Download Full PDF

康雪

王曦悦

余兰心

祁峰 

华东师范大学，中国

 

摘要

本文聚焦国际中文教学中留学生语体能力发展滞后于语法能力的问题，探讨 ChatGPT 等大语言模型辅助语体纠偏的实践效能。研究从 HSK 动态作文语料库中抽取 2000 份作文样本，分析发现留学生书面语口语化偏误主要集中于单双音节对应律、长短律及时空律，其中单双音节对应律偏误发生率达 49.6%。通过对比 ChatGPT、文心一言和讯飞星火三种大语言模型的表现，选定 ChatGPT 作为实验工具，对 31 名汉语水平为 HSK 4-6 级的华东师范大学留学生进行实验。结果表明， ChatGPT 辅助训练能够降低汉语作为第二语言被试的书面语口语化偏误，对华裔被试效果虽不显著但也有一定效果。研究初步支持大语言模型作为语体教学辅助工具的可行性，并为国际中文教学在智能化背景下的实践提供了可参考的实证依据。

 

关键词

ChatGPT，语体教学，人工智能，大语言模型

 

An Empirical Evaluation of ChatGPT's Effectiveness in Assisting International Students with Genre-Style Chinese Writing Correction

 

Xue Kang

Xiyue Wang

Lanxin Yu

Feng Qi

East China Normal University, China

 

Abstract

This study investigates the practical efficacy of ChatGPT and other large language models (LLMs) in assisting genre-style correction in writing, addressing the issue that international students’ genre competence often lags behind their grammatical competence in international Chinese language education. A sample of 2,000 compositions was drawn from the HSK Dynamic Composition Corpus, revealing that the colloquialization errors in students’ written productions predominantly involve the laws of monosyllabic-disyllabic correspondence, length, and temporospatiality, among which violations of the monosyllabic-disyllabic correspondence law account for 49.6% of all errors. A comparative evaluation of three LLMs—ChatGPT, Ernie Bot, and iFlytek Spark—was conducted, and ChatGPT was selected as the experimental tool. An experiment was carried out with 31 Chinese learners at the HSK Level 4-6 from East China Normal University. The results indicate that ChatGPT-assisted training significantly reduces colloquialization errors among Chinese-as-a-second-language (CSL) participants. For heritage learners, although the reduction did not reach statistical significance, a positive trend was observed. This study provides preliminary evidence supporting the feasibility of LLMs as auxiliary tools for genre instruction and offers empirical insights for the intelligent transformation of international Chinese language education.

 

Keywords

ChatGPT, genre teaching, artificial intelligence, large language model