ABSTRACT
Concern has recently developed regarding the possibility that parallel distributed processing models will exhibit massive amounts of retroactive interference. McCloskey & Cohen (in press) have suggested that such models exhibit “catastrophic interference” under realistic training conditions; they conclude that PDP models may not be able to simulate basic aspects of human performance. In this paper we report replications and extensions of simulations on which these claims were based. The new simulations suggest that “catastrophic interference” may be less of a problem than McCloskey & Cohen suggest; specifically, it is related to the use of a rigid training scheme that bears little resemblance to how children actually learn.
