Modelling the intensity difference of Spanish alveolar taps with finite mixture models

The production of stops has been documented to vary considerably in several languages. The difference between the minimum intensity during the consonant and the maximum intensity of the surrounding vowels is an acoustic correlate of taps, whose realizations include plosives, approximants, and deletions. We investigated how lexical, phonetic, and predictability-related factors are associated with changes in the intensity difference of Spanish taps. We conducted an acoustic analysis using a force-aligned corpus of conversational Spanish, with ten percent of tokens hand-corrected to evaluate performance. Model checking with generalized linear models indicated that one distribution could not adequately account for the observed data. As such, we analyzed variation in intensity difference using finite mixture models comprising two skew-normal distributions, which provided a substantially better fit. The results of our modelling approach require a more nuanced interpretation than standard linear models. Our model indicates that Spanish tap production is a complex system where some variables, like frequency, are related to categorical shifts between two potential realizations, and other variables are related to gradient intensity changes within each potential realization of the tap. We also find that articulatory factors like speech rate are responsible for larger intensity changes than lexical properties like frequency.