汉语语篇的连贯性标注困难度分析研究  

Research on difficulty of Chinese text coherence annotation

作　　者：魏天珂 吕学强[1] 周强[2] Wei Tianke;Lv Xueqiang;Zhou Qiang(Beijing Key Laboratory of Internet Culture&Digital Dissemination Research,Beijing Information Science&Technology University,Beijing 100101,China;;Tsinghua National Laboratory for Information Science&Technology(TNList),Center for Speech&Language Technologies,Research Institute of Information Technology,Tsinghua University,Beijing 100084,China)

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101 [2]清华大学信息技术研究院语音与语言技术中心清华信息科学与技术国家实验室(筹),北京100084

出　　处：《计算机应用研究》2018年第10期2969-2974,共6页Application Research of Computers

基　　金：国家"973"计划资助项目(2013CB329304);国家自然科学基金资助项目(61433018;61373075;61411130162;61271304;61671070);北京成像技术高精尖创新中心资助项目(BAICIT-2016003);国家社会科学基金重大资助项目(14@ZH036;15ZDB017);国家语委重大课题资助项目(ZDA125-26)

摘　　要：在人工标注语篇句子连贯性时,句群的划分成为最大的难点,句群内部句间关系的复杂性也增加了篇章标注的困难度。为了识别语篇标注的困难度,从多角度总结了句群边界以及类别的分布特点,分析了句群内部句间的各种连接关系,提出了一种基于人工标注的语篇困难度识别模型,并利用双人标注信息的差异性来验证语篇困难度识别方法的有效性与准确性。数据表明,该模型可以较好地区分不同体裁新闻类语篇的标注难度,为相关语篇内容的分析理解打下了良好的基础。The segmentation of sentence group in a monologue text becomes the most intractable problem when annotating the coherence of the text.The complexity of the inter-sentence relationship within the sentence group also increases the difficulty of the annotation.This paper proposed a new computational model to estimate the text coherence annotation difficulty based on the different statistical features of two independent annotators.The experimental results show that the model can provide better discriminative power for the news texts with different annotation difficulty.The research work lays a good foundation for the analysis and understanding of relevant text content in the future.

关 键 词：句群识别 语篇连贯性 语篇标注困难度 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]