Processing missing vowels: Allophonic processing in Japanese

Abstract

The acoustic realisation of a speech sound varies, often showing allophonic variation triggered by surrounding sounds. Listeners recognise words and sounds well despite such variation, and even make use of allophonic variability in processing. This study reports five experiments on processing of the reduced/unreduced allophonic alternation of Japanese high vowels. The results show that listeners use phonological knowledge of their native language during phoneme processing and word recognition. However, interactions of the phonological and acoustic effects differ in these two processes. A facilitatory phonological effect and an inhibitory acoustic effect cancel one another out in phoneme processing; while in word recognition, the facilitatory phonological effect overrides the inhibitory acoustic effect. Four potential models of the processing of allophonic variation are discussed. The results can be accommodated in two of them, but require additional assumptions or modifications to the models, and primarily support lexical specification of allophonic variability.

Keywords:

• Allophone
• Fine phonetic detail
• Japanese
• Spoken word recognition
• Voiceless vowels

Acknowledgements

We are grateful to James McQueen, Holger Mitterer, Andy Wedel, Tim Vance, Adam Ussishkin, Merrill Garrett, Haruo Kubozono, Erin Good, and the reviewers for helpful suggestions on this material, and to the Cognitive Science Program and the Department of Linguistics at University of Arizona for support for travel to Japan. We would also like to thank Ikuo Hara, Akira Watanabe, Tomoko Yoshino, Miyuki Takasawa, Kyoko Masuda, Keiichi Tajima, Takayuki Arai, and Kazuki Kuwabara for their help in making the experiments in Japan possible. Any errors or misinterpretations are our own.

Notes

¹We will use the ‘(V)’ notation for devoiced/reduced vowels throughout.

²Thus we refer to the environment as a ‘devoicing’ or ‘voicing’ environment, but the vowel itself as ‘reduced’ or ‘unreduced’ in order to avoid confusion about whether we are referring to the environment or the vowel.

³The sequence [] in Japanese is considered to consist of /zi/ phonemically (Vance, 1987). However, we will write ‘’ even in phonemic transcriptions to avoid switching between ‘’ and ‘z’. The restriction of the voicing condition to this preceding consonant is discussed in Experiment 3, below. The phoneme /z/ did occur before other vowels in fillers and other targets, and only 30 out of the 120 tokens of /i/ a listener heard were in this condition and thus followed , so variability of phoneme sequences is maintained.

⁴Thus, extremely fast or slow RTs were included in the error rate for error analysis (as for example in Weber, 2001, and Cho, McQueen, and Cox, 2007). For all error analyses in Experiments 1, 3, 4, and 5, we also analysed the proportion of true errors (without outliers), and results were similar. Those cases with different statistical outcomes will be noted.

⁵When only true errors (without outliers) are analysed, this effect is significant only by subjects, F ₁(1, 45)=11.223, p<.01; F ₂(1, 19)=2.875, p>.1. The rate of true errors is rather low for this environment, so this change in significance may reflect a floor effect. All other statistical outcomes remain the same.