汉语语篇衔接特征的自动分析和应用研究  

Automatic Analysis and Application of Cohesion Features of Chinese Texts

作　　者：彭一平 胡韧奋 吴继峰[2] PENG Yiping;HU Renfen;WU Jifeng

机构地区：[1]北京师范大学,北京100875 [2]首都师范大学,北京100048

出　　处：《语言文字应用》2023年第1期114-129,共16页Applied Linguistics

基　　金：教育部中外语言交流合作中心国际中文教育中外联合研究专项课题“基于新标准的智能化语言分析技术研究”(22YH04ZW);北京市社科基金青年项目“面向汉语国际教育的学术汉语资源库建设研究”(21YYC011)资助。

摘　　要：语篇衔接特征在写作质量预测、文本可读性分析、文本生成、翻译质量评估等任务上有重要作用。为了实现汉语语篇衔接特征的自动分析,本文在国内外语篇衔接连贯理论研究的基础上,从词汇、语法和话题三个维度对汉语的语篇衔接手段进行讨论,提出了27个可量化的测量指标,并基于自然语言处理技术构建了汉语语篇衔接特征自动分析工具。最后,将语篇衔接自动分析工具应用到汉语二语写作质量评估任务中,实验显示,本文所提出的汉语语篇衔接指标具有较强的预测效力(Adjusted R^(2)=0.475),与现有工具中文CTAP中的语篇衔接指标融合后,能够取得更好的表现(Adjusted R^(2)=0.568)。在此基础上,本文进一步讨论了二语作文中的词汇、语法和话题衔接特点。Cohesion plays an important role in essay quality prediction,text readability analysis,text generation,translation quality evaluation and other tasks.To achieve the automatic analysis of Chinese discourse cohesion,this paper discusses the lexical cohesion,grammatical cohesion and topic cohesion features,and proposes 27 cohesion measures.After that,a automatic analysis tool of Chinese discourse cohesion features is built based on natural language processing technologies.We apply this tool to the evaluation of Chinese second language writing quality,and compares the proposed measures with those from an existing automatic text analysis tool.Experiments show that the proposed measures in this paper has strong predictive power on the writing quality(Adjusted R^(2)=0.475),and after integrating the measures from the existing tool,the combined model can achieve a significant improvement(Adjusted R^(2)=0.568).Based on the experimental results,we further discuss the characteristics of lexical,grammar and topic cohesion reflected in Chinese second language writing.

关 键 词：语篇 衔接 自动分析 二语写作 

分 类 号：H087[语言文字—语言学]