MTD-EU-26: Multilingual English–Urdu Text Detoxification Dataset

This dataset accompanies the study “Multilingual Detoxification of English–Urdu Social-Media Text: A Low-Resource Dataset and LLM Evaluation.”

The dataset contains multilingual toxic and detoxified text pairs in English and Urdu collected from social media platforms, including Reddit, Twitter (X), Facebook, and YouTube. It is designed to support research in text detoxification, low-resource natural language processing (NLP), and evaluation of large language models (LLMs).

Contents:

English datasets:

Reddit (3K samples)

Twitter (3K samples)

YouTube (3K samples)

Urdu dataset:

Facebook (3,027 samples)

This dataset is intended for benchmarking detoxification models and advancing multilingual NLP research in low-resource settings.