AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

How Good is Post-Hoc Watermarking With Language Model Rephrasing?

Pierre Fernandez, Tom Sander, Hady Elsahar, Hongyan Chang, Tomáš Souček, Valeriu Lacatusu, Tuan A. Tran, Sylvestre-Alvise Rebuffi, Alexandre Mourachko

ICML 2026 · Large language models › Language, safety and interpretability · approach: Text watermarking · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add context-aware chunking (ablation)post-hoc rephrasing watermark · Dickens excerpts (4000 tokens)detection -log10 p: 24.9 -> 198.3Improved
add context-aware chunking (ablation)post-hoc rephrasing watermark · Dickens excerpts (1500 tokens)SBERT similarity: 0.85 -> 0.91Improved
add context-aware chunking (ablation)post-hoc rephrasing watermark · Dickens excerpts (500 tokens)detection -log10 p: 19.3 -> 19.3No clear effect
add context-aware chunking (ablation)post-hoc rephrasing watermark · Dickens excerpts (4000 tokens)length ratio: 0.70 -> 0.86Improved
increase model size from Llama-3.1-8B to Llama-3.3-70BGumbel-max code watermarking · HumanEval/MBPP code (T=1.4, passed samples)-log10 p-value (median): 1.83 -> 0.44Hurt
increase model size from Llama-3.1-8B to Llama-3.3-70BGumbel-max code watermarking · HumanEval/MBPP code (T=1.4)pass@1: 0.71 -> 0.90Improved
raise sampling temperature from 1.0 to 1.4Gumbel-max code watermarking with Llama-3.1-8B · HumanEval/MBPP codepass@1: 0.92 -> 0.71Hurt
raise sampling temperature from 1.0 to 1.4Gumbel-max code watermarking with Llama-3.1-8B · HumanEval/MBPP code (passed samples)-log10 p-value (median): 0.74 -> 1.83Improved
raise temperature from 0.80 to 1.20Gumbel-max watermarking · Wikipedia ru-log10 p-value (median): 1.355 -> 23.689Improved
raise temperature from 0.80 to 1.20Gumbel-max watermarking · Wikipedia ruSBERT similarity: 0.908 -> 0.836Hurt
rephrase non-English (ru) instead of EnglishGumbel-max watermarking · Wikipedia (T=1.0)SBERT similarity: 0.954 -> 0.888Hurt
add beam search to decoding (ablation)Green-red, SynthID, MorphMark, DiPMark · Dickens rephrasingcross entropy vs watermark strengthImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy