SynthDetoxM: Modern LLMs are Few-Shot Parallel Detoxification Data Annotators

TL;DR

This paper presents SynthDetoxM, a large multilingual detoxification dataset generated using modern LLMs, which improves detoxification performance in low-data scenarios and advances multilingual toxic language mitigation.

Contribution

The paper introduces SynthDetoxM, a novel multilingual detoxification dataset created through a new pipeline using open-source LLMs, enhancing detoxification model training and performance.

Findings

Models trained on SynthDetoxM outperform those trained on human-annotated datasets.

SynthDetoxM enables superior detoxification in few-shot learning scenarios.

The dataset covers four major languages with 16,000 high-quality sentence pairs.

Abstract

Existing approaches to multilingual text detoxification are hampered by the scarcity of parallel multilingual datasets. In this work, we introduce a pipeline for the generation of multilingual parallel detoxification data. We also introduce SynthDetoxM, a manually collected and synthetically generated multilingual parallel text detoxification dataset comprising 16,000 high-quality detoxification sentence pairs across German, French, Spanish and Russian. The data was sourced from different toxicity evaluation datasets and then rewritten with nine modern open-source LLMs in few-shot setting. Our experiments demonstrate that models trained on the produced synthetic datasets have superior performance to those trained on the human-annotated MultiParaDetox dataset even in data limited setting. Models trained on SynthDetoxM outperform all evaluated LLMs in few-shot setting. We release our dataset and code to help further research in multilingual text detoxification.