SSG: Logit-Balanced Vocabulary Partitioning for LLM Watermarking
Chenxi Gu , Xiaoning Du , John Grundy · 2026
Read SSG: Logit-Balanced Vocabulary Partitioning for LLM Watermarking by Chenxi Gu , Xiaoning Du , John Grundy (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-22438.txt.
Description
Gu, Du and Grundy improve language model watermarking, which traces generated content authorship. The KGW scheme is attractive but degrades badly in low-entropy code and mathematical reasoning. A crucial KGW step is random vocabulary partitioning, which adjusts token choice. The authors show the next-token distribution governs token selection changes and sets the lower bound of watermark strength. Their SSG, sort-then-split by groups, partitions the vocabulary into two logit-balanced subsets, lifting the lower bound and aiding detection. Code and mathematical reasoning tests confirm its effectiveness.
Catalog record
| CREATOR | Chenxi Gu , Xiaoning Du , John Grundy |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37415 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.22438