(06-01-2026, 09:47 AM)Koen G Wrote: You are not allowed to view links. Register or Login to view.Definitely, well done Michael! I also like that the writers seem to get some of the nuances correct, rather than going full clickbait. There is hope yet.
Thanks, everyone! In my correspondence with (my fellow) journalists, I have tried hard to ensure that the nuance comes across.
I have good news to report!
The Naibbe cipher seems to support the apparent right-to-left "reading" directionality I reported in this paper: You are not allowed to view links.
Register or
Login to view. (there is no doubt the manuscript was written left to right, I'm talking about reading right to left).
I run a few tests on the glyphs found in Naibbe's unigram encoding table, here are the first conclusions:
- Left-descender 'l': Appears 4.8× more at word ENDINGS (19 vs 4)
- In RTL reading, endings are on the visual left → RTL bias
- Right-descender 'r': Appears 20× more at word ENDINGS (20 vs 1)
- Also suggests RTL visual flow
- Gallows 'q': Appears exclusively at word STARTS (35 vs 0)
- In RTL reading, starts are on the visual right where gallows are drawn → Strong RTL bias
Thanks to Naibbe, we could have the beginning of an explanation: the original clear text (in Latin for instance) could be first written LTR on a wax tablet by the scribe, then encrypted with a Naibbe cipher mimicking the cleartext direction. Finally the cipher would be written LTR on velum, but reading each glyph to be copied... from right to left.
It could make sense to add this level of obfuscation for added confidentiality (in medieval times, not today...).
Thoughts?
The table I used is: You are not allowed to view links.
Register or
Login to view.
(07-01-2026, 05:34 PM)Labyrinthinesecurity Wrote: You are not allowed to view links. Register or Login to view.I have good news to report!
The Naibbe cipher seems to support the apparent right-to-left "reading" directionality I reported in this paper: You are not allowed to view links. Register or Login to view. (there is no doubt the manuscript was written left to right, I'm talking about reading right to left).
I run a few tests on the glyphs found in Naibbe's unigram encoding table, here are the first conclusions:
- Left-descender 'l': Appears 4.8× more at word ENDINGS (19 vs 4)
- In RTL reading, endings are on the visual left → RTL bias
- Right-descender 'r': Appears 20× more at word ENDINGS (20 vs 1)
- Also suggests RTL visual flow
- Gallows 'q': Appears exclusively at word STARTS (35 vs 0)
- In RTL reading, starts are on the visual right where gallows are drawn → Strong RTL bias
Thanks to Naibbe, we could have the beginning of an explanation: the original clear text (in Latin for instance) could be first written LTR on a wax tablet by the scribe, then encrypted with a Naibbe cipher mimicking the cleartext direction. Finally the cipher would be written LTR on velum, but reading each glyph to be copied... from right to left.
It could make sense to add this level of obfuscation for added confidentiality (in medieval times, not today...).
Thoughts?
The table I used is: You are not allowed to view links. Register or Login to view.
Interesting work, thanks!
I was building statistics for the trigrams to find any correlations, and ended up rendering everything as green pixels for any significant trigrams. Horozontal dimensions are individual pages (f1r through f166v), vertical are the trigrams.
[
attachment=13565]
So to support your hypothesys, there would be a strong correlation in behavior between an upper line and a few lines below it. Lines 1 and show strong correlation, those are iin (4341 count) / aii (4311), and another one is line 3 (edy, 4226 count), 12(ched, 1578), 13(eed, 1479). Additionally, 5 (oke) and 6(kee) show some signs of correlation as well, but not as clean.
Here is another pixel visualization of the 'edy', 'ched' and 'eed' occurences word by word, every horizontal line is a single page:
[
attachment=13566]
Other than those I wan't able to find any significant correlations
Kinda cool that the first 48 pages lack those three trigrams altogether (that wast reported before btw)
PS: to view the detailed picture, you may want to download those images and open in a standalone viewer to zoom it in
Subject: Three order-level measures, from someone who arrived at your architecture independently
Dear Mr Greshko,
I measured the Voynich text (ZL3b) this year and then tried to build a cipher
matching what I'd measured, working without reading the literature so as not to
fit my expectations to other people's. Chunked plaintext, several tables chosen at
random, weighted to match glyph frequencies, multiple spellings per unit — I
converged on very nearly your architecture, and was glad to find it already done
properly.
