On localist connectionism and psychological science.

It is important to clarify potential terminological confusion with respect to our use of the terms localist and connectionist: Many cognitive psychologists accept the standard critiques launched against distributed connectionism as being equally applicable to localist connectionism. We demonstrate that this is not the case. The localist models discussed here are connectionist. The fundamental difference between this class of model and other localist models concerns the use of soft constraints as opposed to rule-based procedures. The distinction between connections and rules can be seen as different grain sizes on a continuum ranging from microscopic inferences to macroscopic inferences (note the use of the term microinference [Hinton, McClelland, & Rumelhart, 19861 to refer to the work done by connections in connectionist models). Nevertheless, neither connectionist nor classical theorists would refute that in rule-based systems, exceptions to the rule are by definition treated in a special way (i.e., by using specific routines). Connectionist models do not distinguish rule-following from rule-breaking inferences (e.g., Seidenberg & McClelland, 1989).