拒绝方向在低资源语言中是否普遍存在？

arXiv cs.CL 论文
multilingual safety-alignment refusal-directions llm low-resource-languages interpretability

摘要

该论文在 Qwen2.5-14B-Instruct 上将 PolyRefuse 扩展到另外三种低资源语言（白俄罗斯语、巴什基尔语、塔吉克语），发现拒绝方向的迁移效果存在程度差异，并非普遍有效：来自安全对齐语言的方向能够广泛迁移，而来自严重未对齐语言的方向则无法可靠地调节拒绝行为。

arXiv:2610.28897v1 Announce Type: new Abstract: Refusal directions have been shown to transfer cross-lingually with near-perfect effectiveness across safety-aligned languages (Wang et al., 2025), reportedly regardless of resource level. However, the only truly low-resourced language evaluated is Yoruba, which is strongly misaligned and shows limited transfer, leaving open how refusal directions behave for other low-resourced languages. We extend PolyRefuse to three additional low-resourced languages (Belarusian, Bashkir, and Tajik) and replicate the pipeline of (Wang et al., 2025) on Qwen2.5-14B-Instruct. We find that universality is graded rather than binary: directions from safety-aligned and partially-aligned languages transfer broadly, while those from strongly misaligned languages fail to reliably modulate refusal behavior on any target tested, including other misaligned languages, once ceiling effects are accounted for. Geometric alignment and behavioral transfer are further shown to be partially dissociated. These findings complicate the universality claim and suggest that users of the least-resourced, most strongly misaligned languages may face weaker safety guarantees.
查看原文
查看缓存全文

缓存时间: 2026/10/22 16:21 

# Is Refusal Direction Universal across Low-Resourced Languages?
Source: [https://arxiv.org/abs/2610.28897](https://arxiv.org/abs/2610.28897)
[View PDF](https://arxiv.org/pdf/2610.28897)

> Abstract:Refusal directions have been shown to transfer cross\-lingually with near\-perfect effectiveness across safety\-aligned languages \(Wang et al\., 2025\), reportedly regardless of resource level\. However, the only truly low\-resourced language evaluated is Yoruba, which is strongly misaligned and shows limited transfer, leaving open how refusal directions behave for other low\-resourced languages\. We extend PolyRefuse to three additional low\-resourced languages \(Belarusian, Bashkir, and Tajik\) and replicate the pipeline of \(Wang et al\., 2025\) on Qwen2\.5\-14B\-Instruct\. We find that universality is graded rather than binary: directions from safety\-aligned and partially\-aligned languages transfer broadly, while those from strongly misaligned languages fail to reliably modulate refusal behavior on any target tested, including other misaligned languages, once ceiling effects are accounted for\. Geometric alignment and behavioral transfer are further shown to be partially dissociated\. These findings complicate the universality claim and suggest that users of the least\-resourced, most strongly misaligned languages may face weaker safety guarantees\.

## Submission history

From: Anastasiia Ivanova \[[view email](https://arxiv.org/show-email/4c91cdac/2610.28897)\] **\[v1\]**Tue, 20 Oct 2026 20:19:40 UTC \(1,603 KB\)

相似文章

When Refusal Looks Safe: The Refusal-Cue Shortcut in Safety Guard Models

arXiv cs.AI

This paper identifies a refusal-cue shortcut in safety guard models, where inserting refusal expressions into harmful responses can flip their harmless classification. The authors audit datasets like WildGuardMix and GR-Train, show the issue persists in official models such as LlamaGuard3 and Qwen3Guard, and propose a post-hoc intervention to suppress shortcut-associated components.