(08-07-2026, 08:06 AM)Aga Tentakulus Wrote: You are not allowed to view links. Register or Login to view....
Those examples and your colors suggest that there may be two kinds of
f, distinguished by the position where the curvy stroke starts: at the top of the leg (blue) or at some distance below it (cyan). I don't recall anyone pointing that out before. Are those two kinds of
f distinguished in Rene's IVT file?
I wondered myself whether the presence or absence of a hook at the end of the curvy stroke could be significant. Like,
p without hook =
t,
p with hook =
te. I spent many hours last year re-transcribing the hooked occurrences of
p and
f as w and z, intending to check that theory with statistics. But I never got around to this last step, because I saw signs that the theory was false -- such as there being a continuous gradient between hooked and non-hooked, with no clear separation of the two types.
All the best, --stolfi
[
attachment=16415]
It’s not the only one, after all. The book is full of examples like that.
The differences are the same for both symbols. Two glyphs, four possibilities.
I’m not surprised that no sensible solution can be reached when you’re only working with half the alphabet.
I don’t think Friedman could have worked it out because his source material was of poor quality. I see the same thing in the NSA report.
Deception is the simplest technique in cryptology.
This thread is diverging onto a tangent - let's try to keep it relevant to Vuk's theory. I can see why multiple scribes or varieties of glyphs can be relevant to raise, but now the thread has moved on to debating these when we have other threads for such debates.
There are lots of threads to discuss the question about multiple scribes. Here's one of the You are not allowed to view links.
Register or
Login to view..
I'm sure there have been multiple threads on whether there are subtle but meaningful distinctions in glyphs, though I can't find one. Please feel free to start a new one.
Question about the monte carlo sims, in your 03 stat valid python I see how you're defining any page with a doublet as 'infected' and then have the script take the whole page, remove the duplicate word (so, qokedy qokedy chol daiin -> chol daiin), and carve the page up into 30 word chunks. Is this right? Did you mean for the script to leave one "qokedy" in that example or is it intended to remove both?
Then you take every 30 word chunk and you say if this came from a page that contained a doublet somehwere on that page, model should label it infected. If page didn't contain a doublet, the model should label it clean.
Then your monte carlo (the permutation_test) shuffles the labels and the classifier runs again. But it isn't shuffling voynich words, it's shuffling labels. So you're testing: do 30 word chunks from doublet pages have a different vocabulary profile from 30 word chunks coming from non-doublet pages better than a shuffled label baseline. And I guess, if the model is allowed to see the doublet word as part of the vocabulary profile of the 30 word chunk is tied to whether you're removing one or both instances of th doublet word?
(08-07-2026, 05:29 PM)JoeyB Wrote: You are not allowed to view links. Register or Login to view.Question about the monte carlo sims, in your 03 stat valid python I see how you're defining any page with a doublet as 'infected' and then have the script take the whole page, remove the duplicate word (so, qokedy qokedy chol daiin -> chol daiin), and carve the page up into 30 word chunks. Is this right? Did you mean for the script to leave one "qokedy" in that example or is it intended to remove both?
Then you take every 30 word chunk and you say if this came from a page that contained a doublet somehwere on that page, model should label it infected. If page didn't contain a doublet, the model should label it clean.
Then your monte carlo (the permutation_test) shuffles the labels and the classifier runs again. But it isn't shuffling voynich words, it's shuffling labels. So you're testing: do 30 word chunks from doublet pages have a different vocabulary profile from 30 word chunks coming from non-doublet pages better than a shuffled label baseline. And I guess, if the model is allowed to see the doublet word as part of the vocabulary profile of the 30 word chunk is tied to whether you're removing one or both instances of th doublet word?
@JoeyB
To avoid confusion with the old data and to start fresh with the new statistical infrastructure, I have opened a brand new dedicated thread for this.
I have just replied to your question in detail over there.
You can find my answer and the full updated paper here: You are not allowed to view links.
Register or
Login to view.
Alfredo