@inproceedings{omotoso-etal-2025-improving,
title = "Improving {BGE}-{M}3 Multilingual Dense Embeddings for {N}igerian Low Resource Languages",
author = "Omotoso, Abdulmatin and
Shopeju, Habeeb and
Joshua, Adejumobi Monjolaoluwa and
Oni, Shiloh",
editor = "Zhang, Chen and
Allaway, Emily and
Shen, Hua and
Miculicich, Lesly and
Li, Yinqiao and
M'hamdi, Meryem and
Limkonchotiwat, Peerat and
Bai, Richard He and
T.y.s.s., Santosh and
Han, Sophia Simeng and
Thapa, Surendrabikram and
Rim, Wiem Ben",
booktitle = "Proceedings of the 9th Widening NLP Workshop",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.winlp-main.33/",
pages = "224--229",
ISBN = "979-8-89176-351-7",
abstract = "Multilingual dense embedding models such as Multilingual E5, LaBSE, and BGE-M3 have shown promising results on diverse benchmarks for information retrieval in low-resource languages. But their result on low resource languages is not up to par with other high resource languages. This work improves the performance of BGE-M3 through contrastive fine-tuning; the model was selected because of its superior performance over other multilingual embedding models across MIRACL, MTEB, and SEB benchmarks. To fine-tune this model, we curated a comprehensive dataset comprising Yor{\`u}b{\'a} (32.9k rows), Igbo (18k rows) and Hausa (85k rows) from mainly news sources. We further augmented our multilingual dataset with English queries and mapped it to each of the Yoruba, Igbo, and Hausa documents, enabling cross-lingual semantic training. We evaluate on two settings: the Wura test set and the MIRACL benchmark. On Wura, the fine-tuned BGE-M3 raises mean reciprocal rank (MRR) to 0.9201 for Yor{\`u}b{\'a}, 0.8638 for Igbo, 0.9230 for Hausa, and 0.8617 for English queries matched to local documents, surpassing the BGE-M3 baselines of 0.7846, 0.7566, 0.8575, and 0.7377, respectively. On MIRACL (Yor{\`u}b{\'a} subset), the fine-tuned model attains 0.5996 MRR, slightly surpassing base BGE-M3 (0.5952) and outperforming ML-E5-large (0.5632) and LaBSE (0.4468)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="omotoso-etal-2025-improving">
<titleInfo>
<title>Improving BGE-M3 Multilingual Dense Embeddings for Nigerian Low Resource Languages</title>
</titleInfo>
<name type="personal">
<namePart type="given">Abdulmatin</namePart>
<namePart type="family">Omotoso</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Habeeb</namePart>
<namePart type="family">Shopeju</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Adejumobi</namePart>
<namePart type="given">Monjolaoluwa</namePart>
<namePart type="family">Joshua</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shiloh</namePart>
<namePart type="family">Oni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 9th Widening NLP Workshop</title>
</titleInfo>
<name type="personal">
<namePart type="given">Chen</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Emily</namePart>
<namePart type="family">Allaway</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hua</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lesly</namePart>
<namePart type="family">Miculicich</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yinqiao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Meryem</namePart>
<namePart type="family">M’hamdi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peerat</namePart>
<namePart type="family">Limkonchotiwat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Richard</namePart>
<namePart type="given">He</namePart>
<namePart type="family">Bai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Santosh</namePart>
<namePart type="family">T.y.s.s.</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sophia</namePart>
<namePart type="given">Simeng</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Surendrabikram</namePart>
<namePart type="family">Thapa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wiem</namePart>
<namePart type="given">Ben</namePart>
<namePart type="family">Rim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-351-7</identifier>
</relatedItem>
<abstract>Multilingual dense embedding models such as Multilingual E5, LaBSE, and BGE-M3 have shown promising results on diverse benchmarks for information retrieval in low-resource languages. But their result on low resource languages is not up to par with other high resource languages. This work improves the performance of BGE-M3 through contrastive fine-tuning; the model was selected because of its superior performance over other multilingual embedding models across MIRACL, MTEB, and SEB benchmarks. To fine-tune this model, we curated a comprehensive dataset comprising Yorùbá (32.9k rows), Igbo (18k rows) and Hausa (85k rows) from mainly news sources. We further augmented our multilingual dataset with English queries and mapped it to each of the Yoruba, Igbo, and Hausa documents, enabling cross-lingual semantic training. We evaluate on two settings: the Wura test set and the MIRACL benchmark. On Wura, the fine-tuned BGE-M3 raises mean reciprocal rank (MRR) to 0.9201 for Yorùbá, 0.8638 for Igbo, 0.9230 for Hausa, and 0.8617 for English queries matched to local documents, surpassing the BGE-M3 baselines of 0.7846, 0.7566, 0.8575, and 0.7377, respectively. On MIRACL (Yorùbá subset), the fine-tuned model attains 0.5996 MRR, slightly surpassing base BGE-M3 (0.5952) and outperforming ML-E5-large (0.5632) and LaBSE (0.4468).</abstract>
<identifier type="citekey">omotoso-etal-2025-improving</identifier>
<location>
<url>https://aclanthology.org/2025.winlp-main.33/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>224</start>
<end>229</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Improving BGE-M3 Multilingual Dense Embeddings for Nigerian Low Resource Languages
%A Omotoso, Abdulmatin
%A Shopeju, Habeeb
%A Joshua, Adejumobi Monjolaoluwa
%A Oni, Shiloh
%Y Zhang, Chen
%Y Allaway, Emily
%Y Shen, Hua
%Y Miculicich, Lesly
%Y Li, Yinqiao
%Y M’hamdi, Meryem
%Y Limkonchotiwat, Peerat
%Y Bai, Richard He
%Y T.y.s.s., Santosh
%Y Han, Sophia Simeng
%Y Thapa, Surendrabikram
%Y Rim, Wiem Ben
%S Proceedings of the 9th Widening NLP Workshop
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-351-7
%F omotoso-etal-2025-improving
%X Multilingual dense embedding models such as Multilingual E5, LaBSE, and BGE-M3 have shown promising results on diverse benchmarks for information retrieval in low-resource languages. But their result on low resource languages is not up to par with other high resource languages. This work improves the performance of BGE-M3 through contrastive fine-tuning; the model was selected because of its superior performance over other multilingual embedding models across MIRACL, MTEB, and SEB benchmarks. To fine-tune this model, we curated a comprehensive dataset comprising Yorùbá (32.9k rows), Igbo (18k rows) and Hausa (85k rows) from mainly news sources. We further augmented our multilingual dataset with English queries and mapped it to each of the Yoruba, Igbo, and Hausa documents, enabling cross-lingual semantic training. We evaluate on two settings: the Wura test set and the MIRACL benchmark. On Wura, the fine-tuned BGE-M3 raises mean reciprocal rank (MRR) to 0.9201 for Yorùbá, 0.8638 for Igbo, 0.9230 for Hausa, and 0.8617 for English queries matched to local documents, surpassing the BGE-M3 baselines of 0.7846, 0.7566, 0.8575, and 0.7377, respectively. On MIRACL (Yorùbá subset), the fine-tuned model attains 0.5996 MRR, slightly surpassing base BGE-M3 (0.5952) and outperforming ML-E5-large (0.5632) and LaBSE (0.4468).
%U https://aclanthology.org/2025.winlp-main.33/
%P 224-229
Markdown (Informal)
[Improving BGE-M3 Multilingual Dense Embeddings for Nigerian Low Resource Languages](https://aclanthology.org/2025.winlp-main.33/) (Omotoso et al., WiNLP 2025)
ACL