Multilingual and Explainable Text Detoxification with Parallel Corpora

TL;DR

This work addresses multilingual text detoxification by expanding parallel detoxification data to German, Chinese, Arabic, Hindi, and Amharic, and by conducting an explainability-driven analysis of toxicity and detoxification across nine languages. It introduces a Chain-of-Thoughts prompting framework that leverages cluster-derived descriptive attributes to guide detoxification, improving precision and reducing hallucinations. The authors benchmark a broad set of baselines and demonstrate that cluster-informed CoT prompting often yields the best joint performance quantified by $J = \frac{1}{n} \sum_{i=1}^n \text{STA}(y_i) \cdot \text{SIM}(x_i,y_i) \cdot \text{ChrF1}(x_i,y_i)$, while providing extensive multilingual data and prompts for community use. Overall, the paper advances explainable multilingual detoxification and offers practical insights for deploying safe, culturally aware text moderation across languages.

Abstract

Even with various regulations in place across countries and social media platforms (Government of India, 2021; European Parliament and Council of the European Union, 2022, digital abusive speech remains a significant issue. One potential approach to address this challenge is automatic text detoxification, a text style transfer (TST) approach that transforms toxic language into a more neutral or non-toxic form. To date, the availability of parallel corpora for the text detoxification task (Logachevavet al., 2022; Atwell et al., 2022; Dementievavet al., 2024a) has proven to be crucial for state-of-the-art approaches. With this work, we extend parallel text detoxification corpus to new languages -- German, Chinese, Arabic, Hindi, and Amharic -- testing in the extensive multilingual setup TST baselines. Next, we conduct the first of its kind an automated, explainable analysis of the descriptive features of both toxic and non-toxic sentences, diving deeply into the nuances, similarities, and differences of toxicity and detoxification across 9 languages. Finally, based on the obtained insights, we experiment with a novel text detoxification method inspired by the Chain-of-Thoughts reasoning approach, enhancing the prompting process through clustering on relevant descriptive attributes.

Figures (14)

• Figure 1: Examples of the desired texts detoxification for English and new languages: German, Chinese, Arabic, Hindi, and Amharic.
• Figure 2: In this work, we extend parallel text detoxification data to new languages as well as provide explainability analysis of toxicity and detoxification attributes across all languages. This information helps to improve Chain-of-Thoughts reasoning for automatic text detoxification with LLMs.
• Figure 3: Extracted with GPT-4 toxicity levels and top descriptive features per toxic and non-toxic parts in the multilingual parallel text detoxification data.
• Figure 4: Top-5 extracted keywords from toxic parts.
• Figure 5: Text detoxification with CoT: analyze the input, identify its cluster, and provide the detoxification explanation and cluster example in the prompt.