arXiv:2609.27758v1 Announce Type: cross
Abstract: Safety alignment in large language models is trained primarily in English, and recent work reports that the underlying harmfulness representation survives translation: English-trained probes separate harmful from harmless prompts almost as well in low-resource languages as in English. This has been taken as evidence that cross-lingual refusal failures mainly reflect calibration rather than representation quality. We show that this conclusion depends on the choice of negative examples. Across nine languages spanning three resource tiers, we replicate near-perfect transfer (AUROC > 0.98) when harmless prompts come from an unrelated distribution (easy negatives). With XSTest contrast prompts, which are benign but surface-similar to harmful requests (hard negatives), transfer collapses in low-resource languages while remaining largely stable in high-resource languages. On Qwen2.5-7B-Instruct, mean AUROC drop increases from 0.003 in English to 0.017 in high-resource, 0.042 in mid-resource, and 0.276 in low-resource languages. The pattern replicates on Aya Expanse. Back-translation chrF controls and a matched-chrF comparison across three languages reduce the likelihood that translation quality explains the effect. The collapse remains after controlling for chrF (partial r = 0.70, p = 0.03). Tokenizer fertility correlates with the collapse and explains part of the resource-tier effect, but not all of it. The results show that easy-negative transfer can coexist with substantial degradation under hard negatives. Easy-negative evaluation alone therefore cannot establish that the harmfulness representation survives translation.
Abstract: Safety alignment in large language models is trained primarily in English, and recent work reports that the underlying harmfulness representation survives translation: English-trained probes separate harmful from harmless prompts almost as well in low-resource languages as in English. This has been taken as evidence that cross-lingual refusal failures mainly reflect calibration rather than representation quality. We show that this conclusion depends on the choice of negative examples. Across nine languages spanning three resource tiers, we replicate near-perfect transfer (AUROC > 0.98) when harmless prompts come from an unrelated distribution (easy negatives). With XSTest contrast prompts, which are benign but surface-similar to harmful requests (hard negatives), transfer collapses in low-resource languages while remaining largely stable in high-resource languages. On Qwen2.5-7B-Instruct, mean AUROC drop increases from 0.003 in English to 0.017 in high-resource, 0.042 in mid-resource, and 0.276 in low-resource languages. The pattern replicates on Aya Expanse. Back-translation chrF controls and a matched-chrF comparison across three languages reduce the likelihood that translation quality explains the effect. The collapse remains after controlling for chrF (partial r = 0.70, p = 0.03). Tokenizer fertility correlates with the collapse and explains part of the resource-tier effect, but not all of it. The results show that easy-negative transfer can coexist with substantial degradation under hard negatives. Easy-negative evaluation alone therefore cannot establish that the harmfulness representation survives translation.
This article has been indexed from cs.AI updates on arXiv.org
Read the original article: