Modern Social Science Research
Modern Social Science Research. 2026; 6: (9) ; 10.12208/j.ssr.20260286 .
总浏览量: 34
1香港教育大学中国语言学系 中国香港
2香港科技大学工学院 中国香港
*通讯作者： 李文逸,单位：香港教育大学中国语言学系 中国香港; ；
研究针对通用大语言模型在中文教育领域应用时存在的结构性局限，通过综合案例分析及德尔菲专家评估法，剖析此类模型在对特定教育政策的理解、对不同学段教学内容的适配、与本土文化语境的结合、对个性化学习的支持等关键维度的不足。以香港地区基于学力差异的差异化教学实践为例，结合鲁迅《从百草园到三味书屋》等篇目的教学设计分析，发现通用模型因其训练数据对特定教学场景覆盖不足，且在算法层面难以有效建模复杂的教学逻辑，难以契合香港“两文三语”语言政策下的具体需求，无法精准响应不同学力学生的差异化学习要求。此基础上，研究进一步提出构建垂直领域中文教育大模型的可行路径，详细阐述从底层技术架构设计到上层应用方案，如差异化教学支持系统、文化敏感性适配模块的整体框架，以推动精准化教学并增强文化适应性，从而为研发真正贴合特定区域教育政策与实际需求的智能教育工具，提供兼具理论前瞻性与实践指导意义的参考。
General-purpose large language models are now used more in Chinese language education. But they still have many problems. This study uses case analysis and the Delphi method to find out what these models cannot do well. It looks at four key areas: understanding specific education policies, fitting content for different school levels, matching with local culture, and helping students learn in their own way. We take Hong Kong as an example. In Hong Kong teachers use differentiated instruction to meet different student needs. We look at teaching plans for Lu Xun’s from Baicao Garden to Sanwei Study and other texts. The results show that general models have two main limits. First, their training data does not cover enough teaching scenes. Second, their algorithms cannot model the complex logic of teaching easily. So they cannot meet the needs of Hong Kong’s “biliterate and trilingual” policy. They also cannot give different support to students of different levels. Based on these findings, the study suggests a way to build special models for Chinese education. It gives a clear framework from the bottom technical design to the top application, such as a differentiated instruction support system and a cultural sensitivity module. These can help make teaching more precise and improve cultural adaptability. The study offers both useful ideas for future research and practical advice for building smart education tools that fit real regional policies and classroom needs.
[1] 中华人民共和国教育部. 中国智慧教育白皮书发布[N/OL]. 人民网, 2025-05-16 [2025-12-04].
http://www.moe.gov.cn/jyb_xwfb/xw_zt/moe_357/2025/2025_zt06/mtbd/202505/t20250516_1190804.html.
[2] 温儒敏. 温儒敏谈人文学科与语文教育如何面对AI：跟进，但不盲目追赶[N/OL]. 腾讯新闻, 2025-09-06 [2025-12-04]. https://news.qq.com/rain/a/20250906A06JXR00.
[3] Vygotsky L S. Mind in society: The development of higher psychological processes[M]. Cambridge: Harvard University Press, 1978:34-54.
[4] Sweller J. Cognitive load during problem solving: Effects on learning[J]. Cognitive Science, 1988, 12(2):257-285.
[5] Stiegler B. For a new critique of political economy[M]. Cambridge: Polity Press, 2010:22-28.
[6] Nussbaum M C. Cultivating humanity: A classical defense of reform in liberal education[M]. Cambridge: Harvard University Press, 1997:8-16.
[7] Kohnke L, Moorhouse B L, Zou D. ChatGPT for language teaching and learning[J]. RELC Journal, 2023, 54(2):537-550.
[8] Liu T, Mantuhac P B. Teacher support and student engagement in the conduct of blended learning instruction for english as a foreign language[J]. Journal of Education and Educational Research, 2024, 11(2):22-31.
[9] Mangold A, Gawer L, Weinhold S, 等. From fragmentation to focus: How AI can assist researchers in academic writing[C]//International Conference on Human-Computer Interaction. Cham: Springer Nature Switzerland, 2025:55-71.
[10] Zhang W, Zou L. Generative artificial intelligence enabling fragmented learning: Status, challenges and optimization strategies[C]//Proceedings of the 2nd Guangdong-Hong Kong-Macao Greater Bay Area Education Digitalization and Computer Science International Conference. 2025:94-99.
[11] 王品芝, 邵钰倩. AI时代，六成受访者认为“走心表达”有助对抗信息碎片化的空虚感[N]. 中国青年报, 2025-08-21(004).
[12] Hongbing Z. The development trends, practical dilemmas and promotion strategies of adults’ fragmented learning in the context of big data[J]. Adult Education, 2023, 43(7):6-12.
[13] Xiaowei T, Rui N. On the integration path of teachers’ fragmented learning in the digital age: From the perspective of cognitive surplus[J]. China Educational Technology, 2022(2):69-74,97.