On the Learnability of Watermarks for Language Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Gu, Chenchen, Li, Xiang Lisa, Liang, Percy, Hashimoto, Tatsunori
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866913338550648832
author Gu, Chenchen
Li, Xiang Lisa
Liang, Percy
Hashimoto, Tatsunori
author_facet Gu, Chenchen
Li, Xiang Lisa
Liang, Percy
Hashimoto, Tatsunori
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2312_04469
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle On the Learnability of Watermarks for Language Models
Gu, Chenchen
Li, Xiang Lisa
Liang, Percy
Hashimoto, Tatsunori
Machine Learning
Computation and Language
Cryptography and Security
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.
title On the Learnability of Watermarks for Language Models
topic Machine Learning
Computation and Language
Cryptography and Security
url https://arxiv.org/abs/2312.04469