HOW DO SPEAKERS PAUSE AND HESITATE IN ENGLISH AND JAPANESE? - A COMPARISON USING PARALLEL CORPORA OF ENGLISH AND JAPANESE PRESENTATION SPEECHES

We built an English presentation speech corpus, COPE, to enable contrastive studies of speech disfluencies, particularly fillers, between English and Japanese. Recording settings and linguistic labels are similar to those of informal presentation speeches in CSJ, a huge Japanese spontaneous speech corpus. We introduce COPE in the first section, followed by contrastive studies on silent and filled pauses using COPE and CSJ. The first study revealed larger silent pause duration ratio and higher filler frequency per word in Japanese than in English. The second study indicates that silent pause durations at sentence and clause boundaries are relevant to the immediately following clause initial filler probabilities in both languages. More clauses are preceded by clause-initial fillers after longer silent pauses. The results support the hypothesis that fillers are used to fill uncomfortably long silence.