Deep Reinforcement Learning for Mention-Ranking Coreference Models
Kevin B. Clark, Christopher D. Manning · 2016
Coreference resolution systems are typically trained with heuristic loss functions that require careful tuning.In this paper we instead apply reinforcement learning to directly optimize a neural mention-ranking model for coreference evaluation metrics.We experiment with two approaches: the REINFORCE policy gradient algorithm and a rewardrescaled max-margin objective.We find the latter to be more effective, resulting in a significant improvement over the current stateof-the-art on the English and Chinese portions of the CoNLL 2012 Shared Task.