UCLA

Previous descriptions of Malagasy (Austronesian, Madagascar) describe a vowel devoicing process affecting unstressed high vowels, similar to other languages said to have devoiced vowels (e.g., Japanese, Korean, Turkish). However, a closer inspection of the data reveals that most Malagasy vowels are not truly devoiced; rather, they appear to be nearly deleted, only measurable as coarticulatory residue on adjacent segments. In this dissertation, I present a phonetic and phonological study of this “devoicing” process in Malagasy, arguing that what is heard as vowel devoicing is actually complete gestural overlap of the vowel by the consonants that surround it. The dissertation includes two phonetic studies of vowels said to be devoiced. In the first, I show that Malagasy “vowel devoicing” is, in fact, two distinct processes: in Intonational Phrase-final positions, vowels may be truly devoiced; in non-final positions, however, unstressed vowels seem to disappear altogether, with no segmental interval clearly identified as the vowel. In the second study, I look closely at the acoustics of these non-final “devoiced” vowels, finding that traces of the underlying vowel are present in the acoustic output, even when these vowels appear to be deleted. For example, measures of spectral peak (i.e., the most amplified frequency) of /s/ reveal a lower spectral peak before underlying /u/ compared to /i/, consistent with the effects of typical consonant-vowel coarticulation. This finding suggests that non-final vowels are neither truly devoiced nor deleted in Malagasy; rather, they appear to still be present as an articulatory gesture that is overlapped and obscured by adjacent consonantal gestures. I model this apparent gestural overlap using a theory of Articulatory Phonology, in which speech sounds are represented as “constellations” of gestures that may overlap each other in time: when a vowel’s gestures overlap with adjacent consonants sufficiently, the vowel may be completely obscured, heard only as coarticulation on those consonants. A complete gestural analysis of Malagasy vowel devoicing is presented, accounting not only for the phonetic realization described but also the environment where “devoicing” is most likely to occur: high vowels devoice more than low vowels, unstressed vowels more than stressed vowels and so on. A surprising finding, confirmed quantitatively, is that the likelihood of a vowel undergoing “devoicing” is directly tied to the sonority of the consonants that surround it: vowels are more likely to “devoice” when they follow a sonorant but precede an obstruent. For this, I invoke a theory of Sonority-Driven Gestural Timing (Gu & Durvasula, 2024), arguing that the degree of overlap between two gestures is modulated by the difference in sonority between them. This work has implications for theories of phonological representation and gestural timing.