BanglaNirTox: A Large-scale Parallel Corpus for Explainable AI in Bengali Text Detoxification

2025-11-03

AI Summary

Harmful online content in Bengali is pervasive, yet high-quality detoxification resources and methods remain scarce for low-resource languages. Method: We propose the first explainable text detoxification framework tailored to low-resource languages, integrating Pareto-optimal large language models with chain-of-thought (CoT) prompting. This yields BanglaNirTox—a manually verified, attribution-annotated parallel corpus of 68,000 Bengali detoxification instances. Contribution/Results: Our framework pioneers the integration of multi-objective Pareto optimization with explainable reasoning, significantly improving detoxification quality, consistency, and transparency. BanglaNirTox and the proposed methodology establish critical infrastructure and a novel technical paradigm for explainable AI and content safety research in low-resource linguistic settings.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Language and Vision

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Mitigating online harms, including harassment, spam, and fake reviews

Abstract

Toxic language in Bengali remains prevalent, especially in online environments, with few effective precautions against it. Although text detoxification has seen progress in high-resource languages, Bengali remains underexplored due to limited resources. In this paper, we propose a novel pipeline for Bengali text detoxification that combines Pareto class-optimized large language models (LLMs) and Chain-of-Thought (CoT) prompting to generate detoxified sentences. To support this effort, we construct BanglaNirTox, an artificially generated parallel corpus of 68,041 toxic Bengali sentences with class-wise toxicity labels, reasonings, and detoxified paraphrases, using Pareto-optimized LLMs evaluated on random samples. The resulting BanglaNirTox dataset is used to fine-tune language models to produce better detoxified versions of Bengali sentences. Our findings show that Pareto-optimized LLMs with CoT prompting significantly enhance the quality and consistency of Bengali text detoxification.

Problem

• Detoxifying toxic Bengali text in online environments
• Addressing resource scarcity for Bengali text detoxification
• Generating explainable detoxified Bengali sentences using AI

Innovation

• Pareto-optimized LLMs for Bengali detoxification
• Chain-of-Thought prompting for explainable AI
• Large-scale parallel corpus with toxicity labels