Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

On the Learnability of Watermarks for Language Models

About

Watermarking of language model outputs enables statistical detection of model-generated text, which can mitigate harms and misuses of language models. Existing watermarking strategies operate by altering the decoder of an existing language model. In this paper, we ask whether language models can directly learn to generate watermarked text, which would have significant implications for the real-world deployment of watermarks. First, learned watermarks could be used to build open models that naturally generate watermarked text, enabling watermarking for open models, where users can control the decoding procedure. Second, if watermarking is used to determine the provenance of generated text, an adversary can hurt the reputation of a victim model by spoofing its watermark and generating damaging watermarked text. To investigate the learnability of watermarks, we propose watermark distillation, which trains a student model to behave like a teacher model that uses decoding-based watermarking. We test our approach on three decoding-based watermarking strategies and various hyperparameter settings, finding that models can learn to generate watermarked text with high detectability. We also find limitations to learnability, including the loss of watermarking capabilities under fine-tuning on normal text and high sample complexity when learning low-distortion watermarks.

Chenchen Gu, Xiang Lisa Li, Percy Liang, Tatsunori Hashimoto• 2023

Related benchmarks

TaskDatasetResultRank
Watermark DetectabilityC4 RealNewsLike (Del-0.2) (test)
AUC94.7
28
Text generation quality and watermark detectabilityC4 RealNewsLike
AUC98.1
16
Text generation quality and watermark detectabilityELI5
AUC96.9
16
Watermark DetectabilityC4 RealNewsLike (Del-0.5) (test)
AUC85.1
14
Watermark DetectabilityC4 RealNewsLike (Sub-0.5) (test)
AUC83.7
14
Watermark DetectabilityC4-RealNewsLike Dipper-1 (test)
AUC0.855
14
Watermark DetectabilityC4 RealNewsLike Dipper-2 (test)
AUC59
14
Watermark DetectabilityC4-RealNewsLike Translate (test)
AUC91.2
14
Showing 8 of 8 rows

Other info

Follow for update