Title: Semantic compensation-driven low-distortion linguistic steganography

Authors: Lingyun Xiang; Songjun Liang; Song Jiang

Addresses: School of Computer Science and Technology, Changsha University of Science and Technology, Changsha, 410114, China ' School of Computer Science and Technology, Changsha University of Science and Technology, Changsha, 410114, China ' School of Computer Science and Technology, Changsha University of Science and Technology, Changsha, 410114, China

Abstract: Existing modification-based linguistic steganography (MLS) often suffers from semantic distortion, compromising imperceptibility and security. To address this issue, we propose Low-Distortion Linguistic Steganography(LDStega), which enhances semantic fidelity via semantic compensation substitutions. LDStega utilises an augmented masked language model (MLM) with a dynamic masking strategy to generate contextually coherent candidate words. To ensure substitution integrity, we introduce a BART-based evaluation to quantitatively assess contextual appropriateness. Furthermore, a semantic compensation strategy is proposed to refine subsequent word choices based on semantic fidelity feedback rather than secret messages, effectively mitigating collocation mismatches and unintended distortions. Extensive experiments demonstrate that LDStega outperforms baseline methods in text quality, semantic similarity, and resistance to steganalysis.

Keywords: linguistic steganography; word substitution; semantic distortion; semantic fidelity; MLM; masked language model.

DOI: 10.1504/IJAACS.2026.152851

International Journal of Autonomous and Adaptive Communications Systems, 2026 Vol.19 No.2, pp.259 - 274

Received: 24 Feb 2025
Accepted: 02 Apr 2025
Published online: 13 Apr 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article