ABARUAH at SemEval-2026 Task 9: Multilingual Polarization Detection across Seven Indic Languages using Qwen3

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, medium

Summary

Arup Baruah's study, "ABARUAH at SemEval-2026 Task 9," introduces fine-tuned Qwen3-8B Large Language Models for detecting online polarization, its categories, and manifestation types across seven Indic languages: Bengali, Hindi, Nepali, Oriya, Punjabi, Telugu, and Urdu. The research employed Quantized Low-Rank Adaptation (QLoRA) for parameter-efficient fine-tuning. Experimental results demonstrated the approach's effectiveness, yielding macro F1-scores for polarization detection ranging from 0.76 to 0.90 (specifically 0.82, 0.78, 0.90, 0.76, 0.78, 0.87, and 0.79 for the respective languages). The proposed model outperformed established baseline systems in several subtasks. Furthermore, a unified model, fine-tuned on a concatenated dataset of all seven languages, significantly improved performance over standalone language-specific models, notably achieving a 28.76 point F1-score gain in Subtask 2 for Punjabi, underscoring the benefits of cross-lingual knowledge transfer in low-resource settings.

Key takeaway

For NLP Engineers developing multilingual models for low-resource Indic languages, consider adopting parameter-efficient fine-tuning with QLoRA on models like Qwen3-8B. Your approach should prioritize creating unified models by concatenating datasets across related languages, as this strategy demonstrably improves F1-scores, offering a significant advantage over developing standalone language-specific models. This method enhances performance and addresses linguistic diversity more effectively.

Key insights

Cross-lingual fine-tuning of Qwen3-8B with QLoRA effectively detects polarization in low-resource Indic languages.

Principles

• Parameter-efficient fine-tuning is viable for diverse low-resource languages.
• Cross-lingual knowledge transfer improves performance in low-resource settings.
• Unified models can outperform language-specific models for related languages.

Method

Fine-tune Qwen3-8B using QLoRA on a concatenated dataset of seven Indic languages (Bengali, Hindi, Nepali, Oriya, Punjabi, Telugu, Urdu) for multilingual polarization detection.

In practice

• Apply QLoRA to Qwen3-8B for Indic language NLP tasks.
• Combine low-resource language datasets for cross-lingual transfer.

Topics

• Multilingual NLP
• Polarization Detection
• Indic Languages
• Qwen3-8B
• QLoRA
• Cross-lingual Transfer

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• [P] QLoRA Fine-Tuning of Qwen2.5-1.5B for CEFR English Proficiency Classification (A1–C2) [P] — QLoRA fine-tuning of Qwen2.5-1.5B effectively classifies CEFR English proficiency, despite synthetic data limitations.
• Cross-lingual Relation Extraction with Large Language Models: Zero-Shot, Few-Shot, and Fine-Tuned Evaluation on Romanian — QLoRA fine-tuning significantly boosts cross-lingual relation extraction performance for low-resource languages, outperforming prompt-only LLMs.
• LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification — Iterative imbalance-aware fine-tuning of LLMs, combined with augmentation and post-processing, significantly improves rare-class NLP performance.
• Complex & Intelligent Systems, Volume 12, Issue 3, March 2026 — The issue highlights advancements in AI, optimization, and security across diverse applications and system complexities.
• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific datasets and LoRA fine-tuning significantly enhance LLM performance in specialized tasks like classical poetry analysis.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.