生成式人工智能时代中文动态助词偏误分析与教学应对  

Error Analysis and Teaching Strategies of Chinese Dynamic Particles in the Era of Generative Artificial Intelligence

作　　者：肖锐 侯尚余 张邝弋 Xiao Rui;Hou Shangyu;Zhang Kuangyi(College of International Chinese Language Education,Yunnan University,Kunming Yunnan 650091)

机构地区：[1]云南大学汉语国际教育学院,云南昆明650091

出　　处：《沈阳师范大学学报（教育科学版）》2024年第5期12-20,共9页Journal of Liaoning Educational Administration Institute

基　　金：教育部人文社会科学重点研究基地重大项目(22JJD740016);云南省教育厅科学研究基金项目(2023Y0306);云南大学第三届专业学位研究生实践创新项目(ZC2323415)。

摘　　要：生成式人工智能(GAI)逐渐成为教育技术领域的重要组成部分,给语言教学带来新的机遇。研究聚焦大语言模型(LLMs)对动态助词“了”“着”“过”的偏误检测及其教学应用潜力。基于全球汉语中介语语料库,将动态助词偏误分为六大类型,并选取五种主流LLMs进行对比分析,通过三种提示策略评估其性能。结果显示,ChatGLM表现最佳,GPT4o相对较弱;思维链提示策略显著提升了偏误检测精确度但仍需优化。为增强LLMs中文偏误理解能力,需增加特定结构的训练数据,探索更有效的提示方法。未来,LLMs结合Agent的智能化系统有望提供定制化、交互式的学习体验,优化中文学习效率。Generative Artificial Intelligence has gradually become an important part of educational technology,bringing new opportunities to language teaching.The research focuses on the bias detection of large language models(LLMs)on the dynamic particles“Le”“Zhe”and“Guo”,and their teaching application potential.Based on the global Chinese interlanguage corpus,dynamic particle errors are divided into six types,five mainstream LLMs are selected for comparative analysis,and their performance is evaluated through three prompting strategies.The results show that ChatGLM performs best and GPT4o is relatively weak;the Chain of Thought prompting strategy significantly improves the accuracy of error detection but still needs to be optimized.In order to enhance LLMs'ability to understand Chinese errors,it is necessary to increase training data with specific structures and explore more effective prompting methods.In the future,the intelligent system of LLMs combined with Agent is expected to provide a customized and interactive learning experience and optimize the efficiency of Chinese learning.

关 键 词：大语言模型 中文动态助词 偏误检测 中文教学 

分 类 号：H195[语言文字—汉语] TP391.1[自动化与计算机技术—计算机应用技术]