Plans Toward Automated Chat Summarization

Proceedings of the Workshop on Automatic Summarization for Different Genres, Media, and Languages, pages 1–7, Portland, Oregon, June 23, 2011. c⃝2011 Association for Computational Linguistics Plans Toward Automated Chat Summarization David C. Uthus NRC/NRL Postdoctoral Fellow Washington, DC 20375 david.uthus.ctr@nrl.navy.mil David W. Aha Naval Research Laboratory (Code 5514) Washington, DC 20375 david.aha@nrl.navy.mil Abstract We describe the beginning stages of our work on summarizing chat, which is motivated by our observations concerning the information overload of US Navy watchstanders. We de- scribe the challenges of summarizing chat and focus on two chat-specific types of summa- rizations we are interested in: thread sum- maries and temporal summaries. We then dis- cuss our plans for addressing these challenges and evaluation issues. 1 Introduction We are investigating methods to summarize real- time chat room messages to address a problem in the United States military: information overload and the need for automated techniques to analyze chat messages (Budlong et al., 2009). Chat has be- come a popular mode of communications in the mil- itary (Duffy, 2008; Eovito, 2006). On US Navy ships, watchstanders (i.e., personnel who continu- ously monitor and respond to situation updates dur- ing a ship’s operation, Stavridis and Girrier (2007)) are responsible for numerous duties including mon- itoring multiple chat rooms. When a watchstander reports to duty or returns from an interruption, they have to familiarize themselves with the current sit- uation, including what is taking place in the chat rooms. This is difficult with the multiple chat rooms opened simultaneously and new messages continu- ously arriving. Similarly, Boiney et al. (2008) ob- served that with US Air Force operators, when they returned to duty from an interruption, another oper- ator in the same room verbally updates them with a summary of what had recently taken place in the chat rooms and where they can find the important in- formation. Both of these situations are motivations for chat summarization, since watchstanders and operators could use automatically generated sum- maries to quickly orient themselves with the current situation. While our motivation is from a military perspec- tive, chat summarization is also applicable to other domains. For example, chat is used for communica- tion in multinational companies (Handel and Herb- sleb, 2002), open source meetings (Shihab et al., 2009; Zhou and Hovy, 2005), and distance learning (Osman and Herring, 2007). Summarization could aid people who missed meetings or students who wish to study past material in a summarized format. Even though chat summarization has many poten- tial uses, there has been little research on this topic (Section 3). One possible reason for this is that chat is a difficult medium to analyze: its characteristics make it difficult to apply traditional NLP techniques. It has uncommon features such as frequent use of ab- breviations, acronyms, deletion of subject pronouns, use of emoticons, abbreviation of nicknames, and stripping of vowels from words to reduce number of keystrokes (Werry, 1996). Chat is also characterized by conversation threads becoming entangled due to multiple conversations taking place simultaneously in …

Content truncated to 500 words