課題遂行対話における対話潤滑語の認定

概要

論文の詳細を見る
The Discourse Tagging Working Group has discussed and implemented several methods for tagging dialogue data over the past two years. In this report, our subgroup proposes a method for tagging 'lubricant' words, and show its application to discourse segmentation. We focus on utterances which help smooth the progress of conversation but do not themselves contribute to the objective of the dialogue. This class of utterances include discourse markers, fillers and acknowledgement tokens. Our aim is to formulate a means to tag such utterances. In order to clarify the classification of lubricant words, we first conducted a tagging experiment using the definition from the functional view. The results showed the diversity of the vocabularies identified as lubricant words, but nonetheless suggested the usefullness of lexical information regarding a limited subclass of the words. Thus, we can develop an automatic means to tag lubricant words based on their lexical information. Next, we applied our lubricant words tagging to discourse segmentation. We performed an experiment to check expected correlations between the discourse marker tags and the discourse segment tags. As a result, 64% of the discourse markers were indeed found to be located immediately after topic boundaries, compared to a chance level of 30%. The automatically extracted discourse markers are thus effective for the recognition of discourse segments.
社団法人人工知能学会の論文
1999-09-01

課題遂行対話における対話潤滑語の認定

スポンサーリンク

概要

著者

関連論文

スポンサーリンク