Unverified50% confidenceDecision RuleExact time
The study recommends that evaluations record component language and the amount of translated text, compare matched behaviors at equal numbers of translated spans, and interpret refusal and unsafe compliance separately.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe study proposes a "paired reversal" method that separates the effects of component language identity from text quantity by controlling behavior, English framing, and the number of translated spans.77% similarUnverifiedThe effect of back-translation depends on the quality of the initial parallel corpus and the selection of monolingual data; how to systematically evaluate and filter monolingual data remains an open question.73% similarUnverifiedThe research team proposed a disagreement-guided steering method, which obtains a direction vector by comparing the representation difference between the documents retrieved by that language alone and the consensus documents of other languages, and adds it to that language's query vector at inference time.73% similarUnverifiedThe study used incremental explanatory power (ΔR²) as a stricter criterion, testing whether translated scores still explain additional variance in native-language performance after controlling for covariates such as model scale and language proficiency.72% similarUnverifiedThe exploratory results and differences between models indicate that component-level language analysis is worth incorporating into multilingual safety evaluation frameworks, but the null results of the main experiment contrast with the exploratory findings and require more research to verify.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114732API
curl https://kongchang.com/api/v1/knowledge/claims/1114732MCP
get_claim(id=1114732)