Peeling Back the Layers: Detecting Event Role Fillers in Secondary Contexts

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 1137–1147, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Peeling Back the Layers: Detecting Event Role Fillers in Secondary Contexts Ruihong Huang and Ellen Riloff School of Computing University of Utah Salt Lake City, UT 84112 {huangrh,riloff}@cs.utah.edu Abstract The goal of our research is to improve event extraction by learning to identify sec- ondary role filler contexts in the absence of event keywords. We propose a multi- layered event extraction architecture that pro- gressively “zooms in” on relevant informa- tion. Our extraction model includes a docu- ment genre classifier to recognize event nar- ratives, two types of sentence classifiers, and noun phrase classifiers to extract role fillers. These modules are organized as a pipeline to gradually zero in on event-related information. We present results on the MUC-4 event ex- traction data set and show that this model per- forms better than previous systems. 1 Introduction Event extraction is an information extraction (IE) task that involves identifying the role fillers for events in a particular domain. For example, the Message Understanding Conferences (MUCs) chal- lenged NLP researchers to create event extraction systems for domains such as terrorism (e.g., to iden- tify the perpetrators, victims, and targets of terrorism events) and management succession (e.g., to iden- tify the people and companies involved in corporate management changes). Most event extraction systems use either a learning-based classifier to label words as role fillers, or lexico-syntactic patterns to extract role fillers from pattern contexts. Both approaches, how- ever, generally tackle event recognition and role filler extraction at the same time. In other words, most event extraction systems primarily recognize contexts that explicitly refer to a relevant event. For example, a system that extracts information about murders will recognize expressions associated with murder (e.g., “killed”, “assassinated”, or “shot to death”) and extract role fillers from the surround- ing context. But many role fillers occur in contexts that do not explicitly mention the event, and those fillers are often overlooked. For example, the per- petrator of a murder may be mentioned in the con- text of an arrest, an eyewitness report, or specula- tion about possible suspects. Victims may be named in sentences that discuss the aftermath of the event, such as the identification of bodies, transportation of the injured to a hospital, or conclusions drawn from an investigation. We will refer to these types of sentences as “secondary contexts” because they are generally not part of the main event description. Dis- course analysis is one option to explicitly link these secondary contexts to the event, but discourse mod- elling is itself a difficult problem. The goal of our research is to improve event ex- traction by learning to identify secondary role filler contexts in the absence of event keywords. We cre- ate a set of classifiers to recognize role-specific con- texts that suggest the presence of a likely role filler regardless of whether a relevant event is mentioned or not. For …

Content truncated to 500 words