Three measures defeated my cipher and, it turns out, yours as well; I offer them
as shared ground rather than criticism. I used your `nathist_output_ciphertext.txt`,
cutting both texts into the manuscript's own line lengths so line structure is
matched.
First, though, what your cipher does better than mine did:
measure Voynich Naibbe
h2 (digraph-merged) 2.682 2.581 4% off
mean word length 4.95 5.23 6% off
Zipf slope -0.973 -0.895 8% off
TTR / types 0.309/3706 0.277/3322 10% off
Now the three.
Adjacent word repetition, measured against a within-line shuffle — the same words
of the same line, reordered. The choice of null turns out to matter more than
anything else here: against a global shuffle the manuscript looks anomalous, and
against the line-level one it doesn't.
Voynich 274 observed / 269.8 expected = 1.02x [95% CI 0.92-1.13]
Naibbe 45 observed / 101.5 expected = 0.44x [95% CI 0.38-0.53]
The manuscript sits at chance. Naibbe suppresses adjacency, much as natural
languages do (Latin 0.02x). If this is right, the manuscript's oddity is the
absence of a suppression every language has, rather than an excess of repeats —
which took me an embarrassingly long time to see.
Second, the word junction: mutual information between the last k glyphs of a word
and the first k of the next, minus the same quantity under a word-order shuffle,
the subtraction being there to remove estimator bias.
k=1 excess: Voynich 0.194 | Latin 0.048 | English 0.071 | Naibbe 0.0021
ratio k=3/k=1: Voynich 1.25x | Latin 9.07x | English 10.51x | Naibbe 32.4x
The manuscript's constraint appears to sit on the final glyph rather than on the
ending as a whole: one pair — word-final y followed by word-initial q — carries
about 48% of the junction information, and I found no absolute prohibitions.
A caution I owe you here, since it nearly went into this letter as fact: at 6-12k
words I measured that ratio as 0.63x and believed for a while that the manuscript
was the only corpus below 1. It was sample-size bias in the 3-glyph estimate. The
figures above are the full manuscript with matched samples, and the qualitative
gap survives, but the striking version of the claim did not.
Third, word-length autocorrelation at lag 1. This one I owe to the voynich.ninja
discussion of your preprint, where it was raised and not, so far as I could see,
followed up:
Voynich +0.134 [+0.122, +0.145] Naibbe -0.031 [-0.043, -0.020]
Latin -0.029, German -0.040, English -0.137, Greek -0.151, Italian -0.161
Languages alternate word lengths; the manuscript clusters them, and it is the only
corpus I have measured with a positive value. Strongest in the Stars section
(+0.186), weakest in the Herbal (+0.082).
Two things I should be straight about. The LAAFU comparison isn't meaningful for
either of us, since neither model has a line model at all and both get measured on
artificial line cuts — I mention it only so you know I looked. And on length: your
9,243 plaintext words give 34,764 ciphertext words, so the manuscript's 34,167
would carry roughly 9,000 words of Latin, about twenty book pages across 225
leaves. That may well be fine for a heavily illustrated herbal; I raise it as an
open question rather than an objection, and I don't know how to weigh it.
What strikes me is that all three are about order — what follows what — while the
construction of individual words, which is the harder-looking problem, your cipher
already handles. Whether that is a useful direction or a dead end I genuinely
don't know.
Methods and code are yours if they'd help, and I'd be glad to be told where I've
gone wrong.
With thanks for the paper,
"two things I should be straight about""... Thank you Claude..
Claude slop or just translation? I can't tell about this kind of material.
Despite my misgivings about the possibility of AI-generated text, the underlying findings strike me as plausible enough. The finding of word-length autocorrelation was noted in Bowern and Gaskell's 2022 conference work, and in my paper last year (IIRC in Supplementary Material 1), I noted the low frequency of word reduplication within Naibbe ciphertext relative to the VMS.
At bottom, the big statistical limitation of the original Naibbe cipher is that each draw of a word (when encrypting standalone letters) or of a prefix/suffix (when encrypting letter pairs) is memoryless and independent. This was a known limitation from the beginning; I reported the absence of long-range correlations in Naibbe ciphertext for a reason. To get the VMS's long-range correlations and other observed features, there has to be some kind of memory at play when generating Voynichese tokens.