Lexical effects in Japanese vowel reduction

Thesis (Ph. D.)--University of Washington, 2005.This dissertation reports the results of a study of a vowel reduction in Japanese. Vowel reduction (durational and spectral) has been observed in many languages; however, only a few studies have been conducted research on Japanese phonetic vowel reduction. Furthermore, in these studies, factors influencing duration and formants were not controlled. I investigated vowel reductions in Japanese, while controlling factors influencing duration and formants. In my study, vowels in the functional particles /ga/, /de/, and /to/ were compared with word-final vowels in lexical words.The overall results show that there was a significant effect for LEXICAL on DURATION and Fl values but not on F2 values. When I looked at the lexical effect for individual vowels, vowels [a] and [o] showed lexical effect on DURATION but [e] did not. I sought the reasons for no lexical effect on DURATION of [e] and concluded that lexical effects on [e] were not observed because of phrase final lengthening.For formants, only [a] showed a significant lexical effect on F1 and F2 values but [e] and [o] did not. The lack of observation of a lexical effect was probably due to the short locus distance between alveolar consonants, [d] and [t], and vowels, [e] and [o], respectively. Furthermore, I investigated the possible coarticulation effects of consonants on vowels and V-to-V coarticulation effects.Durational and spectral vowel reductions for [a] were observed, which raised a question: whether duration is the only factor that causes spectral vowel reduction. In the undershoot hypothesis, the short duration of reduced vowels leads to the spectral vowel reduction; in other words, duration is the only factor that causes spectral vowel reduction. I investigated whether or not there was a lexical effect in addition to a durational effect. The results indicate that there was a significant effect for LEXICAL on F1 values. Function particles have a low information value; consequently speakers do not articulate these function particles clearly. In my hypothesis, there are two targets (content and function) instead of one target