Proceedings of the Workshop on Automatic Summarization for Different Genres, Media, and Languages

Welcome to the ACL Workshop on Automatic Summarization for Different Genres, Media, and Languages! Our motivation for organizing this workshop has been the need many researchers in the field have seen, to come together to discuss various new issues that the field is facing as more and more summarization work is being conducted for domains beyond newswire and broadcast news. Extractive summarization of newswire text has dominated summarization research for over a decade. Large corpora of machine- and human-authored summaries have been collected and evaluation has been standardized to a large extent. As work on different genres, media and languages, such as voice mail, email, meetings, broadcast conversations, lectures, chat, blogs and scientific articles becomes more prominent, the need to precisely define tasks, to provide corpora to support comparison between approaches, and to identify desirable evaluation metrics is becoming increasingly urgent. We hope that this workshop will provide a valuable opportunity for all participants to present their work and to engage in discussion about the issues and problems they face, and how we can best support the changing nature of the field. We have an exciting mix of papers. Some introduce novel summarization tasks: abstractive summarization of line graphs from popular media, summarization of Wikipedia articles with increasing popularity, summarization of chat for the military. Others present new approaches to summarization tasks that have been gaining popularity in recent years, such as summarization of spoken meetings and scientific articles. We are fortunate to have a paper from the organizers of the Text Analysis Conferences (TAC). This paper presents an in-depth analysis of the newest task adopted for the evaluation that, while still based on news, promotes the use of abstractive approaches and makes it possible to track the types of information people consider important and summary-worthy. Finally, we have a reminder that continuity in research focus truly helps to understand a domain and sharpen our understanding of prior approaches to the task; we will hear about some significant improvements in the topic model approach that has proven to be so successful for multi-document summarization of news.