AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Multilingual Unlearning in LLMs: Transfer, Dynamics, and Reversibility

Chaoyi Xiang, Olga Ohrimenko, Benjamin I. P. Rubinstein, Lea Frermann

ICML 2026 · Large language models › Language, safety and interpretability · approach: Multilingual LLM data and culture · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
unlearn in a language sharing family and script with the query language (English to German)DPO unlearning · TOFU multilingual forget set, Qwen2.5-7BNLI score changeImproved
unlearn English and query a same-family language (Russian) instead of a distant one (Chinese)DPO unlearning · TOFU multilingual forget set, Qwen2.5-7BNLI score change: -9 ± 2 (RU) vs -4 ± 5 (CH)Improved
unlearn a low-coverage language (English in Turkish-fine-tuned model) to affect the fine-tuned languageDPO unlearning · FT Turkish, query Turkish, Qwen2.5-7BNLI score change: -55 ± 8Improved
use high-coverage language (English, Chinese) as the unlearning source rather than German, Russian or TurkishDPO unlearning · cross-lingual transfer, Qwen2.5-7Btransfer strengthImproved
instruct the model to answer in the fine-tuning language (cross-lingual prompting)Qwen2.5-7B unlearning evaluation · FT English, query GermanNLI score gain: +61Improved
unlearn with DPO on small forget setcross-lingual alignment of hidden states · Qwen2.5-7B, layer-wise cosine similaritycosine similarity early and middle layersNo clear effect
unlearn with DPOrepresentation of forget questions · PCA of hidden statesseparation from base modelMixed
inject extracted steering vector at inference (subtract scaled unlearning direction over layers l..l+2)unlearned Qwen2.5-7B · English forget setNLI score recovery: recovers over half of lost performanceImproved
inject steering vector extracted only from English dataunlearned Qwen2.5-7B · non-English forget setsNLI score recoveryImproved
inject isotropic Gaussian random vector instead of extracted steering vector (ablation)unlearned Qwen2.5-7B · forget setNLI score recoveryNo clear effect
use Gemma2-9B instead of Qwen2.5-7B for steering recoverysteering vector injection · forget setNLI score recovery: 50% (Qwen) -> 90% (Gemma) recoveredImproved
replace DPO with gradient ascent or NPO (ablation)multilingual unlearning · cross-lingual transfer and steeringtransfer patternNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy