arXiv cs.AI· Max Neuwinger, Hanna Yukhymenko, Imanol Schlag·· 14 小时前AI 评分31
用模板扩展多语言安全训练：MSafeTemp 覆盖 60 种语言的拒绝与良性语料
How I Aligned Your Model: Scaling Up Multilingual Safety Training with Templates
AI 导读
研究者提出基于模板的可扩展增强流水线，从有害提示词中提取安全模板、翻译后用文化本地化实体填充槽位，构建出覆盖 60 种语言的 MSafeTemp 语料库，用于多语言安全对齐与评测。实验显示仅用英文评测会误判多语言安全性：部分模型在英文下失败的攻击翻译后即成功，文化本地化变体还暴露出翻译无法覆盖的额外失败。用模板增强安全训练数据可提升英文之外的安全表现。
正文 · 原文
Abstract:Safety alignment for Large Language Models is heavily focused on English with limited multilingual representation, particularly damaging low-resource language performance. Related open training datasets are often limited in size or do not follow region-localized harmful topics to broaden the scope of alignment beyond cultures. We propose a scalable template-based augmentation pipeline to produce massive multilingual refusal and benign corpora, extracting safe templates from harmful prompts, translating them and then filling in the slots with culturally authentic entities, both harmful and benign. As a result, we release MSafeTemp, a validated corpus spanning 60 languages useful for multilingual safety alignment and evaluation. Using the dataset, we show that English-only evaluation misjudges multilingual safety: for several models, attacks that fail in English succeed once translated, while culturally localized variants expose further failures that translation alone misses. We also confirm that augmenting safety training data through templates allows for higher diversity and quality of alignment, resulting in improved safety performance beyond English. MSafeTemp and the respective pipeline are aimed at facilitating research and development of safer and more responsible models for less-represented cultures.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.40516 [cs.CL] |
| (or arXiv:2610.40516v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.40516 arXiv-issued DOI via DataCite |
Submission history
From: Hanna Yukhymenko [view email]
[v1]
Thu, 29 Oct 2026 17:26:02 UTC (1,005 KB)
来源：arXiv cs.AI · arxiv.org