Frustratingly Easy Edit-based Linguistic Steganography with a Masked Language Model

With advances in neural language models, the focus of linguistic\nsteganography has shifted from edit-based approaches to generation-based ones.\nWhile the latter's payload capacity is impressive, generating genuine-looking\ntexts remains challenging. In this paper, we revisit edit-based linguistic\nsteganography, with the idea that a masked language model offers an\noff-the-shelf solution. The proposed method eliminates painstaking rule\nconstruction and has a high payload capacity for an edit-based model. It is\nalso shown to be more secure against automatic detection than a\ngeneration-based method while offering better control of the security/payload\ncapacity trade-off.\n