【作者】 董静； 孙乐； 吕元华； 冯元勇；

【Author】 DONG Jing, SUN Le, LV Yuanhua, FENG Yuanyong Chinese Information Processing Center, Institute of Software. Chinese Academy of Sciences, Beijing 100080

【机构】 中国科学院软件研究所中文信息处理中心；

【摘要】 语义角色标注是对语义理解的一种新的探索,具有很重要的研究意义。本文提出一种基于条件随机场模型上的语义标注方法。该方法通过将句法分析树“压平”,考虑句法树中水平层次上的角色标签之间的马尔科夫依赖关系,提高了系统的精度。另外,本文还分析比较了不同特征对于标注任务的影响并且提出一些新的特征。实验效果表明这些新特征都提高了系统的准确率和召回率。更多还原

【Abstract】 Semantic Role Labeling(SRL) plays an important role in Natural Language Processing as a rising exploration to semantic understanding. This paper presents an approach to SRL based on Conditional Random Fields(CRF) Model, by incorporating tree-to-string alignment and linear dependence among role labels at horizontal level.The experiments indicate that, compared with the Tree CRF model, our approach not only reduce the time and space costs, but also increase the accuracy of the system. Furthermore, the paper also analyzes the effects of various features in the task of labeling and suggests some new features. Our experimental results show that these new features successfully improve the accuracy and recall of the system.更多还原