Semi-supervised Speech Act Recognition in Emails and Forums

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1250–1259, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Semi-supervised Speech Act Recognition in Emails and Forums Minwoo Jeong†∗ Chin-Yew Lin‡ Gary Geunbae Lee† †Pohang University of Science & Technology, Pohang, Korea ‡Microsoft Research Asia, Beijing, China †{stardust,gblee}@postech.ac.kr ‡cyl@microsoft.com Abstract In this paper, we present a semi-supervised method for automatic speech act recogni- tion in email and forums. The major chal- lenge of this task is due to lack of labeled data in these two genres. Our method leverages labeled data in the Switchboard- DAMSL and the Meeting Recorder Dia- log Act database and applies simple do- main adaptation techniques over a large amount of unlabeled email and forum data to address this problem. Our method uses automatically extracted features such as phrases and dependency trees, called sub- tree features, for semi-supervised learn- ing. Empirical results demonstrate that our model is effective in email and forum speech act recognition. 1 Introduction Email and online forums are important social me- dia. For example, thousands of emails and posts are created daily in online communities, e.g., Usenet newsgroups or the TripAdvisor travel fo- rum1, in which users interact with each other us- ing emails/posts in complicated ways in discus- sion threads. To uncover the rich interactions in these email exchanges and forum discussions, we propose to apply speech act recognition to email and forum threads. Despite extensive studies of speech act recogni- tion in many areas, developing speech act recogni- tion for online forms of conversation is very chal- lenging. A major challenge is that emails and forums usually have no labeled data for training statistical speech act recognizers. Fortunately, la- beled speech act data are available in other do- mains (i.e., telephone and meeting conversations ∗This work was conducted during the author’s internship at Microsoft Research Asia. 1http://tripadvisor.com/ in this paper) and large unlabeled data sets can be collected from the Web. Thus, we focus on the problem of how to accurately recognize speech acts in emails and forums by making maximum use of data from existing resources. Recently, there are increasing interests in speech act recognition of online text-based con- versations. Analysis of speech acts for online chat and instant messages and have been studied in computer-mediated communication (CMC) and distance learning (Twitchell et al., 2004; Nastri et al., 2006; Ros´e et al., 2008). In natural language processing, Cohen et al. (2004) and Feng et al. (2006) used speech acts to capture the intentional focus of emails and discussion boards. However, they assume that enough labeled data are available for developing speech act recognition models. A main contribution of this paper is that we ad- dress the problem of learning speech act recog- nition in a semi-supervised way. To our knowl- edge, this is the first use of semi-supervised speech act recognition in emails and online forums. To do this, we make use of labeled data from spo- ken conversations (Jurafsky et al., 1997; Dhillon et al., …

Content truncated to 500 words