基于多策略的藏语语义角色标注研究Multi-Strategic-计算语言学.PDF

文章编号：1003-0077（2011）00-0000-00 基于多策略的藏语语义角色标注研究 1 1 2 1 龙从军 ，康才畯 ，李琳 ，江荻 (1. 中国社科院民族所，北京市 100081; 2.青海师范大学计算机学院，青海省西宁市 810004 ） 摘要：语义角色标注研究对自然语言处理具有十分重要的意义。英汉语语义角色标注研究已经获得了很多 成果。然而藏语语义角色标注研究不管是资源建设，还是语义角色标注的技术探讨都鲜有报道。藏语具有 比较丰富的句法标记，它们把一个句子天然地分割成功能不同的语义组块，而这些语义组块与语义角色之 间存在一定的对应关系。根据这个特点，本文提出规则和统计相结合的、基于语义组块的语义角色标注策 略。为了实现语义角色标注，文中首先对藏语语义角色进行分类，得到语义角色标注的分类体系；然后讨 论标注规则的获得情况，包括手工编制初始规则集和采用错误驱动学习方法获得扩充规则集；统计技术上， 选用了条件随机场模型，并添加了有效的语言特征，最终语义角色标注的结果准确率、召回率和 F 值分别 达到82.78%、85.71%和83.91%。 关键词：藏语；语义角色标注；TBL ；CRFs 中图分类号：TP391 文献标识码：A Multi-Strategic Research on Semantic Role Labeling of Tibetan 1 1 2 1 LONG Congjun , KANG Caijun , Li Lin , JIANG Di （Institute of Ethnology Anthropology Chinese Academy of Social Sciences, Beijing 100081, China; The Computer College of Qinghai Normal University, Xining, Qinghai 810004, China ） Abstract: To study Semantic role labeling is of great significance for natural language processing. The researches of semantic role labeling about English and Chinese have obtained many achievements. However, the resources construction and technological means of semantic role labeling in Tibetan are still in initial stage. Tibetan has rich syntactic markers which naturally segment a sentence to different semantic chunking, and there are certain relationship between these semantic chunking and semantic roles. According to this characteristic, the authors of this paper propose the semantic role labeling strategy based on semantic chunking by combining two means of rules and statistics. In order to realize the semantic role labeling, the authors design classification system of Tibetan semantic roles and then discuss the acquisition of rules, including a manual initial rule sets and expanded rule sets from Transformation-