Multi-faceted Event Recognition with Bootstrapped Dictionaries

Proceedings of NAACL-HLT 2013, pages 41–51, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics Multi-faceted Event Recognition with Bootstrapped Dictionaries Ruihong Huang and Ellen Riloff School of Computing University of Utah Salt Lake City, UT 84112 {huangrh, riloff}@cs.utah.edu Abstract Identifying documents that describe a specific type of event is challenging due to the high complexity and variety of event descriptions. We propose a multi-faceted event recognition approach, which identifies documents about an event using event phrases as well as defin- ing characteristics of the event. Our research focuses on civil unrest events and learns civil unrest expressions as well as phrases cor- responding to potential agents and reasons for civil unrest. We present a bootstrapping algorithm that automatically acquires event phrases, agent terms, and purpose (reason) phrases from unannotated texts. We use the bootstrapped dictionaries to identify civil un- rest documents and show that multi-faceted event recognition can yield high accuracy. 1 Introduction Many people are interested in following news re- ports about events. Government agencies are keenly interested in news about civil unrest, acts of terror- ism, and disease outbreaks. Companies want to stay on top of news about corporate acquisitions, high- level management changes, and new joint ventures. The general public is interested in articles about crime, natural disasters, and plane crashes. We will refer to the task of identifying documents that de- scribe a specific type of event as event recognition. It is tempting to assume that event keywords are sufficient to identify documents that discuss in- stances of an event. But event words are rarely reli- able on their own. For example, consider the chal- lenge of finding documents about civil unrest. The words “strike”, “rally”, and “riot” refer to com- mon types of civil unrest, but they frequently refer to other things as well. A strike can refer to a military event or a sporting event (e.g., “air strike”, “bowl- ing strike”), a rally can be a race or a spirited ex- change (e.g.,“car rally”, “tennis rally”), and a riot can refer to something funny (e.g., “she’s a riot”). Event keywords also appear in general discussions that do not mention a specific event (e.g., “37 states prohibit teacher strikes” or “The fine for inciting a riot is $1,000”). Furthermore, many relevant docu- ments are not easy to recognize because events can be described with complex expressions that do not include event keywords. For example, “took to the streets”, “walked off their jobs” and “stormed par- liament” often describe civil unrest. The goal of our research is to recognize event de- scriptions in text by identifying event expressions as well as defining characteristics of the event. We pro- pose that agents and purpose are characteristics of an event that are essential to distinguish one type of event from another. The agent responsible for an ac- tion often determines how we categorize the action. For example, natural disasters, military operations, and terrorist attacks can all produce human casual- ties and physical destruction. But the agent …

Content truncated to 500 words