This study examines error patterns in the use of the dynamic aspect markers le, zhe, and guo in the interlanguage of Korean learners of Chinese, and compares the corrective strategies and effects of human teachers with those of artificial intelligence systems represented by ChatGPT-5 and DeepSeek in grammatical error correction. The findings indicate that AI demonstrates high efficiency in correcting frequent and rule-based overt errors, producing revisions that are largely comparable to those of teachers in terms of formal grammatical accuracy. However, when errors involve semantic distinctions, aspectual function differentiation, or discourse-level coherence, AI corrections tend to remain at a surface level, lacking systematic explanation and pedagogical orientation. In contrast, human teachers integrate contextual information and aspectual semantics to provide targeted analysis and feedback, which better supports learners’ construction of aspectual understanding. Based on these findings, this study argues that AI is currently more suitable as a supplementary tool for grammatical error correction, and proposes promoting the development of AI correction systems toward greater interpretability and pedagogical integration in order to achieve a human-AI collaborative model of Chinese language teaching.

本研究以韩国汉语学习者的中介语语料为基础，考察动态助词“了、着、过”的偏误 类型，并比较人工教师与以ChatGPT-5、DeepSeek 为代表的人工智能系统在语法纠错中的 修正策略与效果。研究发现，人工智能在高频、规则性强的显性偏误修正方面具有较高效 率，其修改结果在形式正确性上与教师较为接近；但在涉及语义辨析、体貌功能区分及篇 章连贯性的问题上，AI 的修正多停留于形式层面，缺乏系统解释与教学指向。相较之 下，人工教师能够结合语境与体貌语义，对偏误进行有针对性的分析与反馈，更有助于学 习者建构体貌理解。基于研究结果，本文认为当前AI 更适宜作为语法纠错的辅助工具， 并提出推动纠错系统向可解释化、教学化方向发展的应用建议，以实现人机协同的汉语教 学模式。

【目 录】
1. 绪论
2. 理论背景
    1) 动态助词“了”的语法特征
    2) 动态助词“着”的语法特征
    3) 动态助词“过”的语法特征
3. 研究设计
    1) 语料来源与样本选择
    2) 实验流程与提示语设计
4. 韩国学习者动态助词偏误及修正分析
    1) 偏误总体分布
    2) “了”偏误及修正比较
    3) “着”偏误及修正比较
    4) “过”偏误及修正比较
5. 结论与启示
【参考文献】
【论文摘要】

• 章万(韩国加图立大学中语中文系博士研究生) | 장완
• 李炫周(韩国加图立大学中国语言文化系助教授) | 이현주 Corresponding author