跨语言迁移中，语言相似性不决定效果，任务对齐才是关键。
Disentangling Linguistic Relatedness from Task Alignment in Cross-Lingual Transfer
- 用阿拉伯语微调大模型，测试跨语言阅读理解能力。
- 无论是否为闪米特语，强基线模型提升均有限。
- 推理机制与微调效果相似，说明核心是任务适配而非语言知识迁移。
我们通过在阿拉伯语上微调七个大语言模型（参数量4B–671B），评估其在闪米特语和非闪米特语上的零样本阅读理解能力。在密集型和专家混合（Mixture-of-Experts）架构下，均未发现闪米特语特异性迁移现象：基线弱的模型在所有语言上表现显著提升，而基线强的模型仅获得微小增益，且不受语言家族影响。链式思维消融实验进一步表明，从微调中获益最多的模型，在推理时使用思维链也获得同等提升，暗示两者均解决任务格式对齐问题，而非跨语言知识传递。
原文摘要 · Abstract (English)
We study cross-lingual transfer by fine-tuning seven large language models (4B--671B parameters) on Arabic and evaluating zero-shot reading comprehension on Semitic languages and non-Semitic controls. Across dense and Mixture-of-Experts architectures, we find no evidence of Semitic-specific transfer: models with weak baselines improve dramatically across all languages, while strong-baseline models show only marginal gains regardless of language family. A chain-of-thought ablation reinforces this finding -- the same models that benefit most from fine-tuning benefit equally from inference-time reasoning, suggesting both mechanisms address task-format alignment rather than cross-lingual knowledge transfer.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。