aiqbench analysis
This research introduces LUNA, a linguistically adaptive watermarking method designed to identify AI-generated text without degrading output quality or limiting verification to the model provider. By leveraging part-of-speech contexts and external corpora, the technique achieves near-perfect detection accuracy (AUROC 0.9959) while maintaining minimal perplexity shifts across six typologically diverse languages. This advancement addresses critical multilingual deployment challenges, ensuring robust content provenance without compromising the utility of language models in global applications.
Linguistics-Aware Non-Distortionary LLM Watermarking
arXiv:2606.00613v2 Announce Type: replace-cross Abstract: Watermarking should identify language-model output without degrading quality or limiting verification to the model provider. Multilingual deployment makes this harder because morphology, segmentation, and script change where watermark evidence can be naturally embedded. We introduce LUNA, a linguistically adaptive watermark that combines model-free detection with single-token non-distortion under the standard random-key model. LUNA estimates normalized next-tag entropy from part-of-speech contexts in an external corpus and uses it to set the depth of a non-distortionary binary tournament sampler; the detector reconstructs the same schedule from text, a tokenizer, a tagger, and a secret key. We evaluate LUNA on six typologically diverse languages and two domains against eight primary baselines. LUNA attains an AUROC of 0.9959 and the lowest mean absolute median perplexity shift, 0.045, across the 12 settings; its 95% bootstrap interval [0.022, 0.073] lies below all baseline intervals. LUNA also records the lowest mean on Self-BLEU, Distinct-1, surprisal, and entropy shifts; it is the only method that simultaneously achieves AUROC > 0.99 and an absolute median perplexity shift below 0.1 in a majority of settings, reaching this regime in 9 of the 12 settings while no baseline reaches it in more than 2. Our code is available at https://github.com/Shinwoo-Park/luna_watermark.
Abstract:Watermarking should identify language-model output without degrading quality or limiting verification to the model provider. Multilingual deployment makes this harder because morphology, segmentation, and script change where watermark evidence can be naturally embedded. We introduce LUNA, a linguistically adaptive watermark that combines model-free detection with single-token non-distortion under the standard random-key model. LUNA estimates normalized next-tag entropy from part-of-speech contexts in an external corpus and uses it to set the depth of a non-distortionary binary tournament sampler; the detector reconstructs the same schedule from text, a tokenizer, a tagger, and a secret key. We evaluate LUNA on six typologically diverse languages and two domains against eight primary baselines. LUNA attains an AUROC of 0.9959 and the lowest mean absolute median perplexity shift, 0.045, across the 12 settings; its 95% bootstrap interval [0.022, 0.073] lies below all baseline intervals. LUNA also records the lowest mean on Self-BLEU, Distinct-1, surprisal, and entropy shifts; it is the only method that simultaneously achieves AUROC > 0.99 and an absolute median perplexity shift below 0.1 in a majority of settings, reaching this regime in 9 of the 12 settings while no baseline reaches it in more than 2. Our code is available at this https URL.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2606.00613 [cs.CL]
(or arXiv:2606.00613v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.00613 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 297 words
Related coverage
Grammar-Guided Code Watermarking with Green Temperature
arXiv · Oct 7, 2026
How Human-Like Are Large Language Models? A Register-Aware Linguistic Evaluation Framework
arXiv · Sep 1, 2026
AGMark: Attention-Guided Dynamic Watermarking for Large Vision-Language Models
arXiv · Jul 29, 2026
GenoTrace: Inheritable Watermarks for Genome Foundation Model Distillation
arXiv · Sep 30, 2026