LAMAR: An Open Language-Aware Multilingual Alignment Reranker

2026-07-24

AI Summary

This work addresses a critical limitation in existing cross-lingual re-ranking models, which often fail to prioritize query-language documents even when semantically equivalent candidates exist, thereby degrading generation quality. To remedy this, the authors propose a multilingual re-ranking approach that jointly optimizes semantic relevance and language consistency within a multilingual cross-encoder architecture. Their method explicitly models a preference for language alignment through a novel strategy combining English-anchored relevance distillation with language-preference alignment. Experimental results demonstrate that the proposed model achieves state-of-the-art or highly competitive performance on standard multilingual re-ranking benchmarks as well as on tasks specifically designed to evaluate language consistency control, yielding significant improvements across multiple metrics in realistic retrieval scenarios.

Abstract

In multilingual retrieval augmented generation, a retriever can retrieve relevant documents written in multiple languages, which are subsequently reranked before answer generation. However, it remains unclear whether existing multilingual rerankers consider document language when ordering semantically relevant candidates. Our analysis shows that these rerankers do not consistently prioritize documents written in the same language as the query when semantically equivalent documents are available across languages, even though document language can affect answer generation. We release LAMAR, a language aware multilingual cross encoder trained to account for both semantic relevance and language coherence. LAMAR first uses English anchored relevance distillation to establish consistent relevance scoring across multilingual inputs and then applies preference alignment for language coherence to encourage documents written in the same language as the query to receive higher rankings while retaining semantic relevance. In a controlled experiment designed to assess language coherence, LAMAR achieves the best performance overall and across all languages examined individually. LAMAR also remains competitive on established multilingual reranking benchmarks. In practical retrieval settings, LAMAR achieves the best results across all reported metrics when reranking candidates retrieved in the first stage. These results demonstrate that LAMAR accounts for language coherence while achieving strong performance on general multilingual reranking benchmarks.

Problem

• multilingual reranking
• language coherence
• retrieval augmented generation
• semantic relevance
• cross-lingual alignment

Innovation

• language-aware reranking
• multilingual retrieval
• relevance distillation
• preference alignment
• cross-lingual coherence