跨语言迁移中语言相关性与任务对齐的解耦
Disentangling Linguistic Relatedness from Task Alignment in Cross-Lingual Transfer
AI总结：
通过微调大语言模型并在闪语族与非闪语族语言上评估零样本阅读理解，发现跨语言迁移主要提升任务格式对齐而非语言特定知识。
摘要：
我们通过微调七个大语言模型（4B--671B参数）在阿拉伯语上，并在闪语族语言和非闪语族对照语言上评估零样本阅读理解，研究跨语言迁移。在密集架构和混合专家架构中，我们没有发现闪语族特定迁移的证据：基线较弱的模型在所有语言上都有显著提升，而基线较强的模型无论语言族系如何，只有边际提升。思维链消融实验强化了这一发现——从微调中获益最多的模型同样从推理时推理中获益，这表明两种机制都解决了任务格式对齐问题，而非跨语言知识迁移。
英文摘要：
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.
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。
