EVENT COREFERENCE RESOLUTION

Abstract

Event coreference is an important task in event-extraction and other natural language processing tasks. Despite its importance, it was merely discussed in previous studies. In this thesis, we are the first in the literature providing a systematic and computation-oriented study on this challenging task. We present a global coreference-resolution system dedicated to sophisticated event coreference cases. First, seven resolvers are utilized to resolve different event and object Coreference pairs with new instance selection strategy and new linguistic features. Competing classifiers and topic-related event-detection are further imposed to enhance mention-pair resolvers. Second, two global solutions, spectral graph partitioning and random walk model, are employed for the chain formation. Spectral graph partitioning is equipped with heuristic-guidance and model specific manipulations to produce better coreference-chain results. Being the first attempt to apply random walk model for coreference resolution, the modified model utilizes a sampling method, termination criterion and stopping probability to greatly improve the effectiveness of random walk model for event coreference resolution. The new random walk model facilitates a convenient way to incorporate sophisticated linguistic constraints and preferences, the related object mention graph as well as pronoun coreference information not used in previous studies for effective chain formation. Collectively, all the above techniques impose significant B3 F-score improvement over baseline system on the OntoNotes4.0 Corpus.