Improving Event Coreference Resolution by Learning Argument Compatibility from Unlabeled Data

Proceedings of NAACL-HLT 2019, pages 785–795 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 785 Improving Event Coreference Resolution by Learning Argument Compatibility from Unlabeled Data Yin Jou Huang1, Jing Lu2, Sadao Kurohashi1, Vincent Ng2 1Graduate School of Informatics, Kyoto University 2Human Language Technology Research Institute, University of Texas at Dallas huang@nlp.ist.i.kyoto-u.ac.jp, kuro@i.kyoto-u.ac.jp {ljwinnie,vince}@hlt.utdallas.edu Abstract Argument compatibility is a linguistic condi- tion that is frequently incorporated into mod- ern event coreference resolution systems. If two event mentions have incompatible argu- ments in any of the argument roles, they can- not be coreferent. On the other hand, if these mentions have compatible arguments, then this may be used as information toward deciding their coreferent status. One of the key chal- lenges in leveraging argument compatibility lies in the paucity of labeled data. In this work, we propose a transfer learning framework for event coreference resolution that utilizes a large amount of unlabeled data to learn the ar- gument compatibility between two event men- tions. In addition, we adopt an interactive in- ference network based model to better capture the (in)compatible relations between the con- text words of two event mentions. Our exper- iments on the KBP 2017 English dataset con- firm the effectiveness of our model in learn- ing argument compatibility, which in turn im- proves the performance of the overall event coreference model. 1 Introduction Events are essential building blocks of all kinds of natural language text. An event can be described several times from different aspects in the same document, resulting in multiple surface forms of event mentions. The goal of event coreference resolution is to identify event mentions that cor- respond to the same real-world event. This task is critical for natural language processing applica- tions that require deep text understanding, such as storyline extraction/generation, text summariza- tion, question answering, and information extrac- tion. Figure 1 shows a document consisting of three events described by six different event mentions. Among these event mentions, m1, m2 and m4 are Figure 1: A document with three events described in six event mentions. Coreferent event mentions are highlighted with the same color. coreferent, since they all correspond to the event of the KMT party electing a new party chief. Sim- ilarly, m3 and m5 are also coreferent, while m6 is not coreferent with any other event mentions. An event mention consists of a trigger and zero or more arguments. The trigger of an event men- tion is the word/phrase that is considered the most representative of the event, such as the word meet- ing for m3 or the word elected for m6. Triggers of coreferent event mentions must be related, that is, they should describe the same type of events. For example, m1 and m3 cannot be coreferent, since their trigger words — elect and meeting — are not related. Arguments are the participants of an event, each having its role. For example, KMT is the AGENT- argument and new party chief is the PATIENT- argument …

Content truncated to 500 words