James L. McClell
Dept. of Psychol., Carnegie Mellon Univ., Pittsburgh, PA 15213
Connectionist models provide mechanisms in which all of the different types of knowledge thought to be relevant to language understanding---syntax, prosody, semantics, context, etc.---can be brought together in a coherent, integrated manner to address the problem of extracting the intended meaning of an utterance. The approach is different from many other approaches in several respects, one of which is its treatment of the role of words in constructing representations of meaning. Words do not have meanings that show up as constituents of meaning representations; rather, words constrain the construction of learned representations sufficient to exhibit meaningful language use. Earlier work illustrating the ideas [M. D. St. John and J. L. McClelland, Artif. Intell. 46, 217--357 (1990)] and more recent work in collaboration with John Reger will be presented. [Work supported by NSF and NIMH.]