A critical function of human language is to specify who did what to whom. Sentences serve this function by acting as miniature plays, with actors(typically labeled by noun phrases) playing roles in the event labeled b...A critical function of human language is to specify who did what to whom. Sentences serve this function by acting as miniature plays, with actors(typically labeled by noun phrases) playing roles in the event labeled by the verb. In the study of modern syntactic and semantics, theta role occupies quite an important place. This paper is a general overview of thematic role which may make a successful story on further study.展开更多
现有的中文事件抽取方法存在触发词和论元依赖建模不足的问题,削弱事件内的信息交互,导致论元抽取性能低下,特别是论元角色存在重叠的情况下.对此,文中提出基于图注意力和表指针网络的中文事件抽取方法(Chinese Event Extraction Method...现有的中文事件抽取方法存在触发词和论元依赖建模不足的问题,削弱事件内的信息交互,导致论元抽取性能低下,特别是论元角色存在重叠的情况下.对此,文中提出基于图注意力和表指针网络的中文事件抽取方法(Chinese Event Extraction Method Based on Graph Attention and Table Pointer Network,ATCEE).首先,融合预训练字符向量和词性标注向量作为特征输入,并利用双向长短期记忆网络,得到事件文本的强化语义特征.再将字符级建模的依存句法图引入图注意力网络,捕获文本中各组成成分的长距离依赖关系.然后,使用表填充的方法进行特征融合,进一步增强触发词和其对应的所有论元之间的依赖性.最后,将学习得到的表特征输入全连接层和表指针网络层,进行触发词和论元的联合抽取,使用表指针网络对论元边界进行解码,更好地识别长论元实体.实验表明:ATCEE在ACE2005和DuEE1.0这两个中文基准数据集上都有明显的性能提升,并且字符级依存特征和表填充策略在一定程度上可以解决论元角色重叠问题.ATCEE源代码地址如下:https://github.com/event6/ATCEE.展开更多
文摘A critical function of human language is to specify who did what to whom. Sentences serve this function by acting as miniature plays, with actors(typically labeled by noun phrases) playing roles in the event labeled by the verb. In the study of modern syntactic and semantics, theta role occupies quite an important place. This paper is a general overview of thematic role which may make a successful story on further study.
文摘现有的中文事件抽取方法存在触发词和论元依赖建模不足的问题,削弱事件内的信息交互,导致论元抽取性能低下,特别是论元角色存在重叠的情况下.对此,文中提出基于图注意力和表指针网络的中文事件抽取方法(Chinese Event Extraction Method Based on Graph Attention and Table Pointer Network,ATCEE).首先,融合预训练字符向量和词性标注向量作为特征输入,并利用双向长短期记忆网络,得到事件文本的强化语义特征.再将字符级建模的依存句法图引入图注意力网络,捕获文本中各组成成分的长距离依赖关系.然后,使用表填充的方法进行特征融合,进一步增强触发词和其对应的所有论元之间的依赖性.最后,将学习得到的表特征输入全连接层和表指针网络层,进行触发词和论元的联合抽取,使用表指针网络对论元边界进行解码,更好地识别长论元实体.实验表明:ATCEE在ACE2005和DuEE1.0这两个中文基准数据集上都有明显的性能提升,并且字符级依存特征和表填充策略在一定程度上可以解决论元角色重叠问题.ATCEE源代码地址如下:https://github.com/event6/ATCEE.
基金国家高技术研究发展计划(863)(the National High-Tech Research and Development Plan of China under Grant No.2006AA01Z147)国家自然科学基金(the National Natural Science Foundation of China under Grant No.60673041)高等院校博士学科点专项科研基金(the China Specialized Research Fund for the Doctoral Program of Higher Education under Grant No.20060285008)