@inproceedings{baruah-2026-abaruah,
title = "{ABARUAH} at {S}em{E}val-2026 Task 9: Multilingual Polarization Detection across Seven {I}ndic Languages using Qwen3",
author = "Baruah, Arup",
editor = "Kochmar, Ekaterina and
Ghosh, Debanjan and
North, Kai and
Komachi, Mamoru",
booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
month = jul,
year = "2026",
address = "San Diego, California, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.semeval-1.377/",
pages = "3002--3009",
ISBN = "979-8-89176-414-9",
abstract = "Online polarization creates division within the society. As such, it is important to detect and remove polarized messages from social media. This study presents fine-tuned Qwen3-8B Large Language Model (LLM) based models to identify online polarization, its specific categories, and its manifestation types. This study used Quantized Low-Rank Adaptation (QLoRA) to fine-tune the model in seven Indic languages: Bengali, Hindi, Nepali, Oriya, Punjabi, Telugu, and Urdu. The experimental results demonstrate the efficacy of this approach, achieving macro F1-scores of 0.82, 0.78, 0.90, 0.76, 0.78, 0.87, and 0.79, respectively, for polarization detection. The proposed model surpassed the established baseline systems in several of the subtasks, suggesting that parameter-efficient fine-tuning is a viable and powerful strategy for addressing linguistic diversity in low-resource and high-variability Indic language datasets. To leverage cross-lingual transfer, a unified model was developed by fine-tuning on a concatenated dataset of seven Indic languages. This approach proved superior to standalone language-specific models, yielding substantial improvements in F1-score (most notably a 28.76 point gain in Subtask 2 for Punjabi language). This provides strong evidence for the benefits of cross-lingual knowledge transfer in low-resource settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="baruah-2026-abaruah">
<titleInfo>
<title>ABARUAH at SemEval-2026 Task 9: Multilingual Polarization Detection across Seven Indic Languages using Qwen3</title>
</titleInfo>
<name type="personal">
<namePart type="given">Arup</namePart>
<namePart type="family">Baruah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 20th International Workshop on Semantic Evaluation (2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Kochmar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Debanjan</namePart>
<namePart type="family">Ghosh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai</namePart>
<namePart type="family">North</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mamoru</namePart>
<namePart type="family">Komachi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-414-9</identifier>
</relatedItem>
<abstract>Online polarization creates division within the society. As such, it is important to detect and remove polarized messages from social media. This study presents fine-tuned Qwen3-8B Large Language Model (LLM) based models to identify online polarization, its specific categories, and its manifestation types. This study used Quantized Low-Rank Adaptation (QLoRA) to fine-tune the model in seven Indic languages: Bengali, Hindi, Nepali, Oriya, Punjabi, Telugu, and Urdu. The experimental results demonstrate the efficacy of this approach, achieving macro F1-scores of 0.82, 0.78, 0.90, 0.76, 0.78, 0.87, and 0.79, respectively, for polarization detection. The proposed model surpassed the established baseline systems in several of the subtasks, suggesting that parameter-efficient fine-tuning is a viable and powerful strategy for addressing linguistic diversity in low-resource and high-variability Indic language datasets. To leverage cross-lingual transfer, a unified model was developed by fine-tuning on a concatenated dataset of seven Indic languages. This approach proved superior to standalone language-specific models, yielding substantial improvements in F1-score (most notably a 28.76 point gain in Subtask 2 for Punjabi language). This provides strong evidence for the benefits of cross-lingual knowledge transfer in low-resource settings.</abstract>
<identifier type="citekey">baruah-2026-abaruah</identifier>
<location>
<url>https://aclanthology.org/2026.semeval-1.377/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>3002</start>
<end>3009</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ABARUAH at SemEval-2026 Task 9: Multilingual Polarization Detection across Seven Indic Languages using Qwen3
%A Baruah, Arup
%Y Kochmar, Ekaterina
%Y Ghosh, Debanjan
%Y North, Kai
%Y Komachi, Mamoru
%S Proceedings of the 20th International Workshop on Semantic Evaluation (2026)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, USA
%@ 979-8-89176-414-9
%F baruah-2026-abaruah
%X Online polarization creates division within the society. As such, it is important to detect and remove polarized messages from social media. This study presents fine-tuned Qwen3-8B Large Language Model (LLM) based models to identify online polarization, its specific categories, and its manifestation types. This study used Quantized Low-Rank Adaptation (QLoRA) to fine-tune the model in seven Indic languages: Bengali, Hindi, Nepali, Oriya, Punjabi, Telugu, and Urdu. The experimental results demonstrate the efficacy of this approach, achieving macro F1-scores of 0.82, 0.78, 0.90, 0.76, 0.78, 0.87, and 0.79, respectively, for polarization detection. The proposed model surpassed the established baseline systems in several of the subtasks, suggesting that parameter-efficient fine-tuning is a viable and powerful strategy for addressing linguistic diversity in low-resource and high-variability Indic language datasets. To leverage cross-lingual transfer, a unified model was developed by fine-tuning on a concatenated dataset of seven Indic languages. This approach proved superior to standalone language-specific models, yielding substantial improvements in F1-score (most notably a 28.76 point gain in Subtask 2 for Punjabi language). This provides strong evidence for the benefits of cross-lingual knowledge transfer in low-resource settings.
%U https://aclanthology.org/2026.semeval-1.377/
%P 3002-3009
Markdown (Informal)
[ABARUAH at SemEval-2026 Task 9: Multilingual Polarization Detection across Seven Indic Languages using Qwen3](https://aclanthology.org/2026.semeval-1.377/) (Baruah, SemEval 2026)
ACL