state, in terms not involving the target concept, the conditions necessary
and suYcient for something to satisfy that concept. Conceptual reductions
are part of the staple fare of analytic philosophy. Thus, the attempted
analysis of knowledge asjustiWed true belief, is an attempted reduction of
our notion of knowledge to other terms. The historical record of analytic
philosophy does not hold out very much hope of success inWnding concep-
tual reductions of any but a very few concepts. Despite the collective
labour of several generations of analytical philosophers, there is hardly a
single case where a reductive deWnition is generally agreed upon. Typically,
the search for reductive deWnitions has followed the same depressing
pattern. First, a deWnition is proposed (as in ‘knowledge = justiWed true
belief’); then counterexamples to the proposal are found (as in the Gettier-
cases for knowledge); then the analysis is patched up to accommodate
these counterexamples; but then further counterexamples to the new
analysis are found; and so on. So we appear to have quite good inductive
grounds for saying thatthere are no conceptual reductions to be found(or at
least none which are informative). And so,a fortiori, it is unlikely that
there are any conceptual reductions of semantic concepts to be had, either.
A possible explanation for the general unavailability of reductive con-
ceptual analyses is provided by recent work in cognitive psychology, as
both Stich (1992) and Tye (1992) point out. A variety of experimental data
suggest that concepts are stored, not in the form of statements of necessary
and suYcient conditions, but rather asprototypes. A prototype is a rep-
resentation of a prototypical member of the kind, including a (weighted)
set of prototypical properties, and perhaps also including a perceptual
paradigmderived from acquaintance with one or more members of the
kind. A prototype for the conceptdog, for example, would includechases
cats, has dogs as parents, eats bones, barks when angry or afraid, is a
mammal, wags its tail when happy, and so on, together with a perceptual
template, or paradigm, derived from experience of one or more examples.
But there is no suggestion that dogs must necessarily have all of these
features. Rather, deciding whether something is a dog is a matter of
judging whether it is suYciently similar to the prototypical dog. (And
judgements of ‘suYcient similarity’, in turn, may be context-sensitive, and
vary with background purposes.) If the conceptknowledgehas a similar
sort of structure, then it is obvious why all attempts to give a reductive
analysis of the concept, in a statement of necessary and suYcient con-
ditions for knowledge, have failed. And if semantic concepts, too, share
prototype structure, then it looks unlikely that any reductive analysis can
be provided for them, either.
Metaphysical reductions, in contrast, focus on the worldly properties in
question, rather than on our conceptions of them. The classic form of such
reductions is inter-theoretic, as when the gas temperature–pressure laws
Naturalisation versus reduction 185