Large language model and Chinese near synonyms : designing prompts for online CFL learners

Abstract

We propose a novel approach of applying large language models (LLMs) to better identify the Zone of Proximal Development (ZPD) of learners of Chinese as a foreign language (CFL). In particular, we designed prompts that assist LLMs in identifying the correct ZPD for CFL learners in order to provide more effective scaffolding. This study utilizes near synonyms to actuate this scaffolding procedure. By beginning with a base prompt and optimizing it in iterative instances, the models are better able to identify proper use-cases for the nuances of each near synonym, leading to more accurate and practical feedback responses. In three experiments, we used different prompts to test the capability of LLMs to understanding and differentiating near synonyms. We found that prompts containing explanations and guidance of reasoning can significantly improve the performance of these models. We attribute this improvement to the addition of interactive learning in prompt design. Adopting the scaffolding framework of learning, we propose the “Zone of Proximal Development Prompts” that can help LLMs to properly identify the correct ZPD of the CFL learners. 本研究提出了一种创新性的方法，来更好地应用大语言模型识别汉语作为外语学习者的最近发展区以提高学习效果。具体来说，我们通过设计提示来帮助大语言模型识别学习者的正确最近发展区，以提供更有效的学习支架。我们以近义词学习任务为本创新性方法的研究先导，首先给出基础提示，进而使用迭代的方法优化提示，促使大语言模型更好地识别近义词之间的细微差别，进而引导模型给出更为准确且实用的反馈。我们通过三个实验测试了大语言模型在不同提示下对近义词的理解和使用能力，并发现包含解释和思考指引的提示能显著提高模型的表现。我们将这一提高归因于在提示设计中融入了互动学习。采用支架式学习的理论框架，我们提出了“最近发展区提示”，这有助于大语言模型识别汉语学习者的最近发展区。