Answer · Voynich Manuscript
Predictive information reaches beyond the word: a model that sees more of the preceding…
Predictive information reaches beyond the word: a model that sees more of the preceding text predicts the next unit better, on sheets it was not trained on, as it does in languages.
Where it stands
Ruled outNo evidence yet.
What it predicts
- held-out bits per glyph keep falling as the model's context grows past one word (order 6-8); held-out bits per word fall when the previous word is known (order 2 beats order 1)
- by an amount inside the language range
Evidence none for · none against
Glyph entropy is lower than in any of 103 languages
NEUTRALEstablished · weighs little · F-010
Tested by 2 runs
| Run | Measured | Result | Against | Decision | Date |
|---|---|---|---|---|---|
| RUN-018 | held-out bits/glyph at order 1,2,4,6,8 | 3.8944,2.4432,2.2447,2.3618,2.5022 | Lines shuffled: 3.8669,2.272,2.0581,2.1446,2.2818 | See note | 22 Oct |
| RUN-024 | held-out bits/glyph at order 1,2,4,6,8 | 4.0304,2.7036,2.5555,2.7075,2.8533 | Lines shuffled: 4.0202,2.5485,2.381,2.5125,2.6605 | See note | 22 Oct |
The questions it answers
Most likelyThe line is a unit of the system
Cite it
Mysteries Department. (2026). Predictive information reaches beyond the word: a model that sees more of the preceding… [Answer ANS0D8FZY544WRM]. https://mysteriesdept.com/r/ANS0D8FZY544WRM
@misc{mysteriesdept_ans0d8fzy544wrm,
title = {Predictive information reaches beyond the word: a model that sees more of the preceding…},
author = {{Mysteries Department}},
year = {2026},
note = {Answer, catalog key ANS0D8FZY544WRM},
url = {https://mysteriesdept.com/r/ANS0D8FZY544WRM}
}