Abstract: In a classic study of immediate serial recall, Baddeley (1968, QJEP) found no influence of intervening nonconfusable items on the magnitude of interference among confusable items. Henson et al. (1996, QJEP) replicated this result and interpreted it as being incompatible with all "chaining-based" models, including, in their view, recurrent neural networks. Instead, they adopted the increasingly common view that individual list items are transiently associated with content-independent representations of temporal context. However, such accounts have difficulty explaining well-established effects of background knowledge of temporal structure on ISR, such as an advantage recalling letter strings conforming to English bigram frequencies. We apply a recurrent network to ISR and demonstrate that it accounts not only for the bigram frequency effect, but also for Baddeley's and Henson and colleagues' findings. We conclude that simple recurrent networks provide a natural account of human short-term serial memory phenomena because, although they take advantage of temporal structure when it's available, they need not rely on such structure to encode temporal information in unstructured contexts.

Copyright Notice: The documents distributed here have been provided as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.