TriVector@DravidianLangTech 2026: Abusive Tamil Text Detection on Social Media Using Lexicon-Augmented Transformers

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

Summary

TriVector@DravidianLangTech 2026 introduced a system designed for detecting abusive Tamil text targeting women on social media, specifically addressing the challenges inherent in low-resource language abuse detection. This system combines nine handcrafted lexicon features with pretrained multilingual transformer embeddings. Researchers systematically evaluated its effectiveness by comparing model performance both with and without these lexical attributes across multiple transformer architectures. The best-performing model, XLM-RoBERTa-Large, achieved a macro F1-score of 81.71%, securing 15th rank in the DravidianLangTech@ACL 2026 competition. The study's findings highlight that larger multilingual models generalize more effectively to unseen data compared to smaller domain-specific models, and the integration of lexical features yielded only mild performance gains.

Key takeaway

For NLP Engineers developing abusive text detection systems in low-resource languages like Tamil, you should prioritize larger multilingual transformer models such as XLM-RoBERTa-Large. These models demonstrate superior generalization to unseen data compared to smaller, domain-specific alternatives. While handcrafted lexical features can be integrated, expect only mild performance improvements, suggesting that core model architecture and pretraining scale are more critical for robust performance.

Key insights

Larger multilingual transformers outperform smaller domain-specific models for low-resource abusive text detection, with lexical features offering minor benefits.

Principles

• Larger multilingual models generalize better.
• Lexical features offer mild performance gains.
• Low-resource language abuse detection is challenging.

Method

The proposed method integrates nine handcrafted lexicon features with pretrained multilingual transformer embeddings, evaluating performance with and without these attributes across various transformer architectures.

In practice

• Prioritize larger multilingual models.
• Experiment with handcrafted lexicon features.
• Focus on gender-based abuse detection.

Topics

• Abusive Text Detection
• Low-Resource Languages
• Tamil NLP
• Transformer Models
• XLM-RoBERTa
• Lexicon Features
• Social Media Analysis

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

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.
• 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.
• Detecting HIV-Related Stigma in Clinical Narratives Using Large Language Models — LLMs can effectively detect HIV-related stigma in clinical notes, with encoder models outperforming generative models in zero-shot settings.
• Anatomy of Data Repositories for the Analysis and Detection of Toxicity in Portuguese — Lexical diversity and annotation practices significantly impact cross-domain generalization in hate speech detection datasets.
• From Logistic Regression to GPT-2: Building a Complete Spam Detection & Sentiment Analysis Pipeline — For imbalanced text classification, prioritize F1-score and Precision-Recall AUC over accuracy and ROC-AUC.

Open in AIssential →

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