ABSTRACT
It is natural to believe that the current high level of interest in connectionist models of cognitive processes is attributable to the impressive accounts such models can provide of human competencies (e.g. reading, classification, skill acquisition) and that the (arguable) improvement such models provide in explanatory power over nonconnectionist models is the main impetus to their growing acceptance amongst cognitive psychologists. From a historical perspective, however, a more compelling reason for trying to understand high-level human abilities within a connectionist framework is that many decades of research, especially on elementary conditioning in animals, has led to an almost-universal consensus (although see Gallistel, 1990, for an alternative viewpoint) that learning involves moment-by-moment increments and decrements in mental associations. This connectionist view of elementary learning has achieved outstanding success (Dickinson, 1980; Fletcher et al., 2001; Hall, 1991; O’Reilly & Rudy, 2001; Schultz & Dickinson, 2000; Waelti, Dickinson, & Schultz, 2001) and thus, so long as one is willing to envisage some continuity between phylogenetically primitive processes such as conditioning and more “high-level” capacities such as reading acquisition, then pursuing a connectionist research program in cognitive psychology becomes all but mandatory.
