Robust Text Watermarking for Large Language Models via Dual Semantic Embeddings

2026-06-30

AI Summary

This work addresses the vulnerability of copyright protection for large language model (LLM)-generated text to semantic paraphrasing and translation, which often invalidates existing watermarking schemes. To overcome this limitation, the authors propose Dual-Embedding Watermarking (DEW), a novel approach that jointly leverages token-level and context-level semantic embeddings. DEW constructs a robust and imperceptible watermark signal in vector space using a key-driven pseudorandom projection matrix and enables detection through signal processing and statistical hypothesis testing. Evaluated across multiple mainstream LLMs, DEW maintains high text quality while significantly improving watermark detectability in paraphrased or translated outputs, demonstrating superior robustness and graceful degradation compared to current semantic watermarking methods.

Abstract

This work presents Dual-Embedding Watermarking (DEW), a semantic watermarking scheme for large language models (LLMs) that leverages contextual and token-level embeddings to enhance robustness against paraphrasing and translation. DEW utilizes a signal-processing methodology, applying algebraic vector-space operations to \mbox{token and context embeddings to derive a watermark signal that degrades gracefully under semantic shifts. The method obfuscates the watermark by projecting embedding vectors through pseudo-random matrices seeded with a secret key. Relevant distributions derived from the underlying algebra are evaluated and employed for statistical testing and benchmarking of DEW. Experimental results across multiple LLMs indicate that DEW improves post-paraphrase detection while maintaining competitive text quality, and remains detectable after translation, even when prior semantic watermarks degrade significantly. These findings position DEW as a practical and robust solution for safeguarding LLM-generated text and addressing critical issues in responsible AI deployment.

Problem

• text watermarking
• large language models
• semantic robustness
• paraphrasing resistance
• translation invariance

Innovation

• Dual-Embedding Watermarking
• semantic watermarking
• large language models
• robustness to paraphrasing
• vector-space signal processing