5. 1. はじめに
対話データ処理への関心
● SIGDIAL2015, Special Session, Multiing 2015
○ オンラインフォーラム、コールセンターの要約
● ACL2016, 3rd Workshp on Argument Mining
○ 議論データに関するWorkshop
○ 要約に関して2件の発表 [1, 2]
● COLING2016, Invited talk4
○ A Look at Computational Argumentation and Summarisation from a Text-Understanding Perspective
[1] Barker et al., "Summarizing Multi-Party Argumentative Conversations in Reader Comment on News", in Proc of 3rd Workshop on Argument Mining, 2016.
[2] Egan et al., Summarizng the points made in online politial deates", in Proc of 3rd Workshop on Argument Mining, 2016.
23. 3. 対話要約特有の問題
対話要約における3つの問題 [17]
1. 自動音声認識 (ASR) 誤りの問題
○ 音声認識のエラー
1. Disfluencyの問題
○ Filled Pauses (遊び言葉: uh, um, well ...)
○ Repetisions (繰り返し)
1. 抽出単位の問題
○ 質問と回答の一貫性
[17] Nenkova et al., "Automatic Summarization", Foundations and Trends in Information Retrieval, Vol 5, No 2-3, pp. 103-233, 2011.
24. 3. 対話要約特有の問題
1. 自動音声認識 (ASR) 誤りの問題
○ 10% - 40%程度の音声認識誤り [18]
○ AMI Meeting Corpus [19] の例
■ 人手書き起こし
"You look quite funny at the moment, Tim."
■ ASRの結果
"Great can implement that I"
[18] Glass et al., "Recent progress in the MIT spoken lecture processing project", in Proceedings of the Annual Conference of the International Speech Communication
Association, pp. 2553–2556, 2007.
[19] http://groups.inf.ed.ac.uk/ami/corpus/
25. 3. 対話要約特有の問題
2. Disfluencyの問題 (Filled Pauses and Repetitions)
○ Filled Pauses: uh, um, well ... などの遊び言葉
○ Repetitions: 同じ言葉が繰り返される
○ 全体の15 - 25%程度存在する [20]
○ 具体例
A: well I um I think we should discuss this you know with
her.
A’: I think we should discuss this with her.
[20] Zechner et al., "Summarization of spoken language - challenges, methods, and prospects,” Speech Technology Expert eZine, 2002.
26. 3. 対話要約特有の問題
2. Disfluencyの問題 (Filled Pauses and Repetitions)
○ Disfluencyの削除だけで1つの研究分野 [21-23]
○ 特徴量の1つとして利用 [24]
■ Disfluencyが存在する文は重要であるという仮説
■ ROUGE値の向上は1%未満
[21] Johnson et al., “A TAG-based noisy-channel model of speech repairs,” ACL, 2004.
[22] Miller et al., “A syntactic time-series model for parsing fluent and disfluent speech,” in Proceedings of the International Conference on Computational Linguistic,
pp. 569–576, 2008.
[23] Stolcke et al., “Statistical language modeling for speech disfluencies,” in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal
Processing, pp. 405–408, 1996.
[24] Zhu et al., “Summarization of spontaneous conversations,” in Proceedings of the Annual Conference of the International Speech Communication Association,
pp. 1531–1534, 2006.