摘要
现有的中文事件论元抽取方法大多利用句法结构来表示论元和触发词之间的关系,该方法无法抽取与触发词距离较远且不在同一个子句中的论元。为了解决上述问题,基于马尔科夫逻辑网络(MLN),通过学习训练语料中实体填充不同角色的概率和测试语料中部分已知论元信息,来抽取其他可信度低或缺乏有效信息的论元。在ACE2005中文语料上的实验结果表明,所提方法与基准系统相比,系统性能在论元识别和论元角色分配阶段分别提高了6.0%和4.4%。
Currently,previous Chinese argument extraction approaches mainly use syntactic structure as the major features to describe the relationship between trigger and its arguments.However,they suffer much from those inter-sentence arguments which are not in the same sentence or clause of the trigger.To address this issue,this paper brought forward a novel argument inference mechanism based on the Markov logic network.It first learns the probabilities of an entity fulfilling a specific role from the training set and obtains those extracted argument mentions with high confidences in the test set.Then it uses them to extract those argument mentions with lack of effective context information or low confidences.Experimental results on the ACE 2005 Chinese corpus show that our approach outperforms the baseline significantly,with the improvements of 6.0 % and 4.4 % in argument identification and role determination respectively.
出处
《计算机科学》
CSCD
北大核心
2016年第3期252-255,261,共5页
Computer Science
基金
国家自然科学基金(61472265)
国家自然科学基金重点项目(61331011)
江苏省前瞻性联合研究项目(BY2014059-08)
软件新技术与产业化协同创新中心部分资助
关键词
论元抽取
马尔科夫逻辑网络
论元推理
Argument extraction
Markov logic network
Argument inference