【摘要】 针对中文零代词识别任务,提出了一种基于深度神经网络的中文零代词识别模型.首先,通过注意力机制利用零代词的上下文来帮助表示缺省的语义信息.然后,利用Tree-LSTM挖掘零代词上下文的句法结构信息.最后,利用语义信息和句法结构信息的融合特征识别零代词.实验结果表明,相对于以往的零代词识别方法,该方法能够有效提升识别效果,在中文OntoNotes5.0数据集上的F1值达到63.7%.更多还原

【Abstract】 To solve the task of Chinese zero pronoun identification, this paper proposes a Chinese zero pronoun identification model based on deep neural network. Firstly,attention mechanism is applied to learn more semantic information from the context of zero pronoun. Then,Tree-LSTM is used to capture syntactic structure features of the context of the zero pronoun. Finally,semantic information and syntactic structure information are combined to identify the zero pronoun.Compared with the previous zero pronoun identification methods,experiments on Chinese OntoNotes5.0 corpus show that our proposed approach can more effectively improve the recognition effect,and theF1 value reaches 63.7%.更多还原