跳到正文
arXiv cs.CL· Eleanor M. Lin, David Jurgens·· 10 小时前AI 评分31

研究提出 CoRe 与 EnCoRe 语料库及微调框架,教大模型更有效地进行语码转换推理

Think Multilingual, Not Harder: A Framework for Analyzing and Teaching Code-Switched Reasoning

AI 导读

研究者构建了首个从语言学与行为学出发的微调框架,用于识别并教大模型更有效地进行语码转换推理。他们先建立 CoRe 语料库,含 15 个模型、18 种语言、10 种文字体系的 7k 条推理轨迹,再扩展为 EnCoRe,用 40 个数据集在 3 个模型、7 种语言上评估六种干预、共 120 个微调条件。

正文 · 原文

View PDF HTML (experimental)

Abstract:Reasoning capabilities have enabled large language models to solve increasingly complex tasks. Interestingly, while reasoning models are often trained to generate monolingual text, these models have also been observed to code-switch (i.e., mix languages). Prior works have either viewed code-switching as an undesirable error, attempted to control code-switching through modifications to input prompts or the output decoding process, or focus on narrow subsets of languages, domains, tasks, and models. We address these gaps by introducing the first linguistically and behaviorally motivated fine-tuning framework for identifying beneficial code-switched reasoning behaviors in large language models and teaching these models to code-switch more effectively for reasoning. First, we create the Code-Switched Reasoning (CoRe) corpus, consisting of 7k reasoning traces from 15 models, 18 languages, 10 scripts, and diverse reasoning domains, providing insights into potentially helpful code-switching behaviors. Then, we expand CoRe into the Extended Code-Switched Reasoning corpus (EnCoRe) with 40 carefully curated datasets for training and evaluating six interventions on code-switching in reasoning across three models and seven languages, totaling 120 fine-tuning conditions. Across 80k+ reasoning traces from both language/culture-agnostic and -specific evaluations, English-dominated reasoning, semantically accurate code-switching, and a more even mix of languages are positively associated with correct answers, whereas dense switching is associated with incorrect answers. Moreover, we are able to elicit positive behaviors through fine-tuning tasks that do not directly demonstrate code-switching. Our work suggests that small but well-curated datasets can change how reasoning models code-switch, allowing us to reap the benefits of reasoning in the many languages that lack large-scale reasoning data.
Comments: 47 pages, 13 figures; revised with added results and analysis
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2604.15490 [cs.CL]
  (or arXiv:2604.15490v4 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2604.15490

arXiv-issued DOI via DataCite

Submission history

From: Eleanor Lin [view email]
[v1] Thu, 16 Apr 2026 19:44:21 UTC (242 KB)
[v2] Mon, 28 Sep 2026 19:12:27 UTC (523 KB)
[v3] Sat, 3 Oct 2026 01:47:30 UTC (523 KB)
[v4] Thu, 19 Nov 2026 02:02:30 UTC (1,773 KB)

来源:arXiv cs.CL · arxiv.org