ABSTRACT

This paper describes simulations of implicit learning experiments. It compares simulations using connectionist models with existing simulations using symbolic models. It addresses an interesting issue raised by proponents of symbolic models, namely, the claim that implicit learning is better modeled by symbolic rule learning programs. This paper revisits such an issue by quantitatively comparing connectionist simulations with symbolic ones, in the context of the serial reaction time task of Lewicki et al (1987). This comparison is interesting because it helps to clarify, to some extent, some long standing confusions compounded by many claims and counter-claims. It also points to the idea of hybrid connectionist and symbolic models.