Aprendizaje secuencialmodelos de reglas versus modelos conexionistas

Abstract

This study reviews three different accounts of the sequence learning produced in paradigms using deterministic rules to generate the stimulus material. The first one, which assumes the acquisition of algorithms identical to the rules defined by the experimenters (e.g., Lewicki, Hill & Bizot, 1988), is proved to be inadequate considering the results of subsequent studies (e.g., Perruchet, Gallego & Savy, 1990). These authors proposed an alternative view based on the learning of frequency rules for simple movements. However, an in-depth analysis of new results (Ferrer-Gil, 1994) shows that none of these mechanisms could underlie these learning effects. As an alternative, it is proposed that learning could arise from the action of some connectionist mechanisms of learning (Cleeremans, 1993). The simulation of the main results reported by Ferrer-Gil (1994) on a simple recurrent network strengthens this conclusion.