First International Conference on Spoken Language Processing (ICSLP 90)
This paper presents a unified approach to the description and classification of prosodic phenomena in continuous speech, and evaluates its applicability to interpreting telephony for a limited transfer task between equivalent samples of Japanese and English dialogue. An algorithm is proposed which uses the F0 tracings of connected speech dialogue as input and performs speaker independent segmentation into prosodically defined information units. The time- alignment of these units with linguistic structure is established separately for each language, which permits both monolingual classification and bilingual comparison of the prosodic data. A tentative set of transfer roles for the "translation" of prosodic features between Japanese and English is introduced, and directions for further research are indicated.
Bibliographic reference. Huber, Dieter (1990): "Prosodic transfer in spoken language interpretation", In ICSLP-1990, 509-512.