@inproceedings{divya-sree-gupta-2026-nlp,
title = "{NLP}-{MT}{\_}{A}mrita: A {RL}-Inspired {GRPO} based Agentic {RAG} Framework for {M}anipuri and {B}odo",
author = "Divya Sree, Kuruva and
Gupta, Deepa",
editor = "Haddow, Barry and
Kocmi, Tom and
Koehn, Philipp and
Monz, Christof",
booktitle = "Proceedings of the Eleventh Conference on Machine Translation",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wmt-main.149/",
pages = "2101--2110",
ISBN = "979-8-89176-474-3",
abstract = "Low-resource Indic machine translation (MT) remains challenging due to limited parallel corpora, linguistic diversity, and contextual ambiguities that often result in inaccurate translations and hallucinations. This study proposes AgenticRAG-GRPO, an RL-inspired retrieval-augmented translation framework that combines semantic retrieval with GRPO-inspired reward-guided candidate selection for context-aware English{\ensuremath{\leftrightarrow}}Indic MT. The framework dynamically retrieves relevant translation contexts and ranks candidate translations using a reward function based on SacreBLEU and BERT score. Experiments were conducted on the WMT 2026 Low-Resource IndicMT Shared Task for English{\ensuremath{\leftrightarrow}}Manipuri (Mni) and English{\ensuremath{\leftrightarrow}}Bodo (Brx) using IndicTrans2, NLLB-200 Distilled 600M, and Sarvam-Translate. The proposed study achieved s performance, obtaining BLEU scores up to 32.36, METEOR up to 67.01, BERTScore up to 95.04, and COMET up to 80.51 across the languages, while improving semantic faithfulness and reducing hallucinations through retrieval-guided generation. The framework also maintained inference efficiency without requiring task-specific model fine-tuning. These results demonstrate that AgenticRAG-GRPO provides an effective and scalable solution for low-resource English-Indic MT."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="divya-sree-gupta-2026-nlp">
<titleInfo>
<title>NLP-MT_Amrita: A RL-Inspired GRPO based Agentic RAG Framework for Manipuri and Bodo</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kuruva</namePart>
<namePart type="family">Divya Sree</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Deepa</namePart>
<namePart type="family">Gupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Eleventh Conference on Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Barry</namePart>
<namePart type="family">Haddow</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Kocmi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philipp</namePart>
<namePart type="family">Koehn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christof</namePart>
<namePart type="family">Monz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-474-3</identifier>
</relatedItem>
<abstract>Low-resource Indic machine translation (MT) remains challenging due to limited parallel corpora, linguistic diversity, and contextual ambiguities that often result in inaccurate translations and hallucinations. This study proposes AgenticRAG-GRPO, an RL-inspired retrieval-augmented translation framework that combines semantic retrieval with GRPO-inspired reward-guided candidate selection for context-aware English\ensuremathłeftrightarrowIndic MT. The framework dynamically retrieves relevant translation contexts and ranks candidate translations using a reward function based on SacreBLEU and BERT score. Experiments were conducted on the WMT 2026 Low-Resource IndicMT Shared Task for English\ensuremathłeftrightarrowManipuri (Mni) and English\ensuremathłeftrightarrowBodo (Brx) using IndicTrans2, NLLB-200 Distilled 600M, and Sarvam-Translate. The proposed study achieved s performance, obtaining BLEU scores up to 32.36, METEOR up to 67.01, BERTScore up to 95.04, and COMET up to 80.51 across the languages, while improving semantic faithfulness and reducing hallucinations through retrieval-guided generation. The framework also maintained inference efficiency without requiring task-specific model fine-tuning. These results demonstrate that AgenticRAG-GRPO provides an effective and scalable solution for low-resource English-Indic MT.</abstract>
<identifier type="citekey">divya-sree-gupta-2026-nlp</identifier>
<location>
<url>https://aclanthology.org/2026.wmt-main.149/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2101</start>
<end>2110</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T NLP-MT_Amrita: A RL-Inspired GRPO based Agentic RAG Framework for Manipuri and Bodo
%A Divya Sree, Kuruva
%A Gupta, Deepa
%Y Haddow, Barry
%Y Kocmi, Tom
%Y Koehn, Philipp
%Y Monz, Christof
%S Proceedings of the Eleventh Conference on Machine Translation
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-474-3
%F divya-sree-gupta-2026-nlp
%X Low-resource Indic machine translation (MT) remains challenging due to limited parallel corpora, linguistic diversity, and contextual ambiguities that often result in inaccurate translations and hallucinations. This study proposes AgenticRAG-GRPO, an RL-inspired retrieval-augmented translation framework that combines semantic retrieval with GRPO-inspired reward-guided candidate selection for context-aware English\ensuremathłeftrightarrowIndic MT. The framework dynamically retrieves relevant translation contexts and ranks candidate translations using a reward function based on SacreBLEU and BERT score. Experiments were conducted on the WMT 2026 Low-Resource IndicMT Shared Task for English\ensuremathłeftrightarrowManipuri (Mni) and English\ensuremathłeftrightarrowBodo (Brx) using IndicTrans2, NLLB-200 Distilled 600M, and Sarvam-Translate. The proposed study achieved s performance, obtaining BLEU scores up to 32.36, METEOR up to 67.01, BERTScore up to 95.04, and COMET up to 80.51 across the languages, while improving semantic faithfulness and reducing hallucinations through retrieval-guided generation. The framework also maintained inference efficiency without requiring task-specific model fine-tuning. These results demonstrate that AgenticRAG-GRPO provides an effective and scalable solution for low-resource English-Indic MT.
%U https://aclanthology.org/2026.wmt-main.149/
%P 2101-2110
Markdown (Informal)
[NLP-MT_Amrita: A RL-Inspired GRPO based Agentic RAG Framework for Manipuri and Bodo](https://aclanthology.org/2026.wmt-main.149/) (Divya Sree & Gupta, WMT 2026)
ACL