Lost in Translation: Do LVLM Judges Generalize Across Languages?

TL;DR

This paper introduces MM-JudgeBench, a large-scale multilingual benchmark for evaluating vision-language model judges across diverse languages, revealing significant cross-lingual performance variability and limitations of current reward models.

Contribution

The paper presents MM-JudgeBench, the first comprehensive multilingual benchmark for LVLM judges, and provides an extensive analysis of their cross-lingual robustness and limitations.

Findings

Substantial performance variance across languages in LVLM judges.

Model size and architecture poorly predict multilingual robustness.

State-of-the-art judges show inconsistent behavior across languages.

Abstract

Automatic evaluators such as reward models play a central role in the alignment and evaluation of large vision-language models (LVLMs). Despite their growing importance, these evaluators are almost exclusively assessed on English-centric benchmarks, leaving open the question of how well these evaluators generalize across languages. To answer this question, we introduce MM-JudgeBench, the first large-scale benchmark for multilingual and multimodal judge model evaluation, which includes over 60K pairwise preference instances spanning 25 typologically diverse languages. MM-JudgeBench integrates two complementary subsets: a general vision-language preference evaluation subset extending VL-RewardBench, and a chart-centric visual-text reasoning subset derived from OpenCQA, enabling systematic analysis of reward models (i.e., LVLM judges) across diverse settings. We additionally release a multilingual training set derived from MM-RewardBench, disjoint from our evaluation data, to support domain adaptation. By evaluating 22 LVLMs (15 open-source, 7 proprietary), we uncover substantial cross-lingual performance variance in our proposed benchmark. Our analysis further shows that model size and architecture are poor predictors of multilingual robustness, and that even state-of-the-art LVLM judges exhibit inconsistent behavior across languages. Together, these findings expose fundamental limitations of current reward modeling and underscore the necessity of multilingual, multimodal benchmarks for developing reliable automated evaluators.