Challenges to event formalization for information extraction

Abstract

Information Extraction (IE) is a vital technology for dealing with large volume of documents. IE extracts entities, links (relationships) and events of interest from text. Whi le much progress has occurred in recent years in Entity and Link Extraction, Event Extraction remains one of the weakest poi nts of IE. We hypothesize that one of the reasons is the fact that there is little understanding of, and agreement about, whatconstitutes an event. Moreover, in Intelligence and Counterte rrorism environments it is extremely difficult to describe all situations of interest, making monitoring for such situations quite chalenging. We propose a formal definition of event, developed within the framework of Situation Theory, a theory of information flow developed in logic and linguistics. Besides giving a solid y et intuitive foundation, the definition can be put to practical use. We develop a classification of event types on top of our definitio n to let a user (Intelligence Analyst or other) specify events of interest, and sketch an interpreter that can use Information Extraction tools to monitor a collection of documents in order to detect whether the specified events are taking place.