Nominator's statement

Rosch's ideas, like Fodor's, have had more of a sociological than a substantive influence. For although, unlike Fodor's modularity, they were somewhat empirically based, all the empirical evidence (and I mean all of it) ended up going against them: It is not true in general that internal representations" (whatever they turn out to be) are based on prototypes" (if a "prototype" means some sort of template of a general case which can be deformed to match particular cases) -- except in a later series of Rosch-inspired experiments that have DEFINED category membership in terms of prototype-matching (generating artificial stimuli based on template-deformation), which begs the general question. It is even less true that internal representations are based on "family resemblances" itself a bit of a doubtful, if not incoherent, notion in anything but an intuitive sense). If we interpret "family resemblance" in a concrete, coherent way, it can only mean "disjunctive features" -- that is, categories whose members have either this feature or that feature or that feature, not necessarily all of them. This makes sense as a rival to a conjunctive feature" theory of category representation -- according to which every member of a category MUST have all of the "essential" features except that no one ever held such a conjunctivist view. (Note that we are talking about cognition here, and how people categorize things, not about ontology, and what things really "are"). As an alternative to the so-called "classical view" -- that category membership is based on having a set of features that is "necessary and sufficient" for being a member -- it should be obvious that "disjunctivism" (once recognized for what it really is) would be a nonstarter, for disjunctive features are perfectly classical! Besides, "necessity" is a term of logic, whereas categorization is an empirical, probabilistic, and biological capacity that organisms have, nothing to do with necessity (or ontology). Real-world feature-detectors "satisfice"; they do not necessitate; outside mathematics, only God and other ontologists can do the latter. How did the Roschian findings and conclusions side-track us for decades from the so-called "classical" view, as if there had been something wrong with it whereas in reality it is trivial and true)? By changing the subject: Instead of investigating HOW we (or any system) are able to successfully categorize as and when we do, the Roschian studies focused on how QUICKLY we do it, how CONFIDENT we are in our judgments (i.e., how "typical" we judge a member to be) and how we THINK we are doing it (introspection drew only blanks, or demonstrably wrong hunches). Because, empirically, response time is continuous, typicality judgments are matters of degree, and introspection did not come up with any all-or-none features, it was concluded that category membership (and hence its underlying representation") was "graded" -- a matter of degree, rather than all or none. (Tell that to the robins, ostriches and penguins, all of which are birds," even if not 100% typical! And then repeat the exercise for most of the content words in the dictionary. Except for obvious polar continua, such as "big/small," most will prove to be all-or-none, classical, and feature-based, just as they were before the Roschian years began. Membership in categorical categories continues to be categorical, not graded or a matter of degree... Some (but not all) of this might be blameable on Wittgenstein (q.v.), but perhaps the sociological impact of Rosch's work is confirmed by the number of words required to say all this... Stevan Harnad

comments

• Rosch makes the important contribution of the notion of the prototype" to categorization. The prototype is opposed to theories of class membership based on checklist properties. Rather, categorization relies on folk theories, or simple, generalizations of experience. This notion has been very important to linguistics and is a forebearer of Lakoff's work on metaphor.

• Her work is still relevant - even as a reaction point to the development of new models.