Identification of Event Mentions and their Semantic Class

Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP 2006), pages 146–154, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Identification of Event Mentions and their Semantic Class Steven Bethard Department of Computer Science University of Colorado at Boulder 430 UCB, Boulder, CO 80309, USA steven.bethard@colorado.edu James H. Martin Department of Computer Science University of Colorado at Boulder 430 UCB, Boulder, CO 80309, USA james.martin@colorado.edu Abstract Complex tasks like question answering need to be able to identify events in text and the relations among those events. We show that this event identification task and a related task, identifying the seman- tic class of these events, can both be for- mulated as classification problems in a word-chunking paradigm. We introduce a variety of linguistically motivated fea- tures for this task and then train a system that is able to identify events with a pre- cision of 82% and a recall of 71%. We then show a variety of analyses of this model, and their implications for the event identification task. 1 Introduction Research in question answering, machine transla- tion and other fields has shown that being able to recognize the important entities in a text is often a critical component of these systems. Such en- tity information gives the machine access to a deeper level of semantics than words alone can provide, and thus offers advantages for these complex tasks. Of course, texts are composed of much more than just sets of entities, and archi- tectures that rely solely on word and entity-based techniques are likely to have difficulty with tasks that depend more heavily on event and temporal relations. Consider a question answering system that receives the following questions: • Is Anwar al-Sadat still the president of Egypt? • How did the linking of the Argentinean peso to the US dollar in 1991 contribute to economic crisis of Argentina in 2003? Processing such questions requires not only knowing what the important people, places and other entities are, but also what kind of events they are involved in, the roles they play in those events, and the relations among those events. Thus, we suggest that identifying such events in a text should play an important role in systems that attempt to address questions like these. Of course, to identify events in texts, we must define what exactly it is we mean by “event”. In this work, we adopt a traditional linguistic defini- tion of an event that divides words into two as- pectual types: states and events. States describe situations that are static or unchanging for their duration, while events describe situations that involve some internal structure. For example, predicates like know and love would be states because if we know (or love) someone for a pe- riod of time, we know (or love) that person at each point during the period. Predicates like run or deliver a sermon would be events because they are built of smaller dissimilar components: run includes raising and lowering of legs …

Content truncated to 500 words