Syllable frequency effects in a context-sensitive segment production model

In this study we describe a new model of how phonetic knowledge guides speech production. In the Context Sequence model, target acoustic patterns are determined based on selection of previously heard or produced sounds from a memory store. Since signals in the memory correspond to long stretches of continuous speech, individual speech sounds always appear in a larger context. A key property of the model is that the selection of exemplars for production is weighted by the similarity of the contexts in which they originally occurred to the current production context. In two simulations based on realistic amplitude envelope data extracted from a large single-speaker production corpus, we demonstrate that (1) optimal selection of context-appropriate segment-level exemplars requires consideration of about 0.5 s of context material preceding and following exemplars and (2) context-dependent production at this low level may be responsible for a range of frequency effects that have previously been assumed to involve word, syllable, and other higher levels of organization.