Modeling a recognition memory task to investigate differences in working memory

Abstract

Investigating the sources of individual differences in human working memory, Oztekin and McElree (2010) used subject performance in a recognition memory task to argue that the interplay between item familiarity information and episodic retrieval information marked an important difference between individuals with high and low working memory spans. However, they were unable to adjudicate between several hypotheses about the precise sources of those differences. We developed a computational cognitive model using the ACT-R architecture that performed a replication of recognition memory task, and attempted to fit model performance to Oztekin and McElree's human data through parameter changes to the model and the addition of a new recognition memory module. The model was able to produce performance that approximated that of the high span subjects, but had limited success matching low span subject performance. Importantly, parameter changes that produced slower memory retrievals, which was among Oztekin and McElree's hypotheses, were insufficient to produce model performance matching that of low span subjects. This constitutes evidence that a primary difference between high and low span subjects lies in the conflict resolution between familiarity and episodic recall, as opposed to differences in episodic recall initiation or completion time.