Ellen Riloff

University of Utah

 

Abstract
Extracting information about events from text poses several challenges for NLP systems. Finding documents about a specific type of event is difficult because of the wide variety of event phrases and because context often determines the nature of an event. Finding the entities and objects that play important roles in an event is also challenging due to the complexity of narrative texts and discourse phenomena. I will present recent work at the University of Utah on event recognition and event extraction. We have developed a multi-faceted approach to event recognition that identifies documents about a specific type of event by searching for event phrases as well as defining characteristics (facets) of the event type. We use a novel bootstrapping algorithm to automatically learn the necessary dictionaries of event phrases, agent terms, and purpose (reason) phrases for civil unrest events. Experimental results show that multi-faceted event recognition with these bootstrapped dictionaries yields high accuracy. I will also present a new bottom-up architecture for extracting role filler information from event descriptions. Our extraction model includes a structured sentence classifier to identify event-related contexts based on lexical associations, discourse relations, and role filler distributions within and across sentences. This approach yields state-of-the-art performance on the MUC-4 terrorism data set, achieving substantially higher precision than previous systems.

Bio:
Ellen Riloff is an Associate Professor of Computer Science in the School of Computing at the University of Utah. Her primary research areas are information extraction, semantics, sentiment analysis, and coreference resolution. A major emphasis of her research has been automatically acquiring the knowledge needed for natural language processing using bootstrapping methods that learn from unannotated texts. She has served on the NAACL Executive Board, Human Language Technology (HLT) Advisory Board, Computational Linguistics Editorial Board, Transactions of the Association for Computational Linguistics (TACL) Editorial Board, and as Program Co-Chair for the NAACL HLT 2012 and CoNLL 2004 conferences.