On the Learnability of Watermarks for Language Models

arXiv:2312.04469v3 Announce Type: replace-cross
Abstract: 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 …

abstract arxiv cs.cl cs.cr cs.lg decoder detection generate generated language language model language models learn paper statistical strategies text the decoder type watermarking watermarks