Deep Reinforcement Learning for Mention-Ranking Coreference Models
This addresses the issue of heuristic loss functions requiring careful tuning in coreference resolution, offering a more direct optimization approach for researchers and practitioners in natural language processing.
The paper tackled the problem of training coreference resolution systems by applying reinforcement learning to directly optimize a neural mention-ranking model for evaluation metrics, resulting in significant improvements over the state-of-the-art on the English and Chinese portions of the CoNLL 2012 Shared Task.
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 reward-rescaled max-margin objective. We find the latter to be more effective, resulting in significant improvements over the current state-of-the-art on the English and Chinese portions of the CoNLL 2012 Shared Task.