StyleRemix: Interpretable Authorship Obfuscation via Distillation and Perturbation of Style Elements

TL;DR

StyleRemix is an interpretable authorship obfuscation method that perturbs specific style elements using pre-trained modules, outperforming larger models in robustness and controllability.

Contribution

The paper introduces StyleRemix, a novel, interpretable approach for authorship obfuscation that manipulates style features with low computational cost and provides new datasets for research.

Findings

Outperforms state-of-the-art baselines in style obfuscation tasks.

Maintains robustness while controlling stylistic features.

Achieves high performance with low computational cost.

Abstract

Authorship obfuscation, rewriting a text to intentionally obscure the identity of the author, is an important but challenging task. Current methods using large language models (LLMs) lack interpretability and controllability, often ignoring author-specific stylistic features, resulting in less robust performance overall. To address this, we develop StyleRemix, an adaptive and interpretable obfuscation method that perturbs specific, fine-grained style elements of the original input text. StyleRemix uses pre-trained Low Rank Adaptation (LoRA) modules to rewrite an input specifically along various stylistic axes (e.g., formality and length) while maintaining low computational cost. StyleRemix outperforms state-of-the-art baselines and much larger LLMs in a variety of domains as assessed by both automatic and human evaluation. Additionally, we release AuthorMix, a large set of 30K high-quality, long-form texts from a diverse set of 14 authors and 4 domains, and DiSC, a parallel corpus of 1,500 texts spanning seven style axes in 16 unique directions