Is Multilingual LLM Watermarking Truly Multilingual? Scaling Robustness to 100+ Languages via Back-Translation
STEAM, a detection method that uses Bayesian optimisation to search among 126 candidate languages for the back-translation that best recovers the watermark strength, is introduced, compatible with any watermarking method, robust across different tokenizers and languages, non-invasive, and easily extendable to new langu...
Asim Mohamed, Martin Gubri
· 1 citation