Voynich ManuscriptF-124
Evidence · Voynich Manuscript
Without the layout layer, topics follow scribe less
With the line-edge spellings normalised, page topics follow Currier language and scribe less (NMI 0.53 to 0.43 and 0.54 to 0.42, five seeds, sd up to 0.08) and the illustrated section as much as before (0.37 to 0.36); the A-B axis of the vocabulary does not move (0.86). Part of what topic models read as language and scribe is how line edges are written.
How it was found 2 runs
Inductive battery (13 methods) on half C: raw, normalised, interiortopics.nmi_section: 0.302 / 0.402 / 0.493; topics.nmi_lang: 0.587 / 0.548 / 0.466; stylometry.nmi_section: 0.327 / 0.556 / 0.615; seriation.axis1_vs_langB: 0.86 / 0.861 / 0.86; paradigms.dependence: 0.361 / 0.367 / 0.368; layout.mi_line_position_first: 0.1042 / 0.029 / 0.0164Exploratory
ZL3b transcriptionExploration splitLine-edge rules learned on half E (46 minimal alternations at paragraph start, line start and line end)Latin genre texts
Notes from the run
data/analysis/inductive/out/battery_C.csv, battery_C_layout.csv, battery_C_interior.csv
28 Oct
Topic NMI with section / Currier language / scribe, half C, mean of five seedsraw 0.37 / 0.53 / 0.54; normalised 0.36 / 0.43 / 0.42Exploratory
Cite it
Mysteries Department. (2026). Without the layout layer, topics follow scribe less [Evidence EVD0DCC13TM3WQS]. https://mysteriesdept.com/r/EVD0DCC13TM3WQS
@misc{mysteriesdept_evd0dcc13tm3wqs,
title = {Without the layout layer, topics follow scribe less},
author = {{Mysteries Department}},
year = {2026},
note = {Evidence, catalog key EVD0DCC13TM3WQS},
url = {https://mysteriesdept.com/r/EVD0DCC13TM3WQS}
}