Improving Event Detection with Dependency Regularization

Proceedings of Recent Advances in Natural Language Processing, pages 78–83, Hissar, Bulgaria, Sep 7–9 2015. Improving Event Detection with Dependency Regularization Kai Cao Xiang Li Ralph Grishman Computer Science Department New York University 719 Broadway, New York, NY, 10003 {kcao, xiangli, grishman}@cs.nyu.edu Abstract Event Detection (ED) is an Information Extraction task which involves identifying instances of specified types of events in text. Most recent research on Event De- tection relies on pattern-based or feature- based approaches, trained on annotated corpora, to recognize combinations of event triggers, arguments, and other con- textual information. These combinations may each appear in a variety of linguis- tic forms. Not all of these event expres- sions will have appeared in the training data, thus adversely affecting ED perfor- mance. In this paper, we demonstrate the effectiveness of Dependency Regular- ization techniques to generalize the pat- terns extracted from the training data to boost ED performance. The experimen- tal results on the ACE 2005 corpus show that our pattern-based system with the ex- panded patterns can achieve 70.49% (with 2.57% absolute improvement) F-measure over the baseline, which advances the state-of-the-art for such systems. 1 Introduction Event Detection (ED) involves identifying in- stances of specified types of events in text, which is an important but difficult Information Extraction (IE) task. Associated with each event mention is a phrase, the event trigger (most often a single verb or nominalization), which evokes that event. More precisely, our task involves identifying event trig- gers and classifying them into specific types. For instance, according to the ACE 2005 annotation guidelines1, in the sentence “She was killed in an 1https://www.ldc.upenn.edu/ sites/www.ldc.upenn.edu/files/ english-events-guidelines-v5.4.3.pdf automobile accident yesterday”, an event detec- tion system should be able to recognize the word “killed” as a trigger for the event DIE. This task is quite challenging, as the same event might ap- pear in the form of various trigger expressions and an expression might represent different events in different contexts. ED is a crucial component in the overall Event Extraction task, which also re- quires event argument identification and argument role labeling. Most recent research on Automatic Content Extraction (ACE) Event Detection task relies on pattern-based or feature-based approaches to building classifiers for event trigger labeling. Al- though the training corpus is quite large (300,000 words), the test data will inevitably contain some event expressions that never occur in the training data. To address this problem, we propose sev- eral Dependency Regularization methods to help generalize the syntactic patterns extracted from the training data in order to boost ED performance. Among the syntactic representations, dependency relations serve as important features or part of a pattern-based framework in IE systems, and play a significant role in IE approaches. These proposed regularization rules will be applied either to the de- pendency parse outputs of the candidate sentences or to the patterns themselves to facilitate detect- ing the event instances. The experimental results demonstrate that our pattern-based system with the expanded patterns can achieve 70.49% (with 2.57% absolute improvement) F-measure over …

Content truncated to 500 words