PrimeLine@DravidianLangTech 2026: Abusive Tamil Comment Detection Using MuRIL

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, medium

Summary

PrimeLine@DravidianLangTech 2026 presents a system for binary classification of abusive language in Tamil social media comments. This system addresses the challenge of detecting abuse in Tamil, a morphologically rich language often mixed with English and informal romanized forms. The core of the system is MuRIL, a BERT-based encoder pre-trained on 17 Indian languages and their transliterated equivalents, which was fine-tuned for the DravidianLangTech@ACL 2026 shared task. The fine-tuned MuRIL achieved a macro-averaged F1 score of 0.83 on the validation set. This performance significantly surpasses generic multilingual baselines, specifically XLM-RoBERTa, which scored 0.79, and mBERT, which scored 0.77, under identical training conditions. This demonstrates that Indian-language-specific pre-training offers a distinct advantage for abusive language detection in code-mixed Tamil.

Key takeaway

For NLP Engineers developing content moderation systems for Dravidian languages, consider MuRIL as a strong baseline. Your efforts in detecting abusive language in code-mixed Tamil will benefit from models pre-trained on specific Indian languages, as demonstrated by MuRIL's superior F1 score of 0.83 compared to generic multilingual alternatives like XLM-RoBERTa. Prioritize language-specific models to achieve higher accuracy and robustness in challenging linguistic contexts.

Key insights

Indian-language-specific pre-training significantly enhances abusive language detection in code-mixed Tamil.

Principles

• Language-specific pre-training outperforms generic multilingual models.
• Morphological richness and code-mixing complicate NLP tasks.

Method

Fine-tuning MuRIL, a BERT-based encoder pre-trained on 17 Indian languages and their transliterated equivalents, for binary classification of Tamil comments.

In practice

• Use MuRIL for Dravidian language NLP tasks.
• Benchmark against XLM-RoBERTa and mBERT.

Topics

• Abusive Language Detection
• Tamil NLP
• MuRIL
• BERT-based Models
• Code-mixing
• Dravidian Languages

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Multi-Stage Training for Abusive Comment Detection in Indic Languages — Multi-stage training with language preprocessing and model ensembles improves abusive comment detection while minimizing false positives.
• 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.
• Building Singlish2Sinhala: A Machine Learning Approach to Sinhala Transliteration — Informal transliteration systems like Singlish present unique NLP challenges due to spelling inconsistencies.
• Muted: Multilingual Targeted Offensive Speech Identification and Visualization — Muted identifies targeted offensive speech using transformer models and spaCy, requiring no fine-tuning for toxic span detection.
• AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages — Models exhibit a significant "syntax gap" in processing African languages, struggling with their structural diversity.
• MentalMARBERT: Domain-Adaptive Pre-training and Two-Stage Fine-Tuning for Arabic Mental Health Disorders Detection — Domain-adaptive pre-training and hierarchical classification significantly improve Arabic mental health disorder detection.

Open in AIssential →

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