An Event-Oriented Multi-pass Sieve Module for Coreference Resolution

Coreference resolution is one of the key issues in the natural language processing, it can eliminate uncertain problems of event in the event-oriented natural language processing, and that is important for the upper application of event. This paper builds an event-oriented multi-pass sieve module for coreference resolution, and combined with the characteristics of the event, we add the constraint conditions to each sieve to improve the accuracy of each sieve. We use this module to carry on experiment for object elements of the event. Compared with machine learning method based on C4.5 decision tree, it has a very big enhancement on the performance.