AI Paper Atlas

Research map / Large language models / Alignment and preferences

Fluent Alignment with Disfluent Judges: Post-training for lower-resource languages

David Samuel, Lilja Øvrelid, Erik Velldal, Andrey Kutuzov

ICLR 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 1 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
use on-policy RL instead of SFT on machine-translated dataNorMistral 11B · Norwegian fluency (human)pairwise win-rate: 67.5Improved
use on-policy RL instead of SFT on machine-translated dataNorMistral 11B · Norwegian fluency (automatic)fluency score: 85.7% -> 92.2%Improved
use a disfluent judge (Mistral Nemo 12B) rather than the most fluent judge Mistral Large for reward (ablation)on-policy RL · policy fluencyfluency of trained policy: 92.2 vs 94.2; Pearson correlation 0.067No clear effect
train the SFT stage for 2 epochs instead of 1 on English data (ablation)initial SFT before RL · RL fluencyfluency: 94.2 -> 93.2Hurt
train the SFT stage for 4 epochs instead of 1 on English data (ablation)initial SFT before RL · RL fluencyfluency: 94.2 -> 92.8Hurt
use translated Norwegian dataset instead of English for initial SFT (ablation)initial SFT before RL · RL fluencyfluency: 94.2 -> 91.0Hurt
translate with Tower-Plus 72.7B instead of OPUS Eng-Gem 0.1B (ablation)translated SFT · Norwegian fluencyfluency: 68.2 -> 85.7Improved
translate with Tower-Plus instead of MADLAD-400 (ablation)translated SFT · Norwegian fluencyfluency: 82.4 -> 85.7Improved
translate with Tower-Plus instead of Seed-X (ablation)translated SFT · Norwegian fluencyfluency: 73.4 -> 85.7Improved
continue RL training beyond first 50 stepson-policy RL with Mistral Large judge · fluency over trainingfluency score: 87.5% -> around 93%No 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