@inproceedings{saud-etal-2026-gabu,
title = "{GABU}@{WAT} 2026: Zero Shot and {QL}o{RA} Adaptation of Gemma 4 {E}4{B} for Multimodal {E}nglish to {I}ndic Translation",
author = "Saud, Shiv Ram and
Mahato, Aashish and
Ghimire, Prajwal and
Dawadi, Sundeep and
Saud, Ramesh and
Regmi, Sunil",
editor = "Nakazawa, Toshiaki and
Goto, Isao",
booktitle = "Proceedings of the Thirteenth Workshop on {A}sian Translation ({WAT} 2026)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wat-1.10/",
pages = "145--154",
ISBN = "979-8-89176-446-0",
abstract = "We describe \textit{GABU}, our WAT2026 English-to-Indic multimodal translation system (CITATION), which translates image grounded English captions into Hindi, Bengali, Malayalam, and Odia. The task follows the WAT English-to-Indic multimodal track setup with text only, image captioning, and multimodal tracks (CITATION). Our system builds on the pretrained, instruction-tuned vision-language model \texttt{google/gemma-4-E4B-it}. Each input combines a cropped visual region with the English caption, rendered through a chat template with a script constrained translation instruction. We explore two approaches which includes a zero shot baseline and a fine tuned approach of the model using QLORA (rank 64, alpha 128, 4-bit NF4, double quantization), trained for 3,000 steps due to computational resource constraints, reflecting our aim to adapt the model{'}s existing instruction-following ability rather than train it from scratch. We evaluate both systems using BLEU and CHRF3 under greedy decoding and report macro averaged scores across the target languages. The fine-tuned model substantially outperforms the zero shot baseline and achieves a BLEU score of 43.97 and 69.99 CHRF3 on the test split, and 40.37 BLEU and 68.25 CHRF3 on the challenge split. These results correspond to improvements of 24.39 BLEU and 18.72 CHRF3 points on the test split, and 16.70 BLEU and 11.90 CHRF3 points on the challenge split."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="saud-etal-2026-gabu">
<titleInfo>
<title>GABU@WAT 2026: Zero Shot and QLoRA Adaptation of Gemma 4 E4B for Multimodal English to Indic Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shiv</namePart>
<namePart type="given">Ram</namePart>
<namePart type="family">Saud</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aashish</namePart>
<namePart type="family">Mahato</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Prajwal</namePart>
<namePart type="family">Ghimire</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sundeep</namePart>
<namePart type="family">Dawadi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ramesh</namePart>
<namePart type="family">Saud</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunil</namePart>
<namePart type="family">Regmi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Thirteenth Workshop on Asian Translation (WAT 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Toshiaki</namePart>
<namePart type="family">Nakazawa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Isao</namePart>
<namePart type="family">Goto</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-446-0</identifier>
</relatedItem>
<abstract>We describe GABU, our WAT2026 English-to-Indic multimodal translation system (CITATION), which translates image grounded English captions into Hindi, Bengali, Malayalam, and Odia. The task follows the WAT English-to-Indic multimodal track setup with text only, image captioning, and multimodal tracks (CITATION). Our system builds on the pretrained, instruction-tuned vision-language model google/gemma-4-E4B-it. Each input combines a cropped visual region with the English caption, rendered through a chat template with a script constrained translation instruction. We explore two approaches which includes a zero shot baseline and a fine tuned approach of the model using QLORA (rank 64, alpha 128, 4-bit NF4, double quantization), trained for 3,000 steps due to computational resource constraints, reflecting our aim to adapt the model’s existing instruction-following ability rather than train it from scratch. We evaluate both systems using BLEU and CHRF3 under greedy decoding and report macro averaged scores across the target languages. The fine-tuned model substantially outperforms the zero shot baseline and achieves a BLEU score of 43.97 and 69.99 CHRF3 on the test split, and 40.37 BLEU and 68.25 CHRF3 on the challenge split. These results correspond to improvements of 24.39 BLEU and 18.72 CHRF3 points on the test split, and 16.70 BLEU and 11.90 CHRF3 points on the challenge split.</abstract>
<identifier type="citekey">saud-etal-2026-gabu</identifier>
<location>
<url>https://aclanthology.org/2026.wat-1.10/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>145</start>
<end>154</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T GABU@WAT 2026: Zero Shot and QLoRA Adaptation of Gemma 4 E4B for Multimodal English to Indic Translation
%A Saud, Shiv Ram
%A Mahato, Aashish
%A Ghimire, Prajwal
%A Dawadi, Sundeep
%A Saud, Ramesh
%A Regmi, Sunil
%Y Nakazawa, Toshiaki
%Y Goto, Isao
%S Proceedings of the Thirteenth Workshop on Asian Translation (WAT 2026)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-446-0
%F saud-etal-2026-gabu
%X We describe GABU, our WAT2026 English-to-Indic multimodal translation system (CITATION), which translates image grounded English captions into Hindi, Bengali, Malayalam, and Odia. The task follows the WAT English-to-Indic multimodal track setup with text only, image captioning, and multimodal tracks (CITATION). Our system builds on the pretrained, instruction-tuned vision-language model google/gemma-4-E4B-it. Each input combines a cropped visual region with the English caption, rendered through a chat template with a script constrained translation instruction. We explore two approaches which includes a zero shot baseline and a fine tuned approach of the model using QLORA (rank 64, alpha 128, 4-bit NF4, double quantization), trained for 3,000 steps due to computational resource constraints, reflecting our aim to adapt the model’s existing instruction-following ability rather than train it from scratch. We evaluate both systems using BLEU and CHRF3 under greedy decoding and report macro averaged scores across the target languages. The fine-tuned model substantially outperforms the zero shot baseline and achieves a BLEU score of 43.97 and 69.99 CHRF3 on the test split, and 40.37 BLEU and 68.25 CHRF3 on the challenge split. These results correspond to improvements of 24.39 BLEU and 18.72 CHRF3 points on the test split, and 16.70 BLEU and 11.90 CHRF3 points on the challenge split.
%U https://aclanthology.org/2026.wat-1.10/
%P 145-154
Markdown (Informal)
[GABU@WAT 2026: Zero Shot and QLoRA Adaptation of Gemma 4 E4B for Multimodal English to Indic Translation](https://aclanthology.org/2026.wat-1.10/) (Saud et al., WAT 2026)
ACL