ABSTRACT
Concepts are generally viewed as stable representations in long-term memory. As such they are assumed to participate in 'higher' cognitive tasks such as linguistic comprehension. Language understanding is commonly taken to require accessing and combining the concepts associated with linguistic input. Within this framework, a concept typically consists of a set of 'features'. In the most classical (and dominant) interpretation of this last statement, the features of a concept define its boundaries in terms of necessary and sufficient conditions. In light of the problems entailed by such rigidity, the notion of prototypes has been proposed. In his recent work, Franks (1995) argues that neither of these approaches can adequately address the difficulties of assembling concepts. Instead, he puts forth a "quasi-classical" approach to concept combination. In his model, features of a concept each receive one value, but may be defeated or modified through contextual effects. His detailed formal approach focuses in particular on the treatment of "privative" combinations such as stone lion, fake gun, apparent friend, etc. Complex rules and structures are developed to address these problematic examples, which are too often ignored in existing symbolic or connectionist work. More precisely, Franks suggests distinguishing between the "lexical concept" (i.e., the stable information represented in the mental lexicon) and the "sense" of a linguistic unit in context. Most importantly, the sense is derived from the lexical concept. A theory of concept combining thus consists in specifying the rules of derivation of sense from lexicon, subject to contextual constraints.
