{"slug": "no-tokenizer-still-a-gap-cross-lingual-fairness-of-homoglyph-and-invisible", "title": "No Tokenizer, Still a Gap: Cross-Lingual Fairness of Homoglyph and Invisible-Character Watermarking", "summary": "Homoglyph and invisible-character watermarks still produce cross-lingual inequity even though they bypass tokenizers, according to an arXiv paper (2610.36432v1) measuring 40 language varieties across 24 scripts. The study found watermark capacity varies 3.3-fold, with script accounting for 96% of that variation; within Latin script resource level made no detectable difference (ρ=+0.095, p=0.78), but across scripts capacity rose with resource level (ρ=+0.54, 95% CI [+0.14,+0.83]) because lower-resourced languages tend to use scripts with few homoglyphs. After ordinary copy-paste strips invisible characters, a keyed detector needs about 2 code points of English or Persian text but 93 of Lao, leading the authors to conclude that removing the tokenizer does not remove the inequity, only changes its source.", "body_md": "arXiv:2610.36432v1 Announce Type: new \nAbstract: Homoglyph and invisible-character watermarks are edit-based watermarks. Instead of changing how a model generates text, they hide a signal in the text after it is written. They swap some characters for homoglyphs, which look the same, or very similar, but have different Unicode code points, or they insert characters that do not display, all without access to the model that wrote the text. Generation-time watermarks work less well for lower-resourced languages due to how tokenizers split their text. Because they work on individual characters, these watermarks never touch a tokenizer, so we ask whether they escape this gap. Across 40 language varieties in 24 scripts, we measure capacity (the hidden bits each character can carry), how much of it survives routine handling and targeted removal, and how much text a keyed detector needs. Capacity varies 3.3-fold and script accounts for 96\\% of that variation. Within Latin script, resource level makes no detectable difference ($\\rho=+0.095$, $p=0.78$), but across scripts capacity rises with it ($\\rho=+0.54$, 95\\% CI $[+0.14,+0.83]$), because lower-resourced languages tend to use scripts with few homoglyphs. Ordinary copy-paste, which strips invisible characters, widens the gap: a detector then needs about 2 code points of English or Persian but 93 of Lao. Removing the tokenizer does not remove the inequity, but instead only changes its source.", "url": "https://wpnews.pro/news/no-tokenizer-still-a-gap-cross-lingual-fairness-of-homoglyph-and-invisible", "canonical_source": "https://arxiv.org/abs/2610.36432", "published_at": "2026-10-28 04:00:00+00:00", "updated_at": "2026-10-28 04:12:51.669748+00:00", "lang": "en", "topics": ["ai-safety", "natural-language-processing", "ai-research", "large-language-models"], "entities": ["arXiv", "Lao", "Persian", "English"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/no-tokenizer-still-a-gap-cross-lingual-fairness-of-homoglyph-and-invisible", "markdown": "https://wpnews.pro/news/no-tokenizer-still-a-gap-cross-lingual-fairness-of-homoglyph-and-invisible.md", "text": "https://wpnews.pro/news/no-tokenizer-still-a-gap-cross-lingual-fairness-of-homoglyph-and-invisible.txt", "jsonld": "https://wpnews.pro/news/no-tokenizer-still-a-gap-cross-lingual-fairness-of-homoglyph-and-invisible.jsonld"}}