人机协作的游戏文化元素本地化新探——以《原神》活动对联英译为例

An Exploration of Human-AI Collaboration in the Localization of Cultural Elements in Games—Taking English Translation of Couplets in Genshin Impact Events as an Example

摘要: 大语言模型正在深刻改变翻译与语言服务行业的发展范式。这一变革在网络游戏本地化领域日趋显著——游戏剧情文本兼具多模态、交互性与沉浸感特征，其文化元素融合现实与虚拟世界，对本地化工作者的创译能力提出了挑战。研究以开放世界角色扮演游戏《原神》为例，通过构建提示词探索DeepSeek模型在游戏文化元素本地化中的应用潜力。研究表明，大语言模型能够辅助生成多版本适配文本，有效提升不同玩家群体的剧情体验，并在保持文本风格一致性方面展现优势。然而，研究同时揭示了大模型在文化适配精准度、幻觉等方面的局限性，表明当前技术仍需与专业译者的人机协作才能实现最优本地化效果。这一发现为人工智能时代的游戏本地化实践提供了重要的方法论参考。

Abstract: Advancement of large language models (LLMs) are profoundly reshaping the language services industry. The transformation is increasingly evident in the field of video game localization—game narratives, characterized by multimodality, interactivity, and immersiveness, integrate cultural elements spanning both real and virtual worlds, posing challenges to the transcreation capabilities of localization professionals. This study takes the open-world role-playing game Genshin Impact as an example and explores the potential of DeepSeek in the localization of in-game cultural elements by constructing tailored prompts. The findings demonstrate that LLMs can assist in generating multiple adapted versions of text, effectively enhancing the narrative experience for diverse player demographics while exhibiting strengths in maintaining stylistic consistency. However, the study also reveals limitations in LLMs’ precision in cultural adaptation and susceptibility to hallucinations, indicating that optimal localization outcomes still require human-AI collaboration with professional translators at present. These insights provide significant methodological references for game localization practices in the era of artificial intelligence.