Researchers propose Cross-Lingual Ranking Preference Optimization (CRPO), a framework that uses English-centric data to improve language model performance in other languages. CRPO optimizes both intra- and inter-lingual preferences through a hierarchical structure of parallel preference pairs across languages. The authors claim their method outperforms standard approaches in instruction-following and knowledge utilization capability, with consistent gains observed across various weighting schemes. --- Why it matters: This matters to AI engineers because it addresses the issue of English-centric data dominating language model performance, which can lead to suboptimal results in other languages. CRPO's hierarchical design and relative ranking signal could improve language adaptation and output quality for multilingual applications. Source: https://arxiv.org/abs/2608.23149