类型感知的汉越跨语言事件检测方法  

Chinese-Vietnamese cross-language event detection method basedon event type awareness

作　　者：张磊[1,2] 高盛祥[1,2] 余正涛[1,2] 刘畅 陈瑞清[1,2] ZHANG Lei;GAO Shengxiang;YU Zhengtao;LIU Chang;CHEN Ruiqing(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,P.R.China;Yunnan Key Laboratory of Artificial Intelligence,Kunming University of Science and Technology,Kunming 650500,P.R.China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]昆明理工大学云南省人工智能重点实验室,昆明650500

出　　处：《重庆邮电大学学报（自然科学版）》2022年第5期803-811,共9页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金(61972186,61761026,61762056);国家重点研发计划(2018YFC0830105,2018YFC0830101,2018YFC0830100);云南高科技人才项目(201606);云南省重大科技专项计划(202002AD080001-5);云南省基础研究计划(202001AS070014,2018FB104)。

摘　　要：针对汉越跨语言事件检测缺少平行语料,越南语标注困难,需要统一跨语言语义空间,且触发词存在较大的歧义和局限性等问题,提出基于事件类型感知的汉越跨语言事件检测方法。构造类型感知的注意力机制突显事件特征,融入汉越的词位置、词性和命名实体信息,并通过梯度反转(gradient reversal layer,GRL),实现有标注汉语和无标注越南语之间的对抗训练,将从大量汉语新闻文本中学到的语言无关的事件类型特征融入到联合特征提取器中,进行汉越跨语言的无触发词事件检测,缓解越南语的数据稀缺和触发词的局限性。实验中提出的方法较最好的基线模型在准确率上提升了4.32%。In view of the lack of parallel corpus for Chinese-Vietnamese cross-language event detection,the difficulty of Vietnamese labeling,the need to unify the cross-language semantic space,and the large ambiguity and limitations of trigger words,this paper proposes an event-type-aware-based Chinese-Vietnamese cross-language event detection method.First,an event-type-aware attention mechanism is constructed to highlight event features,integrate Chinese and Vietnamese language differences into position,part of speech,and named entity information,and achieve confrontation training between tagged Chinese and unlabeled Vietnamese through gradient reverse layer(GRL).The language independent event type features learned from a large number of Chinese news texts are integrated into the joint feature extractor to detect Chinese Vietnamese cross language non-trigger word events.It alleviates the data scarcity and the limitation of trigger words in Vietnamese.The accuracy of the proposed method is 4.32%higher than that of the best baseline model.

关 键 词：汉越跨语言事件检测 无触发词 事件类型感知 梯度反转 语言对抗 

分 类 号：TN391.3[电子电信—物理电子学]