Chinese Portion

Coreference resolution systems are typically trained with heuristic loss functions that re- quire careful tuning. In this paper we in- stead apply reinforcement learning to directly optimize a neural mention-ranking …...