Year: 2003
Type: article
Abstract: Marcus (2001) argues that only those connectionist models that incorporate (classical) rules can account for the phenomenon of transfer of learning in infants.Seidenberg and Elman (1999) have tried to counter to Marcus by means of a simple recurrent network (SRN) trained on a categorization task.In this paper we show how a prediction-SRN, trained on a simple but structured pre-training set, can preserve its computational equivalency with respect to classical counterparts while eschewing the need to posit rule-governed underlying mechanisms; a criticism that has been raised against Seidenberg and Elman's categorization-based reply.
Authors: Francisco Calvo, Eliana Colunga
Institutions: Universidad de Murcia, University of Colorado Boulder
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2622930309