中文语义角色标注的树核空间研究

摘要

使用基于树核函数的方法来进行语义角色标注,有效的树核空间的设计是影响系统性能的关键.探索树核空间在中文语义角色标注上的应用,考虑到同一谓词的各论元间的相互影响,提出多论元-谓词特征(AAPF)空间,并在此基础上提出了三种受平面特征启发的树核空间设计方法.基于中文PropBank语料的实验表明,加入一些重要平面特征信息的树核空间,性能有了明显的提高,分类精确率由90.96％提高到92.54％.最后使用复合核将特征启发的树核与特征向量结合起来,精确率达到95.21％,性能高于同类系统.%Effective design of tree kernel space plays a critical role in system performance when conducting SRL with tree kernel-based approaches. This paper explores the application of tree kernel space in Chinese SRL. Considering the dependence among the arguments of a predicate, in this paper we propose an All-Arguments Predicate Feature (AAPF) space and present three design methods of tree kernel space with flat features inspiration on this basis. Experiments on Chinese PropBank corpus show that the performance improvement of tree kernel space is noticeable when some important flat features information are added to tree kernel space, the classification accuracy rate increases from 90.96% to 92. 54%. At last, hybrid kernel is adopted to combine the feature-inspired tree kernel with eigenvector, its accuracy achieves 95. 12% , the performance outperforms the similar systems.