AI Paper Atlas

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

Mitigating Translationese Bias in Multilingual LLM-as-a-Judge via Disentangled Information Bottleneck

Hongbin Zhang, Kehai Chen, Xuefeng Bai, Youcheng Pan, Yang Xiang, Jinpeng Wang, Min Zhang

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
add bias-capture and disentanglement terms to compression-only objective (ablation)DIBJudge · DIB objective ablationbias severity S_bias: 0.124 -> 0.031Improved
add disentanglement term to compression-only objective (ablation)DIBJudge · DIB objective ablationaccuracy: 85.25 -> 88.55Improved
add compression term to bias+disentangle (ablation)DIBJudge · DIB objective ablationaccuracy: 89.10 -> 89.85Improved
add bias-capture term to compression+disentangle (ablation)DIBJudge · DIB objective ablationaccuracy: 88.55 -> 89.85Improved
add disentanglement term to compression+bias (ablation)DIBJudge · DIB objective ablationbias severity S_bias: 0.091 -> 0.031Improved
add both proxy tasks CLA and LPBC (ablation)DIBJudge · Proxy task ablationbias severity S_bias: 0.421 -> 0.147Improved
add both proxy tasks CLA and LPBC (ablation)DIBJudge · Proxy task ablationaccuracy: 87.12 -> 89.18Improved
add LPBC proxy task alone (ablation)DIBJudge · Proxy task ablationbias severity S_bias: 0.421 -> 0.279Improved
add CLA proxy task alone (ablation)DIBJudge · Proxy task ablationbias severity S_bias: 0.421 -> 0.312Improved
replace vanilla SFT with DIBJudgeQwen3 judge · Skywork held-out, length biasSpearman rho: 0.553 -> 0.314Improved
replace vanilla IB with DIBJudgeQwen3 judge · Skywork held-out, self-preferencebias severity S_bias: 0.276 -> 0.219Improved
replace vanilla IB with DIBJudgeQwen3 judge · Translationese biasbias severity S_bias: 0.168 -> 0.083Improved

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