大语言模型在跨文化语言理解中的潜力与局限

Potentials and Limitations of Large Language Models in Cross-Cultural Language Understanding

摘要: 随着全球化的发展，跨文化语言理解在日常交际、教育和人工智能应用中愈发重要。大语言模型(LLM)如ChatGPT在生成跨文化表达方面展现出一定潜力，但其对文化语境和隐含意义的理解仍存在局限。本文以礼貌表达、幽默和隐喻为研究对象，通过让ChatGPT生成典型跨文化表达示例，结合跨文化语言理解理论，对其生成内容进行分析。研究发现，LLM能够生成符合文化预期的语言表达，但在面子、礼貌策略、情感微调和语境依赖上仍表现不足。这一研究为跨文化交流、语言教学及人工智能在跨文化应用中的改进提供了参考，同时提出了大语言模型在理解文化内涵方面的局限性及优化方向。

Abstract: With the progression of globalization, cross-cultural language understanding has become increasingly vital in daily communication, education, and artificial intelligence applications. Large Language Models (LLMs) such as ChatGPT have demonstrated significant potential in generating cross-cultural expressions; however, their understanding of cultural context and implicit meanings remains limited. Focusing on politeness expressions, humor, and metaphors, this paper analyzes the content generated by ChatGPT through typical cross-cultural expression examples, grounded in theories of cross-cultural linguistic understanding. The findings indicate that while LLMs are capable of generating language expressions that align with cultural expectations, they still exhibit deficiencies in handling face-saving, politeness strategies, emotional fine-tuning, and contextual dependency. This study provides insights for the improvement of cross-cultural communication, language teaching, and the cross-cultural application of AI, while highlighting the limitations of LLMs in grasping deep cultural connotations and proposing directions for optimization.