The Voynich Ninja

I'm curious how the generator theories can account for the positional tendencies people here keep finding, and I didn’t want to blow up someone else’s thread about their generator. I'm not arguing against a generator, I just can't picture one rule producing all of them, since they seem to work in different ways.

Starting right at the front of the VMS, the two paragraphs of folio f2r:

[attachment=16276]

What's been noticed by the pros here:

1. Tavie's vertical impact: a line's opener avoids the glyph directly above it, and it's directional (o→q is common, q→o isn't; o-o is avoided but s-s is fine). You can watch it in the opener column, they keep changing all the way down.

2. Anton's opener order (from "Regaining the lost order"): openers advance through a set sequence, k y d ch o q s sh. Anton reads You are not allowed to view links. Register or Login to view. as k-d-q in both paragraphs (which you can see), and says paras tend to open with a gallows. The first paragraph here it runs k→d→q→[c]→o→s→sh.

3. Top row: gallows p/f cluster on the first line of a paragraph (Tavie, Currier, Feaster). Here that's p in ypchol/ypchaiin (line 1) and f in fodan (line 8).

4. Feaster's drift: big picture within a line, sh→ch, qo→o, k→t shift rightward, and word makeup drifts as you go down the paragraph.

5. Line ends: the last word tends to differ from mid words (Currier).

So my question: how does a generator that writes left to right, one line at a time, produce these, especially the vertical one? What ties right drift to opener order to being allergic to verticality… maybe it’s separate habits, coincidence, astral phenomena…





// FWIW here is the rabbit hole of other threads I didn’t want to blow up with this question, and sources for positionality such I’ve been sorting through on this. This started with me just privately trying to aggregate rules and tendencies to harmonize over a nice pour of whiskey:

A. Other positional tendencies I left off the image:

benches (ch/sh) are rare as line-openers but spike later: You are not allowed to view links. Register or Login to view.

first words of lines run longer, second words shorter, Vogt 2012 (plain handwriting effect?): You are not allowed to view links. Register or Login to view.

long-range word correlations look like real topic structure, Montemurro & Zanette 2013: You are not allowed to view links. Register or Login to view.

q-start and m-end lines are anti-correlated (MarcoP and pfeaster in the vertical-impact thread below)

B. Sources for what's on the image:

vertical impact, Tavie: You are not allowed to view links. Register or Login to view.

opener order, Anton: You are not allowed to view links. Register or Login to view.

drift, Feaster 2022: You are not allowed to view links. Register or Login to view.

line as a functional unit, Currier 1976

C. Related threads: 

ch/sh interchangeability You are not allowed to view links. Register or Login to view.

copy/mutate ledger You are not allowed to view links. Register or Login to view.

hoax/generator debate You are not allowed to view links. Register or Login to view.

generators: Timm & Schinner 2019 (self-citation) You are not allowed to view links. Register or Login to view.

Cardan grille, Rugg 2004 You are not allowed to view links. Register or Login to view.

ReneZ 2021 write-up You are not allowed to view links. Register or Login to view.

Schinner 2007, lots here and the stochastic generator observation You are not allowed to view links. Register or Login to view.

I'm not sure I understand the question. It's trivial to adjust a generator to create any positional dependency. As far as I can see, most (all?) generator theories do not really explain anything. I've yet to see a generator theory that would go "so, they implemented this complex system that randomly selects chunks from N lists, because..." Mostly they just attempt to replicate Voynichese using some algorithm. I think I can easily adjust any of the generative algorithms to create a similarity of these positional dependencies, but what for? Some generator theories show that certain features "naturally follow" from certain set of rules, but this again doesn't explain anything, because the set of rules remains absolutely arbitrary and no rationale for its existence is ever provided, at least in most cases. I can't say I have reviewed every generator theory out there. So, I'd say generator theories can trivially incorporate the positional dependencies, but this won't tell us anything new.

(01-07-2026, 06:59 PM)JoeyB Wrote: You are not allowed to view links. Register or Login to view.I'm curious how the generator theories can account for the positional tendencies people here keep finding, and I didn’t want to blow up someone else’s thread about their generator. 

Markov chain generators don't account for positional tendencies. Ie. my generator works purely word-by-word and indeed it fails badly @quimqu's position-based tests. One could tweak these generators to get positional dependencies I guess, but see @oshfdk post above, it'd probably be an exercise in futility over an exercise in futility.

Timm Torsten's generator has explicit rules to account for some of the positional tendencies: iirc it pre-defines where paragraphs will begin in the text and preferentially starts a paragraph with a gallows. It also fares much better than Markov chains at @quimqu's tests, but I think this is due much more to the copy-and-modify mechanism itself than to the 'gallow at paragraph start' rule (which imo just 'beautifies' the text). I think this method has the potential for creating positional tendencies, just I see no smoking guns at the moment.

Thanks, both. Maybe I was over-reading what generator work claims to show. Some of it does come with big conclusions attached, e.g., Torsten's strong version "The high regularities of the VMS text is therefore our key argument for a meaningless text" (You are not allowed to view links. Register or Login to view.) and kostrubaty wrote "Results strongly indicate that the text was generated via simple methodology (order-3 markov chain)" and the "results seem to leave little to imagination" (You are not allowed to view links. Register or Login to view.). 

If the claim is only that a math machine can produce text that looks like this, fair enough and most of my question dissolves. It's the stronger conclusions that seem to run headlong into everything this forum keeps finding about position.

(02-07-2026, 09:51 PM)JoeyB Wrote: You are not allowed to view links. Register or Login to view.Thanks, both. Maybe I was over-reading what generator work claims to show. Some of it does come with big conclusions attached, e.g., Torsten's strong version "The high regularities of the VMS text is therefore our key argument for a meaningless text" (You are not allowed to view links. Register or Login to view.) and kostrubaty wrote "Results strongly indicate that the text was generated via simple methodology (order-3 markov chain)" and the "results seem to leave little to imagination" (You are not allowed to view links. Register or Login to view.). 

If the claim is only that a math machine can produce text that looks like this, fair enough and most of my question dissolves. It's the stronger conclusions that seem to run headlong into everything this forum keeps finding about position.

Claims are claims. There is no consensus. I think the manuscript is likely meaningful and if so it's most certainly a cipher, but I have no direct proof. I don't see the arguments for meaningless as strong enough. What is more, specifically Torsten's results are You are not allowed to view links. Register or Login to view., so I a bit paradoxically see them as evidence in favor of a cipher. In the end, I don't think there is a published hypothesis that would explain the manuscript well enough.

kostrubaty's results don't look conclusive at all to me, I don't understand how anything can be claimed based on these results.