AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasing
For developers of text watermarking systems, AliMark provides a more robust method against paraphrasing attacks, addressing a key vulnerability in prior approaches.
Existing sentence-level watermarking methods are vulnerable to structural perturbations like sentence splitting and merging under strong paraphrasers. AliMark reformulates watermarking as bit sequence encoding and alignment, using multi-candidate alignment to improve robustness, substantially outperforming state-of-the-art baselines.
Existing sentence-level watermarking methods enhance robustness to paraphrasing by anchoring watermarks in sentence semantics. However, their prefix-based designs remain vulnerable to structural perturbations, such as sentence splitting and merging, which commonly arise under strong paraphrasers like DIPPER and GPT-3.5. To mitigate this issue, we propose AliMark, a framework that reformulates sentence-level watermarking as a bit sequence encoding and alignment problem between a potentially watermarked text and a secret bit sequence. Notably, our approach adopts a two-stage detection strategy: we generate multiple restructured text variants and adaptively align their extracted bit sequences with the secret bit sequence to minimize alignment cost. This multi-candidate alignment design naturally improves robustness to sentence merges and splits. Extensive experiments demonstrate that AliMark substantially outperforms state-of-the-art baselines under diverse paraphrasing attacks.