Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

An Annotation Scheme of Spoken Dialogues With Topic Break Indexes

Yoichi Yamashita, Michiyo Murai

Dep. of Computer Science, Ritsumeikan University, Kusatsu-shi, Shiga, Japan

This paper proposes a scheme of annotating spoken dialogues with discourse level information in terms of the discourse segment. Dialogues are coded with topic break index (TBI), which indicates the degree of topic break between the discourse segments, instead of marking a beginning and an ending utterances of the segment. TBI is graded by two levels, 1 and 2, and TBI=2 indicates a large change of the topic. Two methods are tried for assigning a TBI value for segment boundaries. In the method-I, the coder directly describes TBI according to the di erence of contents between the adjacent segments. In the method-II, the coder classifies relative change of the topic break between the adjacent segments into three categories. Then, the relative changes are automatically converted into TBIs by extraction of local maximum change of the topic break. Two annotation methods are evaluated with the agreement score and the relation to prosodic parameters.

Full Paper

Bibliographic reference.  Yamashita, Yoichi / Murai, Michiyo (2000): "An annotation scheme of spoken dialogues with topic break indexes", In ICSLP-2000, vol.1, 569-572.