DLRG@DravidianLangTech 2026: Explainable Transformer-Based Detection of Abusive Tamil Text Targeting Women on Social Media

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

Summary

The paper "DLRG@DravidianLangTech 2026: Explainable Transformer-Based Detection of Abusive Tamil Text Targeting Women on Social Media" by Mirudhula Sankar and Ratnavel Rajalakshmi presents a transformer-based system for binary classification of Tamil comments as abusive or non-abusive. This system addresses the challenge of detecting online abuse in low-resource languages like Tamil, which features informal writing, spelling variations, and culturally specific expressions. The approach fine-tunes MuRIL, a multilingual transformer pretrained for Indian languages, to achieve effective contextual representation with minimal preprocessing. It incorporates a post-hoc Explainable AI component, using a perturbation-based method with log-odds differences to identify influential words. Experimental results show the model achieved a validation accuracy exceeding 81% and a strong macro-F1 score. The research demonstrates that combining contextual multilingual representations with simple interpretability methods is a viable solution for detecting abusive Tamil text. The system's implementation is publicly available at https://github.com/mirud5173/Abusive-Tamil-Comment-Detection-using-Transformer-Models.

Key takeaway

For NLP Engineers developing content moderation systems for regional languages, you should consider fine-tuning multilingual transformers like MuRIL. This approach, combined with a perturbation-based Explainable AI component, provides both high accuracy (over 81% validation accuracy) and crucial transparency for detecting abusive text in challenging low-resource contexts like Tamil. Implement simple interpretability methods to understand model decisions and build trust in your automated systems.

Key insights

Combining multilingual transformers with XAI effectively detects abusive text in low-resource languages.

Principles

• Contextual multilingual models improve low-resource language processing.
• Post-hoc XAI enhances transparency in abuse detection systems.
• Perturbation-based methods identify influential prediction factors.

Method

The system fine-tunes MuRIL for binary classification of Tamil comments. It integrates a perturbation-based XAI component using log-odds differences to explain predictions.

In practice

• Fine-tune MuRIL for similar low-resource language tasks.
• Implement perturbation-based XAI for model transparency.
• Utilize DravidianLangTech dataset for Tamil NLP.

Topics

• Abusive Language Detection
• Tamil Language Processing
• Transformer Models
• Explainable AI
• MuRIL
• Content Moderation

Code references

• mirud5173/Abusive-Tamil-Comment-Detection-using-Transformer-Models

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.
• Understanding Sarcasm in Generative AI Models: How ChatGPT Interprets Human Intent — Sarcasm detection in AI requires understanding non-literal meaning through contextual learning, moving beyond simple keyword analysis.
• Building Singlish2Sinhala: A Machine Learning Approach to Sinhala Transliteration — Informal transliteration systems like Singlish present unique NLP challenges due to spelling inconsistencies.
• An End-to-End Hybrid Framework for Rumour Detection in Low-Resources Algerian Dialect — A hybrid framework combining transformer embeddings with classical classifiers enables effective rumour detection in low-resource Algerian dialect.
• 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.
• Socially Responsible and Explainable Automated Fact-Checking and Hate Speech Detection — Explainable NLP methods and high-quality datasets can jointly optimize performance and interpretability for Portuguese hate speech and fact-checking.

Open in AIssential →

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