Effects of predictability on vowel reduction

Abstract

High vowel devoicing in Japanese, where unaccented /i, u/ in a C1VC2 sequence reduce when both C1 and C2 are voiceless, has been studied extensively, but whether the target vowel is truly devoiced (oral gesture is maintained) or deleted completely (oral gesture and voicing are both lost) is still debated. This study examines the effects of vowel predictability on the degree of vowel reduction. Native Japanese speakers (N = 8) were recorded in a sound-proof booth reading sentences containing lexical stimuli. C1 of the stimuli were [k, ∫], after which either high vowel can occur, and [ϕ, s, ç], after which only one of the two is possible. Half of the stimuli contained a devoicing environment with a voiceless C2. Center of gravity, the amplitude weighted mean of frequencies present in a signal, was measured for the first half (COG1) and the second half (COG2) of C1. Results show that COG2 is significantly lower than COG1 for all consonants when a full vowel follows, as well as for [k, ∫] in devoicing stimuli. In contrast, COG remained stable for [ϕ, s, ç] in devoicing environments, suggesting a complete lack of vowel gestures. Predictable vowels, therefore, seem to delete, while unpredictable vowels devoice.