I’m proud to announce that our latest article TAROT: Task-Oriented Authorship Obfuscation Using Policy Optimization Methods has been accepted at PrivateNLP 2025 (Sixth Workshop on Privacy in Natural Language Processing, Colocated with NAACL 2025).
TL;DR: The paper introduces TAROT, an unsupervised method for authorship obfuscation that balances privacy and utility in text rewriting. While strong obfuscation techniques protect identity but harm text quality, weaker ones maintain utility but risk de-anonymization. TAROT uses policy optimization to fine-tune small language models, regenerating text to obscure authorship while preserving usefulness for downstream tasks. Experiments show it effectively reduces attacker accuracy while maintaining utility. The code and models are publicly available.
🇬🇧 TAROT: Task-Oriented Authorship Obfuscation Using Policy Optimization Methods. Gabriel Loiseau, Damien Sileo, Damien Riquet, Maxime Meyer, Marc Tommasi. Sixth Workshop on Privacy in Natural Language Processing (PrivateNLP 2025).
Links:
