Abstract
Language systems seem to accumulate and implement at least two types of phonological knowledge for serial ordering. One is knowledge of inter-element associations (e.g., which phoneme tends to follow which) and the other is context–element associations (e.g., which phoneme tends to occupy which position). The present study examined whether and how such phonological knowledge is accumulated and implemented by a computational model for serial order: the simple recurrent network. The model simulated the pattern of behavioral data that demonstrated the contribution of phonological knowledge in a phonological short-term memory task. The model replicated the interaction between the two types of knowledge. It also replicated dissociable effects of inter-element associations, which were evident in later within-word positions, from those of context–element associations. The only point that required further modeling was the early-effectiveness of context–element associations.
