Assessing prosodic and text features for segmentation of Mandarin broadcast news

• Authors: Gina-Anne Levow
• Venue: SpeechIR '04 Proceedings of the Workshop on Interdisciplinary Approaches to Speech Indexing and Retrieval at HLT-NAACL 2004
• Year: 2004

Automatic topic segmentation, separation of a discourse stream into its constituent stories or topics, is a necessary preprocessing step for applications such as information retrieval, anaphora resolution, and summarization. While significant progress has been made in this area for text sources and for English audio sources, little work has been done in automatic segmentation of other languages using both text and acoustic information. In this paper, we focus on exploiting both textual and prosodic features for topic segmentation of Mandarin Chinese. As a tone language, Mandarin presents special challenges for applicability of intonation-based techniques, since the pitch contour is also used to establish lexical identity. However, intonational cues such as reduction in pitch and intensity at topic boundaries and increase in duration and pause still provide significant contrasts in Mandarin Chinese. We first build a decision tree classifier that based only on prosodic information achieves boundary classification accuracy of 89--95.8% on a large standard test set. We then contrast these results with a simple text similarity-based classification scheme. Finally we build a merged classifier, finding the best effectiveness for systems integrating text and prosodic cues.