SmurfCat at PAN 2024 TextDetox: Alignment of Multilingual Transformers for Text Detoxification

TL;DR

This paper introduces a multilingual text detoxification method using data augmentation, fine-tuning of sequence-to-sequence models, and alignment techniques, achieving top results in the PAN-2024 competition.

Contribution

It presents a novel multilingual dataset creation process and applies advanced alignment and fine-tuning techniques to improve text detoxification performance.

Findings

Achieved state-of-the-art results for Ukrainian detoxification.

Secured first place in automated evaluation at PAN 2024.

Obtained second place in human evaluation at PAN 2024.

Abstract

This paper presents a solution for the Multilingual Text Detoxification task in the PAN-2024 competition of the SmurfCat team. Using data augmentation through machine translation and a special filtering procedure, we collected an additional multilingual parallel dataset for text detoxification. Using the obtained data, we fine-tuned several multilingual sequence-to-sequence models, such as mT0 and Aya, on a text detoxification task. We applied the ORPO alignment technique to the final model. Our final model has only 3.7 billion parameters and achieves state-of-the-art results for the Ukrainian language and near state-of-the-art results for other languages. In the competition, our team achieved first place in the automated evaluation with a score of 0.52 and second place in the final human evaluation with a score of 0.74.