly realist in their approach to concepts are contempo-
rary researchers holding what Komatsu calls an “explanation-based
view” of concept structure. Komatsu (1992) characterizes this view by
quoting Keil:

No individual concept can be understood without some understanding of how
it relates to other concepts. Concepts are not probablistic distributions of fea-
tures or properties, or passive reflections of feature frequencies and correlations
in the world; nor are they simple lists of necessary and sufficient features. They
are mostly about things in the world, however, and bear nonarbitrary relations
to feature frequencies and correlations, as well as providing explanations of
those features and correlations. If it is the nature of concepts to provide such
explanations, they can be considered to embody systematic sets of beliefs “ be-
liefs that may be largely causal in nature. (Keil 1989, p. 7)

Note that the view is not just that concepts designate kinds for which
there exist explanations of property correlations, but that the concept
actually consists in essential part of an understanding or, looking beyond
Page 1 of Keil™s text, a partial understanding of these explanations. In-
terpreting this in the terms of the last chapter, the concept consists in
part of a partial understanding of the ontological ground of induction
that underlies the concept. Of particular interest to the explanation the-
orists, for example, has been Medin™s work showing that people behave
as though believing that beneath their categories there are hidden
essences making the things in the categories what they are (e.g., Medin
and Ortony 1989). Keil, Carey, Gelman, and Markman are among those
who have done very interesting work tracing the development of chil-
dren™s natural kind concepts and artifact concepts, for example, docu-
menting the transition from reliance on superficial characteristic prop-
erties for identification of these kinds to use of rudimentary and then

more sophisticated “theories” about the underlying causes of the unity
of the kind.
These advocates of explanation-based views have remained strongly
influenced by the characteristic mid-twentieth-century doctrine that a
“theory” is a set of inference connections among concepts, that the net
of theory in which a concept is caught up determines its “meaning,” and
that the meaning of a concept determines its reference. Thus, to intro-
duce or change theories threatens to change meanings hence reference:

How can one be sure that one is even talking about the same concept at all if all
concepts are relative to theories? . . . We do not want every change in theoretical
beliefs to make the concepts embedded in them completely different from those
that were embedded before the change; yet no precise method is offered [by
Smith, Carey, & Wiser 1985] for making a decision. . . . These are difficult issues,
and it is hardly surprising that they are not yet resolved. (Keil 1989, pp. 21“2)

Following Smith, Carey, and Wiser, Keil speaks of “ ˜tracking™ con-
cepts across theory change” and agrees with them that probably
“[d]escent can be traced . . . because of several properties of theories
that stay fixed through change” (Smith, Carey, & Wiser 1985, p. 182).
And he agrees with Fodor that it is not obvious how the classical
view could be true that “children and adults could have different
kinds of concepts for the same terms,” for that makes it seem as
though [quoting Fodor 1972] “they must misunderstand each other
essentially” (Fodor, 1972 p. 88; Keil 1989, pp. 15“16). Again, the view
here is conceptionist. There is no suggestion that the extent of the
concept, its “meaning” in the most fundamental sense, might be di-
rectly fixed by the extent of a natural unit in nature, reference re-
maining the same while conceptions change. (For an exception to
this, see Gopnik & Meltzoff 1996.)
My claim is that all this traditional work, supposedly on the nature of
concepts, has actually concerned merely conceptions. For example,
changes in theories about the underlying grounds of induction for spe-
cific kinds of substances are changes in conceptions, not in what the
corresponding substance concepts are of. You cannot have a theory
about something, say, about what makes dogs to be dogs (remember that
Aristotle had a false theory about this), unless you can first think of the
thing you would theorize about. Changes in your theory are not, then,
changes in your concept or in its extension. It does not follow that these
psychological studies, when they have concerned substance concepts,

have not had, and will not continue to have, great value. Nor does it fol-

<< . .

( 29)

. . >>

Copyright Design by: Sunlight webdesign