Context-aware Attention Model for Coreference Resolution

Yufei Li, Xiangyu Zhou, Jie Ma, Yu Long, Xuan Wang, Chen Li

Keywords: Coreference resolution, Feature Attention

TL;DR: We demonstrate an attention model reweighing features around different contexts to reduce the wrongful predictions between similar or identical texts units

Abstract: Coreference resolution is an important task for gaining more complete understanding about texts by artificial intelligence. The state-of-the-art end-to-end neural coreference model considers all spans in a document as potential mentions and learns to link an antecedent with each possible mention. However, for the verbatim same mentions, the model tends to get similar or even identical representations based on the features, and this leads to wrongful predictions. In this paper, we propose to improve the end-to-end system by building an attention model to reweigh features around different contexts. The proposed model substantially outperforms the state-of-the-art on the English dataset of the CoNLL 2012 Shared Task with 73.45% F1 score on development data and 72.84% F1 score on test data.