New statistical analysis: transition grammar and bidirectional self-clustering in the Voynich Manuscript

[DataWeaver22]

Hi all,

I'm Amy Laird, an independent researcher. I've been working on a statistical analysis of sequential constraints in the Voynich Manuscript and wanted to share it with this community before the arXiv preprint goes up.

The paper is: Transition Grammar of the Voynich Manuscript: Sequential Constraints and Bidirectional Self-Clustering Symmetry

Everything — paper, data, code, and an interactive dashboard — is at:
https://github.com/amy2213/Voynich-Transition-Grammar

The three strongest findings:

1. Two distributed transition rules. CHEDY→QOK attraction at 2.63x above chance and AIIN→QOK repulsion at 0.50x. Both hold across every manuscript section, both scribal hands, and all line lengths. They're distributed across 77% of individual CHEDY tokens and 369 unique word pairs — a class-level constraint, not fixed phrases.

2. Lines are grammatical units. The CHEDY→QOK attraction is 2.54x within a line but drops to 0.85x across line breaks. The grammar resets at every line boundary regardless of line length or section. Each line appears to be a clause or phrase, not arbitrary text wrapping.

3. Bidirectional self-clustering symmetry. When I measure how often similar-looking words cluster together, Voynich clusters equally at word beginnings (1.52x) and word endings (1.54x), ratio 0.99. I tested this against 13 natural-language comparators including Arabic, Latin, Finnish, Estonian, Turkish, Hungarian, Hebrew, Swahili, and others. Every natural language with positive clustering is suffix-dominant. Voynich is the only tested system with elevated, balanced clustering in both directions.

The analysis also covers suffix agreement between adjacent words (adjacent words agree on their endings more than chance predicts, and this cascades through 3-token chains), productive morphological paradigms, and glyph-layer architecture.

I use the Zandbergen-Landini EVA transliteration — so René's work underpins all of this. Happy to discuss methodology or findings.

Amy
https://github.com/amy2213/Voynich-Transition-Grammar