LLM Safety Alignment in Low-Resource Languages: A Systematic Literature Review
LLM Safety Alignment in Low-Resource Languages: A Systematic Literature Review
低资源语言中的大语言模型安全对齐：一项系统性文献综述
Large Language Models (LLMs) have achieved substantial progress in safety alignment, yet their safety guarantees remain significantly weaker in low-resource and multilingual settings than in high-resource languages. 大语言模型（LLMs）在安全对齐方面已经取得了实质性进展，然而在低资源和多语言环境下，其安全保障能力仍显著弱于高资源语言。
In this paper, we conduct a Systematic Literature Review (SLR) of LLM safety alignment in low-resource languages by adopting the PRISMA 2020 methodology. Out of roughly 1,500 papers identified from Semantic Scholar, arXiv, and OpenAlex, 50 relevant studies have been selected and analyzed. 在本文中，我们采用 PRISMA 2020 方法论，对低资源语言中的大语言模型安全对齐进行了系统性文献综述（SLR）。在从 Semantic Scholar、arXiv 和 OpenAlex 检索到的约 1,500 篇论文中，我们筛选并分析了 50 篇相关研究。
Our review is organized around four themes: safety alignment methods, multilingual safety risks, evaluation benchmarks, and cross-lingual transferability. We further propose a taxonomy of safety alignment approaches based on three adaptation mechanisms: data adaptation, objective optimization, and mechanistic alignment. 我们的综述围绕四个主题展开：安全对齐方法、多语言安全风险、评估基准以及跨语言迁移能力。此外，我们基于数据适配、目标优化和机制对齐这三种适配机制，提出了一套安全对齐方法的分类体系。
Across literature, translated English benchmarks fail to sufficiently represent culturally rooted harms, and multilingual models are more vulnerable to cross-lingual jailbreaks, code-switching attacks, and safety degradation in underrepresented languages. 纵观现有文献，翻译后的英语基准测试无法充分反映根植于特定文化的危害；同时，多语言模型在面对跨语言越狱、语码转换攻击以及代表性不足语言中的安全性退化时，显得更为脆弱。
These failures are driven by several key factors, including uneven multilingual pre-training coverage, insufficient native-language preference data, poor transfer of safety representations, and a lack of culturally aware evaluation frameworks. 这些缺陷由多个关键因素导致，包括多语言预训练覆盖不均衡、母语偏好数据不足、安全表征迁移效果差，以及缺乏具备文化意识的评估框架。
The review also notes that many low-resource languages, especially African languages, have fewer safety benchmarks available than other multilingual regions. Overall, the results reveal a persistent multilingual safety gap, and suggest that future progress will require culturally grounded benchmarks, participatory data collection, balanced multilingual pre-training, and scalable multilingual alignment methods. 综述还指出，许多低资源语言（尤其是非洲语言）可用的安全基准测试远少于其他多语言地区。总体而言，研究结果揭示了持续存在的多语言安全鸿沟，并表明未来的进展需要依赖于基于文化的基准测试、参与式数据收集、均衡的多语言预训练以及可扩展的多语言对齐方法。