The Tatoxa System for Text Detoxification in Low-Resource Languages: The Case of Tatar

2026-06-24

AI Summary

This study addresses the critical gap in effective text detoxification techniques for low-resource languages, focusing on Tatar, which faces significant challenges in mitigating online harmful content. The work proposes Tatoxa, the first dedicated detoxification system for Tatar, alongside the creation of the language’s inaugural detoxification dataset. Leveraging large language models fine-tuned exclusively on this native Tatar data, Tatoxa demonstrates substantial performance gains over cross-lingual transfer approaches. Experimental results show that Tatoxa outperforms both existing open-source and commercial large language models on key evaluation metrics, underscoring the pivotal role of language-specific, locally sourced data in ensuring content safety for low-resource languages.

Abstract

Text detoxification, the automated detection and mitigation of abusive and harmful content, is essential for ensuring the safety of online communities and protecting users. However, low resource languages such as Tatar have received little research attention. In this paper we present Tatoxa, a novel state-of-the-art system for text detoxification in the Tatar language. Comparative experiments show that the proposed approach outperforms existing open source and proprietary commercial LLMs on key quality metrics. We also introduce a new dataset for text detoxification in Tatar, designed for fine tuning and evaluation in low resource settings. Finally, cross lingual transfer experiments indicate that transfer from other languages, including the culturally close Russian, performs significantly worse than training on native Tatar data even when a large Russian corpus is available.

Problem

• text detoxification
• low-resource languages
• Tatar
• abusive content
• online safety

Innovation

• text detoxification
• low-resource languages
• Tatar
• cross-lingual transfer
• dataset creation