<i>Time</i> and <i>thyme</i> again: Connecting English spoken word duration to models of the mental lexicon

Effects of word frequency on spoken word duration are well documented and have long informed theories of the mental lexicon. In this study, we discuss the two theoretical constructs, ‘frequency’ and ‘word’, that are implicated by the notion of lexical frequency, in light of recent models of the lexicon that do not contain stable, discrete lexical representations, and in which lexical frequency therefore has no place. We compare two approaches (localist spreading-activation vs. discriminative learning (DL) models integrating distributional semantics) by assessing regression models of spoken word duration of English homophones grounded in each. We further show that the relationship between a homophone's form and its semantics is predictive of its duration, consistent with predictions of the DL-based model.