Hypervector Analysis of the Voynich Manuscript

[dvallis]

But I am interested in looking into the other original files as well.

My source files for the Caucasian languages was primarily this site.

You are not allowed to view links. Register or Login to view.

Steps
Download individual PDF files
Convert them to .txt with You are not allowed to view links. Register or Login to view.
Combine the text files
Remove anything not belonging by hand
Transliterate the txt
Clean it up with my "normalize" text

Takes quite a while for for each language

[-JKP-]

dv, you've clearly spent a lot of time hunting things up, so it is appreciated that you share it.

[Aga Tentakulus]

Aga,

These guys dont  understand Schwyzer dütsch.

ohhh, it doesn't matter. They don't understand Voynich either.

And already we have a match.  

[dvallis]

JKP I'm going to do Scenario 3 (Latin-ish) tonight. Do you want v substituted for space ' ' or nothing ''?

Starting with properly capitalized EVA TT. You said:

c/C
e/c
v/^ <= Can't do this. Must be ' ' or ''
n/v
r/er
m/ris
g/cis
q/con
s/cir
p/pro
f/per
y/um


After that I'll convert all to lower case, remove anything not [a-z,' '], normalize and run.

[dvallis]

Here's a plot of the Latin-ish transliteration of TT-EVA proposed by JKP. It still falls near EVA and CUVA. Hypervectors are very resistant to slight letter jiggering it seems.

Hypervector data file is You are not allowed to view links. Register or Login to view.
EVA to Latin-ish transliteration is You are not allowed to view links. Register or Login to view.
Transliteration script is You are not allowed to view links. Register or Login to view.

[-JKP-]

It's very helpful to have some baseline variations and particularly interesting to see how little effect the expansions had on the VMS position for this particular flavor of computational attack.


If you had asked me to guess where VMS might end up before showing us any of these graphs, I might have guessed that it would end up near Kannada (it does have a few properties in common with this language group), but that is apparently not the case (at least not with the EVA/CUVA/Expanded versions).


Of course, the "a" and "o" and other familiar shapes might be a smoke-screen. The vowel shapes might be consonants and the consonant-like shapes might be vowels (or they might all be numbers), we really don't know, but even swapping vowels and consonant-like shapes might not shift the overall position very much.

[-JKP-]

I just took a look at the Agul/Aghul language. I can't read a single word of it!

[dvallis]

I'm encouraged by the algorithms emergent behavior. I literally had no idea what would happen after feeding in language files. The fact it self organizes data into recognizable language groups means valid information is definitely being extracted. I have no idea what to do next with Voynich being similar to Caucasian languages, but hey, its a data point.

[dvallis]

I just took a look at the Agul/Aghul language. I can't read a single word of it!

Hey, not an Agul expert. If you provide a properly transliterated sample I'll run it. 

The source files are that russian link I posted a while back, run through unidecode.

[-JKP-]

I glanced at the Omniglot Romanized version. Unfortunately, I'm working right now so I can only pop in for a few moments and then I'm gone. I will look at these languages in more depth on the weekend.