汉语篇章微观话题结构建模与语料库构建

发布时间：2018-02-16 20:46

本文关键词： 篇章话题结构主位-述位理论主位推进话题链语料库构建　出处：《计算机研究与发展》2017年08期 　论文类型：期刊论文

【摘要】：篇章话题结构分析是自然语言理解的前沿基础,而大规模高质量的适用于汉语篇章分析的语料资源缺乏,严重制约了相关篇章话题计算模型的研究.针对上述问题,首先研究了汉语篇章话题结构的理论表示体系.分析了主述位理论、英语修辞结构理论和宾州篇章树库体系的优势,结合汉语复句句群理论以及汉语自身特点,提出了一种基于主述位理论的汉语篇章微观话题结构表示方式,并借助微观话题链构建了汉语篇章话题结构表示体系.随后,在此基础上,采用自顶向下、后向搜索的标注策略和人机结合的语料库标注方式,构建了基于篇章微观话题表示体系的汉语篇章话题结构语料库(Chinese discourse topic corpus,CDTC).CDTC共包含500个文档,对其进行了详细统计分析并展示了语料库的标注情况.与宾州篇章树库体系、广义话题结构理论的对比表明,所提篇章微观话题结构表示体系在理论上具有一定的优越性,并且符合汉语特点;一致性检验表明CDTC能够充分体现汉语篇章话题分析问题本身的难度,并能够为相关研究提供语料资源支持.
[Abstract]:Discourse topic structure analysis is the front line of natural language understanding. However, the lack of large scale and high quality corpus resources for Chinese text analysis seriously restricts the research of relevant discourse topic computing models. Firstly, this paper studies the theoretical representation system of Chinese discourse topic structure, analyzes the advantages of subject and rheme theory, English rhetoric structure theory and Pennsylvania text tree library system, and combines the theory of sentence group of Chinese complex sentences and the characteristics of Chinese itself. In this paper, we propose a topic structure representation method of Chinese text based on thematic and rheme theory, and construct a Chinese discourse topic structure representation system with the help of micro topic chain. Then, on this basis, we adopt the top-down approach. The tagging strategy of backward search and the corpora tagging method of human-computer combination are used to construct a corpus of Chinese discourse topic Corpus CDTCf.CDTC containing 500 documents, which is based on the text micro-topic representation system. Compared with the Pennsylvania text tree system and the generalized topic structure theory, the proposed text micro-topic structure representation system has some advantages in theory. The consistency test shows that CDTC can fully reflect the difficulty of the topic analysis of Chinese text, and can provide corpus-based support for related research.
【作者单位】：苏州大学计算机科学与技术学院;苏州科技大学计算机科学与工程系;苏州市虚拟现实智能交互及应用技术重点实验室;
【基金】：国家自然科学基金项目(61331011,61673290,61472264)~~
【分类号】：TP391.1

【参考文献】