Lexically dependent estimation of acoustic information in speech II: Minimal pair confusability

We aim to develop a framework for the analysis of phonetic contrast systems that is fundamentally lexical and does not depend on assumptions of inventory homogeneity and independence of distribution in words and higher-order systems. Previously (Redmon and Jongman, 2018, JASA) we reported results of an open-class identification experiment on a 240-word sample of the 26,793-word single-speaker database in Tucker et al. (2018). Here, we present results of the second experiment in the project: a 2AFC task where the choice set is limited to obstruent-contrastive minimal pairs. This task forms the opposite end of a continuum from least restricted utilization of acoustic or higher-order information (Exp. 1), to localized attention to a particular contrast in the lexicon. Just as the first experiment provided estimates of a lower bound on listeners’ sensitivity to different cues in the signal, the results of this experiment provide an upper bound on those estimates. Participants were presented with 480 stimuli balanced between contrastive obstruents in #CV, VCV, and VC# positions. The results were then used to determine network edge weights on a phonological lexicon on the model of Vitevitch (2008), which emphasizes the interaction between acoustic features, neighborhood topology, and higher-order information in the lexicon.