Hampton, James, and Danièle Dubois. 1993. "Psychological Models of Concepts: Introduction," in Categories and Concepts: Theoretical Views and Inductive Data Analysis, editors Ivan Van Mechelen, James Hampton, Ryszard S. Michalski, and Peter Theuns, 11-33. ix, 364 pp. Cognitive Science Series, General Editors. Barry Richards, and Keith Stenning, London, UK: Academic Press.
Chapter introduces first half of book, which is generally devoted to psychological models of concepts and categories. Authors present definitions of commonly-used terms, differing views of concepts, brief commentary on data-analytic methods (subject of the last half of the book), and challenges to existing views of concepts. In the course of discussing these topics, they preview later chapters. The definitions and challenges sections are extremely useful, clearly written summaries of the literature.

Following are some definitions of key terms provided in the first section, in order of appearance. Any computer science analogies are my own:
concept—"idea or notion by which an intelligence is able to understand some aspect of the world" (13), related to intension.
category—"a class or set of entities... which are grouped together on the basis of some criterion or rule" (13), related to extension. Whether concepts determine categorization or vice versa is a "central issue."
intension—the metadata, database schema, "information that is used for classification and the possible inferences that classification allows" (14).
extension—the data, database extent, "the members of a category" (14).
instance—an individual category member, as in an instance of an object class in Smalltalk.
exemplar—synonym for instance, although either may also be used to denote a subclass.
typicality—category gradedness, aka prototypicality, representativeness, goodness of example; "primarily defined as an extensional phenomenon.... can also, however be applied to the intensional attributes of concepts. Some attributes will be commonly considered to be more central to the definition of a concept than others..." such as feathers on a bird. An extra-logical phenomenon regarded by philosophers to be irrelevant.
property—"any predicate that can be asserted of some or all of the members of a category" (15), which may involve involve an implicit "theory" or knowledge of cultural context. This is a general term that encompasses attribute and feature.
attribute—an enumerated type, a dimension, which may take on one of many mutually exclusive values.
value—"The attributes form the dimensions or aspects on which entities in a domain may differ, while the values provide the alternative forms those aspects can take" (15).
feature—a boolean ("two-valued") attribute
frame—a "structural relation", a complex data type such as a C++ object class, complete with default attribute values and validation rules concerning interactions between different attributes, e.g. for oranges the attribute value "unripe" implies a value of "green" for the color attribute. See Barsalou & Hale (Ch. 5) for more on frames.

Under "Differing views of concepts," authors outline issues around "realism and psychologism," and the possible relationships between intension and extension (i.e., classical view, prototype view). Most cognitive scientists work from the assumption that concepts exist a priori, "either in the 'real' physical world, or in an ideal (platonistic)one, waiting, in each case, for our minds to discover them. The alternative (constructivist) view that concepts are conventional creations of human societies, and are therefore relative to particular cultures and historical contexts, is less common in cognitive science" (17) [explaining why ethnomusicologists reflexively tune out on cognitive science, maybe]. The two kinds of a priori knowledge, empiricist and rationalist, are understood as bottom-up and top-down processes, respectively, both of which a "cognitivist" uses to model concept representation. Different people may hold different representations of the same concept, hence to speak of the concept of chair is to speak of an "abstract societal norm" (17). In contrast to cognitivism, philosophical realism dispenses with mental representations and insists that all true concepts exist in the real world and that have a status independent of language, and that language can mislead us into becoming preoccupied with pseudo-concepts. Realism is associated with the so-called "classical view" of concepts and categories as the logical application of necessary and sufficient conditions, which has been criticized for being inflexible, insensitive to context, and incapable of accounting for vagueness. Prototypes are definitely a psychological theory, and have been criticized for not providing an adequate basis for logical reasoning, and for relying too much on attribute clustering and similarity and ignoring how such structures relate to "naïve theories" of objects, object history ("boiled celery is no longer crisp"), function and behavior, other objects. Frame theorists advocate for recursive frames as a solution to the shortcoming of prototype models.

With respect to data-analytic models, which have developed outside of psychology, authors explain that such models are limited to applying a list of features to a data set, which can be useful for revealing "conceptual structure," although there are problems with interpreting results if the semantics underlying the binary coding (e.g., true/false vs. present/absent vs. large/small) are not made explicit. In a footnote, they comment that the selection of features by the analyst already makes the procedure less than purely inductive. "However, this inductive procedure for forming categories cannot be taken seriously as a psychological model of concept formation.... Our perception, identification, and interpretation of an object or event depend crucially on general and particular theoretical beliefs that we possess.... [yet] There may be situations in which we lack any a priori beliefs and have to rely on the data to show us inductive generalizations that can be made" (23). The jury is out as to "what kinds of naive data-analytic method we possess ourselves," that is whether or not we learn to categorize by intuitively accumulating statistics on occurrences of objects in the world.

Methodological criticisms to categorization research: 1) individual introspection [such as eliciting attribute lists, I suppose] as a source of data is highly limited, variable, and inconsistent, and must be averaged to have any significance. 2) Averaging data, however, leads to the possibility that graded category structure "may just reflect the variability with which individuals have appropriated the social norm, and structural properties such as extensional gradedness within a category may reflect distributions of different representations within the group, rather than individual structure" (24-25). 3) The universality of findings is in question due to the limited sample population of the body of studies done (mostly college students). 4) Studies depend heavily on language: "Words are used to stand for concepts, and it can be argued that the results reflect facts about word meanings rather than about concepts" (25). That is, a word with several different meanings does not necessarily represent a vague concept, rather it could be seen as representing different well-defined concepts in a context-dependent way. Authors see a crucial problem therefore in tasks where subjects are asked to give sets of ratings or categorizations with no context provided other than that of the experiment [as I found when I tried to give a single "prototypicality" rating for a recording, not knowing how weight the various kinds of things that I noticed]. They see that cross-linguistic studies may help to disentangle language from concepts. 5) There is a neglect of process modelling: how do subjects arrive at their ratings? Prototype ratings have been found to be unstable within the same subject, so is the variability in the process or in the data used to make the decision?

Theoretical criticisms of categorization research: 1) defining concepts in terms of other concepts, such as in attribute lists or recursive frames, is a potentially circular activity, therefore "it is commonly supposed that [at bottom there must be] perceptual and behavioural components, mainly determined at a subsymbolic perceptual level, and constrained by physiological structure" (28). 2) concepts understood as theories, which relate them to other concepts which are expressed in terms of other theories, are also prey to circularity. Ways out: platonic primitives, or cultural designation of the concepts that are not further decomposable. 3) realist (logico-philosophical) analysis of concepts not grounded in empirical data leaves no basis for judging one analysis over another; "There is nothing in the realist position to prevent the contruction of arbitrarily absurd conceptual structures, provided that they are logically consistent. (This problem derives from the reliance on logic, which explicitly eschews semantic content.)" (29). Authors take a dim view of realist position, saying that even studies of expert knowledge show that the classical view is not a useful model for how diagnoses or decisions are or can be made. They cite an unpublished manuscript by Chater and Oaksford, who conclude that "common-sense ontology will never provide a proper basis for a correct understanding of the world" (30), although it is unclear if this statement supports or runs counter to their argument against the realists.
P128.C37.C38 1993 Ed-P.