End-to-end neural event coreference resolution

Conventional event coreference systems commonly use a pipeline architecture and rely heavily on handcrafted features, which often causes error propagation problems leads to poor generalization ability. In this paper, we propose neural network-based end-to-end (E3C) that can jointly model detection resolution tasks learn extract features from raw text automatically. Furthermore, because mentions are highly diversified is intricately governed by long-distance semantically-dependent decisions, type-enhanced mechanism further proposed in our E3C network. Experiments show method achieves new state-of-the-art performance both standard datasets.