Summarizing Decisions in Spoken Meetings

Proceedings of the Workshop on Automatic Summarization for Different Genres, Media, and Languages, pages 16–24, Portland, Oregon, June 23, 2011. c⃝2011 Association for Computational Linguistics Summarizing Decisions in Spoken Meetings Lu Wang Department of Computer Science Cornell University Ithaca, NY 14853 luwang@cs.cornell.edu Claire Cardie Department of Computer Science Cornell University Ithaca, NY 14853 cardie@cs.cornell.edu Abstract This paper addresses the problem of summa- rizing decisions in spoken meetings: our goal is to produce a concise decision abstract for each meeting decision. We explore and com- pare token-level and dialogue act-level au- tomatic summarization methods using both unsupervised and supervised learning frame- works. In the supervised summarization set- ting, and given true clusterings of decision- related utterances, we find that token-level summaries that employ discourse context can approach an upper bound for decision ab- stracts derived directly from dialogue acts. In the unsupervised summarization setting,we find that summaries based on unsupervised partitioning of decision-related utterances per- form comparably to those based on partitions generated using supervised techniques (0.22 ROUGE-F1 using LDA-based topic models vs. 0.23 using SVMs). 1 Introduction Meetings are a common way for people to share in- formation and discuss problems. And an effective meeting always leads to concrete decisions. As a re- sult, it would be useful to develop automatic meth- ods that summarize not the entire meeting dialogue, but just the important decisions made. In particular, decision summaries would allow participants to re- view decisions from previous meetings as they pre- pare for an upcoming meeting. For those who did not participate in the earlier meetings, decision sum- maries might provide one type of efficient overview of the meeting contents. For managers, decision summaries could act as a concise record of the idea generation process. While there has been some previous work in summarizing meetings and conversations, very lit- tle work has focused on decision summarization: Fern´andez et al. (2008a) and Bui et al. (2009) in- vestigate the use of a semantic parser and machine learning methods for phrase- and token-level deci- sion summarization. We believe our work is the first to explore and compare token-level and dialogue act-level approaches — using both unsupervised and supervised learning methods — for summarizing de- cisions in meetings. C: Just spinning and not scrolling , I would say . (1) C: But if you’ve got a [disfmarker] if if you’ve got a flipped thing , effectively it’s something that’s curved on one side and flat on the other side , but you folded it in half . (2) D: the case would be rubber and the the buttons , (3) B: I think the spinning wheel is definitely very now . (1) B: and then make the colour of the main remote [vocal- sound] the colour like vegetable colours , do you know ? (4) B: I mean I suppose vegetable colours would be orange and green and some reds and um maybe purple (4) A: but since LCDs seems to be uh a definite yes , (1) A: Flat on …

Content truncated to 500 words