【摘要】 针对汉越跨语言新闻事件因果关系识别中，汉越跨语言的文本语义空间难以统一、新闻之间的因果关联特征捕获困难的问题，提出了基于类型矩阵转移的汉越跨语言新闻事件因果关系识别方法。通过跨语言预训练统一汉越跨语言的文本语义空间，使用树形长短期记忆循环神经网络提取汉越文本中的句法结构化特征，融入汉越句法特征并结合基于事件类型转移的注意力机制，对汉越事件句对的因果关系进行识别。实验结果表明，该方法在汉越跨语言新闻事件因果关系的识别上较基线模型准确率有所提升。更多还原

【Abstract】 To address the problems in the cross-language news event causality identification task for Chinese-Vietnamese, such as the difficulty in unifying the text semantic space across Chinese-Vietnamese and capturing the causal correlation features between news, we propose a Chinese-Vietnamese cross-lingual news event causality identification method based on type matrix transfer. First, the text semantic space across Chinese and Vietnamese languages is unified through cross-lingual pre-training. Second, a tree-shaped long-short-term memory recurrent neural network is used to extract syntactically structured features in Chinese-Vietnamese texts. Finally, the event causality between Chinese-Vietnamese temporal sentence pairs is identified by incorporating Chinese-Vietnamese syntactic features and combining the attention mechanism based on temporal type transfer. Experimental results demonstrate that our method improves the accuracy of identifying causal relationships between Chinese and Vietnamese cross-language news events when compared to the best baseline model.更多还原