The ISLA model of phonological encoding and word durations

Every spoken word is assigned a form through the process of phonological encoding; however, this process is not fully understood. This paper presents a theoretical, computational account of phonological encoding that is informed by empirical findings on word durations, which reflect many linguistic and cognitive factors and serve as a useful window into the phonological encoding process. We construct a connectionist model that accounts for two phenomena that have traditionally been considered separate: repetition reduction and phonological overlap lengthening. We propose a singular incremental learning mechanism to account for these phenomena through the updating of phonological sequences. We show that this mechanism correctly predicts shortening and lengthening of words as indexed by the difficulty the model faced in production of a phonological sequence in both a small-scale, toy-like lexicon, and in a larger, more realistic lexicon. We additionally discuss the implications of this theory for phonological encoding and propose areas for future directions.