基于语义的中文事件触发词抽取联合模型

中文事件触发词抽取是一项具有挑战性的任务.针对中文事件触发词抽取中存在的事件论元语义信息难以获取以及部分贫信息事件实例难以抽取的问题,提出了基于语义的中文事件触发词抽取联合学习模型.首先,根据中文句子结构灵活和句法成分多省略的特点,提出了基于模式匹配的核心论元和辅助论元抽取方法,这两类论元可以较好地表示论元语义,进一步提高中文事件触发词抽取性能;其次,根据同一文档中关联事件实例间存在的高度一致性,构造了一个关联事件语义驱动的中文事件触发词识别和类型分配二维联合模型,用于抽取贫信息事件实例.在ACE 2005中文语料上的实验结果表明:与现有最好的中文事件抽取系统相比,所提出方法的性能得到了明显提升.

Semantics-Based Joint Model of Chinese Event Trigger Extraction

Chinese event trigger extraction is a challenging task. To tackle the difficulties of obtaining the semantic information of event arguments and extracting those context-poor event mentions in Chinese event trigger extraction, this paper proposes a semantics-driven joint model to integrate the components of Chinese event trigger extraction. First, considering the nature of Chinese language (e.g., flexible sentence structure and ellipsis), it provides a pattern-based method to identify core arguments and supplement arguments to better represent argument semantics, and applies the method to improve the performance of Chinese trigger extraction. Secondly, regarding the consistency among relevant event mentions in a document or discourse, it introduces the semantics among relevant event mentions to formulate a 2-dimensional joint model of Chinese trigger detection and type allocation to extract those context-poor event mentions. Finally, it provides experimental results on the ACE 2005 Chinese corpus to show that the presented model significantly outperforms the state-of-the-art system.