基于层次化聚类的稀疏谓词语义角色标注方法

Rare predicate's semantic role labeling based on hierarchical clustering

中文语义角色标注中,稀疏谓词的标注性能要远远低于其它谓词,而在实际应用中,标注系统经常需要处理大量的稀疏谓词,因此,稀疏谓词问题大大限制了语义角色标注系统的应用效果。为解决上述问题,提出一种基于聚合层次化聚类的方法。通过聚合层次化聚类建立起稀疏谓词与常见谓词的联系,稀疏谓词可以泛化为与之语义相近的常用谓词,缓和语义角色标注系统中的稀疏谓词问题。在中文命题库上的实验结果表明,该方法可有效处理中文语义角色标注中的稀疏谓词问题。

For Chinese semantic role labeling(SRL),the labeling performance of rare predicates’semantic role labeling is much poorer than common predicates.However,a labeling system often faces a lot of rare predicates.Therefore,this problem restricts SRL’s application performance.A method based on hierarchical clustering was proposed.Rare predicates and common ones were linked through hierarchical clustering and a rare one was generalized into a common one,which relieved the above problem in Chinese semantic role labeling.Experiments results on Chinese Propbank show that the proposed method can improve significantly rare predicates’semantic role labeling performance.