MENLO: From Preferences to Proficiency -- Evaluating and Modeling Native-like Quality Across 47 Languages

Ensuring native-like quality of large language model (LLM) responses across many languages is challenging. Our evaluation reveals that zero-shot LLM judges benefit significantly from pairwise evaluation and our structured annotation rubrics, yet they still underperform human annotators on our dataset. We demonstrate substantial improvements through fine-tuning with reinforcement learning, reward shaping, and multi-task learning approaches. Additionally, we show that RL-trained judges can serve as generative reward models to enhance LLMs' multilingual proficiency, though discrepancies with human judgment remain. Our findings suggest promising directions for scalable multilingual evaluation and preference alignment. We release our dataset and evaluation framework to support further research in multilingual LLM evaluation.Dataset https://huggingface.co/datasets/facebook/menlo In order for LLMs to be most useful across the globe, they need to be able to provide high-quality responses in many languages. Responses should be relevant (Zhuang et al., 2024), factually accurate (Jacovi et al., 2025), and natural (Marchisio et al., 2024; Guo et al., 2025), among other considerations. Ultimately, for interaction in any language to be seamless, responses need to be indistinguishable from those of a native speaker (Novikova et al., 2016; Liu et al., 2021). Language proficiency in humans has traditionally been evaluated via standardized tests (Jamieson et al., 2000). While such tests have been applied to evaluating LLMs (Anil et al., 2023; Mayor-Rocher et al., 2024; Lothritz & Cabot, 2025), they are difficult to scale and do not readily correspond to real-world conversations.