CLFeb 16, 2025

Investigating Language Preference of Multilingual RAG Systems

arXiv:2502.11175v416 citationsh-index: 4Has CodeACL
Originality Incremental advance
AI Analysis

This addresses inconsistent outputs in multilingual AI systems for users handling cross-lingual information, representing an incremental improvement.

The paper tackles the problem of language preference in multilingual Retrieval-Augmented Generation (mRAG) systems, which causes inconsistent responses due to linguistic variations and conflicts in multilingual sources. The result is DKM-RAG, a framework that fuses translated passages with model knowledge, mitigating language preference and enhancing performance across diverse linguistic settings.

Multilingual Retrieval-Augmented Generation (mRAG) systems enhance language models by integrating external multilingual information to produce context-aware responses. However, mRAG systems struggle with retrieving relevant information due to linguistic variations between queries and documents, generating inconsistent responses when multilingual sources conflict. In this work, we systematically investigate language preferences in both retrieval and generation of mRAG through a series of experiments. Our analysis indicates that retrievers tend to prefer high-resource and query languages, yet this preference does not consistently improve generation performance. Moreover, we observe that generators prefer the query language or Latin scripts, leading to inconsistent outputs. To overcome these issues, we propose Dual Knowledge Multilingual RAG (DKM-RAG), a simple yet effective framework that fuses translated multilingual passages with complementary model knowledge. Empirical results demonstrate that DKM-RAG mitigates language preference in generation and enhances performance across diverse linguistic settings. Code is available at https://github.com/jeonghyunpark2002/LanguagePreference.git

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes