超越双语迁移：指令微调中的多语言代码切换
Beyond Bilingual Transfer: Multilingual Code-Switching in Instruction Tuning
AI总结：
本研究通过跨四种语言的句子级多语言代码切换指令微调，验证了多语言代码切换能有效提升大语言模型的多语言理解性能，超越了传统双语迁移设置。
摘要：
近期研究表明，代码切换数据（CSD）——即在同一上下文中混合多种语言——可以改善大语言模型（LLMs）的跨语言迁移和多语言对齐。然而，现有研究主要关注英语与目标语言之间的双语迁移，涉及三种或更多语言的多语言设置在很大程度上尚未被探索。在本工作中，我们研究了跨四种语言（英语、日语、韩语和中文）的多语言代码切换指令微调。我们在Belebele上评估多语言理解能力。我们的实验表明，简单的句子级多语言CSD持续提高了所有四种语言的平均多语言性能，表明多语言代码切换在双语迁移设置之外也能有效。
英文摘要：
Recent studies have shown that code-switching data (CSD), in which multiple languages are mixed within the same context, can improve cross-lingual transfer and multilingual alignment in large language models (LLMs). However, existing studies primarily focus on bilingual transfer between English and a target language, leaving multilingual settings involving three or more languages largely unexplored. In this work, we investigate multilingual code-switching instruction tuning across four languages: English, Japanese, Korean, and Chinese. We evaluate multilingual understanding on Belebele. Our experiments show that simple sentence-level multilingual CSD consistently improves average multilingual performance across all four languages, indicating that multilingual code-switching can be effective beyond bilingual transfer settings.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
