Connectionism, classical cognitive science and experimental psychology

There has been an enduring tension in modern cognitive psychology between the computational models available and the experimental data obtained. Standard computational models have assumed the symbolic paradigm: that it is constitutive of cognitive processes that they are mediated by the manipulation of symbolic structures. Such schemes easily handle formal inferences, and memory for arbitrary symbolic material. However, context-sensitive defeasible inference and content-addressable memory retrieval have remained problematic. By contrast, in the empirical data on human memory and inference, the opposite profile is observed. Everyday mundane reasoning is both context dependent and defeasible, and yet is performed easily and naturally, whereas subjects are typically unable to perform the simplest formal reasoning task (Wason and Johnson-Laud 1972; Evans 1982). In memory, content-addressable access in knowledge-rich domains seems natural and unproblematic for human subjects, whereas people can retain only very small quantities of arbitrary material. Despite this tension between experiment and theory, Fodor and Pylyshyn (1988) have recently reaffirmed what they term the “classical symbolic paradigm”. That is, they argue that symbolic cognitive processes are autonomous from their implementation. Thus they question the relevance of connectionist theorizing for psychology, and suggest that connectionism should be viewed as a theory of implementation for autonomous classical architectures