The Secret's in the Word Order: Text-to-Text Generation for Linguistic Steganography

Linguistic steganography is a form of covert communication using natural language to conceal the existence of the hidden message, which is usually achieved by systematically making changes to a cover text. This paper proposes a linguistic steganography method using word ordering as the linguistic transformation. We show that the word ordering technique can be used in conjunction with existing translation-based embedding algorithms. Since unnatural word orderings would arouse the suspicion of third parties and diminish the security of the hidden message, we develop a method using a maximum entropy classifier to determine the naturalness of sentence permutations. The classifier is evaluated by human judgements and compared with a baseline method using the Google n-gram corpus. The results show that our proposed system can achieve a satisfactory security level and embedding capacity for the linguistic steganography application.