Data-driven approaches to pitch modelling in two Mexican Spanish ethnolects: K-means Clustering & GAMMs

This study models F0 data for documentation purposes. Using K-means clustering and Generative Additive Mixed Models, it documents differences in F0 contours between two Mexican Spanish ethnolects: Afro-Mexican Spanish and Altiplateau Mexican Spanish. Both dialects distinguish focus condition through contour shape and pitch range: rising contours and high frequency ranges are associated with in-focus conditions, and lower frequency plateaus with post-focus. AMS however shows delayed peak alignment in open syllables and overall higher pitch contours for those with post-vocalic /s/. We posit that these fine-grained differences in segmental alignment and pitch range would not be captured through traditional methods, e.g., Sp_ToBI. With this study's focus on phonetic detail, the ToBI model risks both the absence of phonetic variation and the assumption of underlying phonological structure. We therefore argue that this variation is best captured through clustering and GAMMs.