Keio Research Highlights

Yukiko Sugiyama's interest in Japanese pitch accent began during her graduate studies in the US. "Spoken forms of languages can be roughly divided into either stress accent languages like English, or tone based languages like Chinese," explains Sugiyama. "But how do we classify pitch accent languages like Japanese? My goal is to answer this question."

Japanese employs the so-called fundamental frequency (F0) as the main cue to differentiate words that contrast only in accent. In the absence of F0, which is roughly analogous with the speech produced in whisper, listeners can use other acoustic information to infer the meaning of words. Is this true also for Japanese?

"Reports based on syllable duration and intensity ― acoustics of speech ― do not show if listeners can use them to distinguish words that differ only in accent," says Sugiyama. "My approach is completely different. I am focusing on the perception of speech instead of its production."

Now, writing in Phonetica, Sugiyama describes the use of acoustic stimuli obtained by removing the periodicity of F0 from natural Japanese speech and substituting it with white noise, while keeping all the other acoustic properties intact, to see if a group of native Tokyo Japanese speakers would still be able to identify words.

Sugiyama selected 14 two-syllable words that differ only in respect to the presence or absence of an accent on the last syllable. Out of all the disyllabic words that differ only in accent, she selected those that cannot be produced with more than one accent type, and then restricted the set to words that would be likely familiar to participants, and excluded words that were typically used in compounds. Because the onset and offset of an utterance exhibit acoustic properties specific to their locations in a sentence, the words were embedded in the middle of a sentence equivalent to the English 'he wants...' or 'he has a good...'.

Each listener was exposed to blocks of both natural and edited speech; the accuracy for edited speech was about 65%, compared to a little over 95% for natural speech. The fact that listeners' responses exceeded chance level indicates the presence of other cues for accent besides F0.

Analysis of the duration and amplitude of test words showed no significant differences in syllable duration between accented and unaccented words; a reliable difference in amplitude was, however, observed. The implication of this research is that amplitude acts as a secondary cue for Japanese accent, whereas duration does not play an important role.

"For practical purposes, I hope my research can help in the development of high performance hearing aids and cochlear implants."

1. Yukiko Sugiyama, Perception of Japanese Pitch Accent without F0.Phonetica 74, 107-123 (2017). | article