When Does Mixing Help? Analyzing Query Embedding Interpolation in Multilingual Dense Retrieval

Abstract

This paper presents a ratio-controlled study of mixed-language querying on mMARCO, building mixed queries as interpolations of monolingual embeddings. With BGE-M3, an optimal mixing ratio outperforms the best monolingual endpoint in 88 of 105 cases, and the gains show a clear asymmetry driven by English dominance: mixing helps for non-English document indices, while indices containing English are best served by pure English queries.

Tongyao Zhu

IPP Doctoral Student (Jan ‘23; SEA)

PhD Candidate January 2023 Intake

Chao Ming Huang

FYP Alumnus (Apr ‘26). Thesis: Evaluating and Enhancing Information Retrieval in Code Mixed Contexts
OCP Awardee

FYP Student

Min-Yen Kan

Associate Professor

WING lead; interests include Digital Libraries, Information Retrieval and Natural Language Processing.