Chinese Zero Pronoun Resolution Based on Biaffine Attention Mechanism

Abstract

To solve the task of Chinese zero pronoun resolution, this paper proposes a Chinese zero pronoun resolution based on biaffine attention mechanism. First, Bert is employed to get representation of the text, and then build a candidate antecedent set for the target zero pronoun, and then get the representation of the local information and global information respectively for each antecedent candidate. Finally, biaffine attention mechanism is used in classification of the zero pronoun resolution. The Experiments on Chinese OntoNotes5.0 dataset show that our proposed approach can effectively improve the effect of Chinese zero pronoun resolution, and the F1 value reaches 59.7%.