A Computational Model of Routine Procedural Memory

Abstract

Cooper and Shallice (2000) implemented a computational version of the Norman and Shallice’s (1986) Contention Scheduling Model (CSM). The CSM is a hierarchically organized network of action schemas and goals. Botvinick and Plaut (2004) instead took a connectionist approach to modeling routine procedural behavior. They argued in favor of holistic, distributed representation of learned step co-occurrence associations. Two experiments found that people can adapt routine procedural behavior to changing circumstances quite readily and that other factors besides statistical co-occurrence can have influence on action selection. A CSM-inspired ACT-R model of the two experiments is the first to postdict differential error rates across multiple between-subjects conditions and trial types. Results from the behavioral and modeling studies favor a CSM-like theory of human routine procedural memory that uses discrete, hierarchically-organized goal and action representations that are adaptable to new but similar procedures.