Disentangling sociophonetic and physiological variation in /s/ acoustics across 12 languages

Sibilants, especially /s/, display gendered variation across languages. From acoustic data alone, it can be hard to determine whether this variation is due to performance of social identity or to physiology, e.g. vocal tract length (VTL) variation. In causal terms, there is a direct effect of gender on sibilant acoustics and an indirect effect mediated by VTL, which must be teased apart. To this end, a speaker's VTL can be estimated from their vowel formants, though the consistency of this measure's relationship to sibilant acoustics is largely untested. Using a new multilingual phonetic database of 12 languages (> 1,300 speakers) and the tools of causal mediation analysis, we investigate the relationship between VTL and /s/ peak frequency and its causal role in gender effects. We find that estimated VTL predicts peak well across languages. While there are gender differences in peak across nearly all languages, the relative contributions of physiology and performance vary considerably.