Is vowel devoicing in Japanese lexically specified?—An eye-tracking study

Abstract

In Tokyo Japanese, high vowels can be devoiced typically between voiceless consonants. Traditionally, devoiced vowels are treated as allophones of corresponding vowels in which the devoicing process is a result of phonetic implementation commonly affected by speech rate. An opposing view is that devoicing is lexically specified since there are cases of interactions between devoicing and compounding that can be naturally accounted for by lexical phonology. To investigate this issue, the present study conducted an eye-tracking experiment. Edited mismatch stimuli where vowels were voiced before a voiceless consonant (M1) or devoiced before a voiced consonant (M2) were auditorily presented with pictures of the target word, a competing word, and a distractor presented visually. This design is based on a prediction that M1 and M2 invoke confusion in the recognition process if devoicing is specified lexically. Sixteen participants were recruited for an experiment of about 200 trials. Results showed that only M1 stimuli invoked a different pattern of fixation rates than M2 and other natural stimuli. This can be interpreted as partial support for the traditional view of devoicing. The asymmetry between M1 and M2 calls for an account based on markedness theory which is independent of the lexical specification.