Effects of word position on high vowel devoicing rates and related duration patterns in Tokyo Japanese

High vowels /i/ and /u/ are often devoiced between two voiceless obstruents in Tokyo Japanese. We examined the rate of devoicing across word positions, in either the first or second mora of the word. We also examined the durations of intervals crossing segments over a /C1VC2/ sequence, where both C1 and C2 were stops, such as a /C1V/ mora and the interval from C1 burst to the onset of C2 closure. Eight speakers of Tokyo Japanese produced 24 four-mora nonce words 15 times each. Items had the structure of /C1VC2ateko/ or /taC1VC2ateko/. Eight items contained an environment conducive to high vowel devoicing. The rate of devoicing was dependent on word position. It was nearly 100% in the first mora, but was much lower in the second mora, just above 50% with large cross-speaker variability. Regardless of the word position, devoicing had a consistent shortening effect on the cross-segmental intervals. Moreover, intervals with voiced high vowels in the devoicing environment were shorter than voiced vowels in other environments, indicating that the devoicing environment has an effect on timing even when devoicing fails.