An investigation of the dynamics of vowel nasalization in Arabana using machine learning of acoustic features

This paper presents exploratory research on temporally dynamic patterns of vowel nasalization from two speakers of Arabana. To derive a dynamic measure of nasality, we use gradient tree boosting algorithms to statistically learn the mapping between acoustics and vowel nasality in a speaker-specific manner. Three primary findings emerge: (1) NVN contexts exhibit nasalization throughout the entirety of the vowel interval, and we propose that a similar co-articulatory realization previously acted to resist diachronic change in this environment; (2) anticipatory vowel nasalization is nearly as extensive as carryover vowel nasalization, which is contrary to previous claims; and (3) the degree of vowel nasalization in word-initial contexts is relatively high, even in the #_C environment, suggesting that the ongoing sound change *#ŋa > #a has involved the loss of the oral constriction associated with ŋ but not the complete loss of the velum gesture.