In a classic experiment, Saffran, Aslin, and Newport (1996)used a headturn preference procedure to show that infants candiscriminate between familiar syllable sequences (“words”)and new syllable sequences (“non-words” and “part-words”).While several computational models have simulated aspects oftheir data and proposed that the learning of transitional prob-abilities could be mediated by neural-net or chunking mech-anisms, none have simulated the absolute values of infants’listening times in the different experimental conditions. In thispaper, we used CHREST, a model based on chunking, to sim-ulate these listening times. The model simulated the fact thatinfants listened longer to novel words (non-words and part-words) than familiar words. While the times observed with themodel were longer than those observed with infants, we makea novel finding with regard to phonological store trace decay.We also propose how to modify CHREST to produce data thatfits closer to the human data.

Creating a plausible Unified Theory of Cognition (UTC) requires considerable effort from large, potentially distributed, teams. Computational Cognitive Architectures (CCAs) provide researchers with a concrete medium for connecting different cognitive theories to facilitate development of a robust, unambiguous UTC. However, due to wide dissemination of research effort, and broad scope of cognition as a psychological science, keeping track of CCA contributions is difficult. We compare the structuring of long-term memory (LTM) in two CCAs: ACT-R and CHREST. LTM structuring is considered in particular since it is an essential component of CCAs and underpins most of their operations. We aim to consolidate knowledge regarding LTM structuring for these CCA’s and identify similarities and differences between their approaches. We find that, whilst the architectures are similar in a number of ways, providing consensus for some concepts to be included in a UTC, their differences highlight important questions and development opportunities