Event Coreference Resolution with Multi-Pass Sieves

Event Coreference Resolution with Multi-Pass Sieves Jing Lu and Vincent Ng Human Language Technology Research Institute University of Texas at Dallas Richardson, TX 75083-0688, USA {ljwinnie,vince}@hlt.utdallas.edu Abstract Multi-pass sieve approaches have been successfully applied to entity coreference resolution and many other tasks in natural language processing (NLP), owing in part to the ease of designing high-precision rules for these tasks. However, the same is not true for event coreference resolution: typically lying towards the end of the standard information extraction pipeline, an event coreference resolver assumes as input the noisy outputs of its upstream components such as the trigger identification component and the entity coreference resolution component. The difficulty in designing high-precision rules makes it challenging to successfully apply a multi-pass sieve approach to event coreference resolution. In this paper, we investigate this challenge, proposing the first multi-pass sieve approach to event coreference resolution. When evaluated on the version of the KBP 2015 corpus available to the participants of EN Task 2 (Event Nugget Detection and Coreference), our approach achieves an Avg F-score of 40.32%, outperforming the best participating system by 0.67% in Avg F-score. Keywords: event coreference resolution, discourse processing, information extraction 1. Introduction Within-document event coreference resolution is the task of determining which event mentions in a text refer to the same real-world event. Compared to entity coreference res- olution, event coreference resolution is not only much less studied, but it is arguably more challenging. The chal- lenge stems in part from the fact that an event coreference resolver typically lies towards the end of the standard in- formation extraction pipeline, assuming as input the noisy outputs of its upstream components. More specifically, an event coreference resolver assumes as input not only the event triggers, their types/subtypes, and their arguments, but also entity coreference information. Different corpora have been used to train and evaluate event coreference resolvers, but as Liu et al. (2014) pointed out, not all of them were carefully annotated. As will be dis- cussed in more detail in Section 2, OntoNotes and ECB have only be partially annotated with event coreference links. Among the publicly-available corpus, the ACE 2005 corpus is arguably the one that is most complete with re- spect to the annotation of event coreference links. In fact, the majority of recent work on event coreference was eval- uated on the ACE 2005 corpus. As an event coreference corpus, ACE 2005 has a major weakness: it adopts a strict notion of event identity. Specif- ically, two event mentions were annotated as coreferent if and only if “they had the same agent(s), patient(s), time, and location” (Song et al., 2015), and their event attributes (polarity, modality, genericity, and tense) are not incompat- ible. This is arguably an overly strict definition of event coreference, as some event mentions are intuitively coref- erent even if their time and/or location arguments are not identical. The new KBP 2015 event coreference corpus was created in response to the aforementioned weakness of the ACE 2005 corpus (Song et al., …

Content truncated to 500 words