Near-imperceptible Neural Linguistic Steganography via Self-Adjusting Arithmetic Coding
This work addresses the challenge of secure and imperceptible communication for applications like privacy and security, representing an incremental advancement over existing neural-based methods.
The paper tackles the problem of hiding secret messages in natural language texts by introducing a neural linguistic steganography method using self-adjusting arithmetic coding, achieving improvements of 15.3% in bits/word and 38.9% in KL metrics over previous state-of-the-art methods on four datasets.
Linguistic steganography studies how to hide secret messages in natural language cover texts. Traditional methods aim to transform a secret message into an innocent text via lexical substitution or syntactical modification. Recently, advances in neural language models (LMs) enable us to directly generate cover text conditioned on the secret message. In this study, we present a new linguistic steganography method which encodes secret messages using self-adjusting arithmetic coding based on a neural language model. We formally analyze the statistical imperceptibility of this method and empirically show it outperforms the previous state-of-the-art methods on four datasets by 15.3% and 38.9% in terms of bits/word and KL metrics, respectively. Finally, human evaluations show that 51% of generated cover texts can indeed fool eavesdroppers.