摘要
针对线性条件随机场模型不能清楚表达语义角色内部结构关系的问题,提出一种基于树状条件随机场模型的语义角色标注方法。对句法依存树上的层次依赖关系和兄弟依赖关系进行标注,处理状态变量之间的长距离依赖,利用CRFs模型能添加任意特征的优点,在系统中添加新的组合特征和介词短语角色。在CoNNL 2008 Shared Task语料库上进行实验,结果证明该方法能有效提高系统的准确率和召回率。
Based on the deficiency of Conditional Random Fields（CRFs） can not describe structure relationship of the internal semantic roles more exactly, this paper proposes an approach to Semantic Role Labeling（SRL） which is based on Tree Conditional Random Fields（TCRFs） model. By labeling Hierarchical dependencies and Brother dependencies of syntactic dependency tree, it can deal with the long-distance dependencies between different state variants effectively. Meanwhile, taking advantage of CRFs model can add any features, some new combinative features and preposition phrase role are added to the system. Experimental results which are based on CoNNL 2008 Shared Task show that the proposed method can improve precision and recall rate of the system.
出处
《计算机工程》
CAS
CSCD
北大核心
2010年第18期41-42,45,共3页
Computer Engineering
基金
甘肃省自然科学基金资助项目(0809RJZA018)
关键词
语义角色标注
特征选择
树状条件随机场
semantic role labeling
feature selection
Tree Conditional Random Fields（TCRFs）
