+- The Voynich Ninja (https://www.voynich.ninja)
+-- Forum: Voynich Research (https://www.voynich.ninja/forum-27.html)
+--- Forum: Analysis of the text (https://www.voynich.ninja/forum-41.html)
+--- Thread: Voynich text generation (/thread-2684.html)
RE: Voynich text generation - geoffreycaveney - 16-03-2019
(16-03-2019, 09:00 AM)-JKP- Wrote: You are not allowed to view links. Register or Login to view.I didn't want to try this until I was back home, but I simply can't resist.
Okay, this is NOT good Voynichese, but I did this in less than 15 minutes and I know if I had an hour or so to work on it (which I don't right now) I could do much better, but here is a quick and dirty example...
These are grammatically correct sentences in a major medieval language. The passage is from a well-known text:
Thank you Rene and JKP for posting these examples.
JKP, the first things that jump out at me upon my first reading of your text, are the word-final characters in [olcth], [olsh], and [yt]. It is very rare to have word-final [-h] (any [-h] character) or word-final [t] (or any gallows character) in the actual Voynich ms.
RE: Voynich text generation - -JKP- - 16-03-2019
Geoffrey, I know this. I am so intimately acquainted with VMS text that I can see at a glance which patterns are "legal" and which are not.
But René's challenge was for us to write READABLE/TRANSLATABLE text (not simply glyph combinations that look like Voynichese as in the Hyde/Rugg and Timm algorithms) and the text I posted above is grammatically correct text in a medieval language and I did it in less than 15 minutes.
Can you create readable, translatable text (at least a couple of sentences) in a medieval language that looks like Voynichese (or at least very close to Voynichese) in less than 15 minutes?
RE: Voynich text generation - geoffreycaveney - 16-03-2019
JKP, your text was impressive! Especially in 15 minutes. I'd love to try to decipher it, but that would be quite a challenge with such a short amount of text. Likewise with Rene's encrypted Italian text.
Sure I can create a couple sentences in a medieval language that looks like Voynichese in less than 15 minutes:
The language is Judaeo-Greek written in the Hebrew script.
The text is
"eiApan tis ipeirous otan skiiAis tis , epan oAn
par Atous & Atees tAr oikous ok-eis tees , tes eiApan"
And the Voynichese is
"teeodaiin shey epairody osaiin yteeoey shey epaiin oaiin
[font=Arial, Helvetica, sans-serif]daiir okeody qoekeeg sar oeteody oteey keey key keeodal[/font]"
Just kidding, folks, just kidding!!
RE: Voynich text generation - DonaldFisk - 17-03-2019
In a few minutes , I was able to generate this: "qodom qodeey otaiin olk otaiin shodeey okeey otaiin qodom qodeey otaiin otaiin ykain shotol qotol otaiin ydaim aiin ldor qotar qokaiin okos chokeey qokaiin aiin ldor qotar qokaiin otary ty odol qokaiin shodain tal okaiin olk otaiin chokain shotol qotol dam qokar qokaiin shodeey okeey otaiin aiin shodain qotain dain dal aiir chokair qold otey ldy dain daiir odol qotey otaiir okeey otaiin otaiin tal otar ydol tor qodol dam qokar qokaiin qokaiin al oiin oteey okal otaiin yky das chokor chodol dol chodaiin shodor qotaly choky doiin d qokal qokaiin shodol olkar qokal aiin qoty dam otor dam qodor qodain d chokar olkain deey ary tain daiir odol qotey otaiir aiin qokaiin aiin shodain qotain dar chokain ytol".
I'll tell you how I did it. Each pair of plaintext letters was encoded into a single Voynichese word. The first letter was the prefix, the next letter was the suffix. Spaces between words in the plaintext are significant, and diacritics were removed before encryption. Plaintext letters, and Voynichese prefixes and suffixes, were ordered roughly in descending order of frequency. So, it's a verbose cipher. It might be too short to be breakable, but it's the beginning of a well-known text, roughly contemporary with the Voynich Manuscript, in a well-known European language.
RE: Voynich text generation - -JKP- - 17-03-2019
Donald, I was thinking some people might like to see this in the original Voynich font, so I have converted it:
Quote:DonaldFisk: In a few minutes , I was able to generate this: "qodom qodeey otaiin olk otaiin shodeey okeey otaiin qodom qodeey otaiin otaiin ykain shotol qotol otaiin ydaim aiin ldor qotar qokaiin okos chokeey qokaiin aiin ldor qotar qokaiin otary ty odol qokaiin shodain tal okaiin olk otaiin chokain shotol qotol dam qokar qokaiin shodeey okeey otaiin aiin shodain qotain dain dal aiir chokair qold otey ldy dain daiir odol qotey otaiir okeey otaiin otaiin tal otar ydol tor qodol dam qokar qokaiin qokaiin al oiin oteey okal otaiin yky das chokor chodol dol chodaiin shodor qotaly choky doiin d qokal qokaiin shodol olkar qokal aiin qoty dam otor dam qodor qodain d chokar olkain deey ary tain daiir odol qotey otaiir aiin qokaiin aiin shodain qotain dar chokain ytol".
You've made an interesting point... it is much easier to generate Voynich text by turning them into big chunks (full tokens, a highly verbose code).
For those who are interested, my example is not done that way. I actually tried to incorporate some aspects of how I think Voynichese might be encoded. I could be wrong about them (and I didn't have sufficient time to do a really thorough and completely "legal" job of it), but for the time allotted, I tried as best I could.
RE: Voynich text generation - Emma May Smith - 17-03-2019
Donald, I have to say I'm impressed with your text. At first glance it seems very convincing and it's only with some study that a few (and only a few) doubts emerge.
RE: Voynich text generation - -JKP- - 17-03-2019
If Don is taking entire Voynich tokens to encode the letters, then the tokens are "legal" from the get-go.
If the VMS were created this way, then the amount of text in it would be very small.
What one has to look for in a token-based reproduction is sequence and length. That's where it starts to crack at the seams, but sequence is also more difficult to quantify.
Most computational attacks look at individual tokens or at generalized statistics. Not many have aimed at mathematically describing the sequence of tokens—what has to follow what.
RE: Voynich text generation - ReneZ - 17-03-2019
Indeed.
My experiments are also character based. I have a second iteration, below. The source text is, as before, from Dante. The method does not generate any of the following characters:
g m f k b
They are all considered to be variants of other characters to be used in special cases. (I consider the same to be true for n by the way.)
Anyway, here goes:
Quote:ol ar dar dchear sey dqokol sol okshar sheol ckhol dotokaiin ol
otaiin doty okaiin oteol y dsol chear ddar aiin ckhey choteain dotar
sheey chaiin ckhy okaiin shy ckhol aiin aiin seseaiin dsar daiin ol
sheaiin otear ol sheeol ctheain ol aiin choteol y sain dshar chchy cthol ain dotar
chaiin sar ckhy qoty oksheol daiin ckhaiin ckhey chchaiin ckhaiin chchy
sheey sheaiin ckhy aiin sol oteaiin ain okaiin ol chshar okshar sheoteain dotar
ol ar dar dchear sey dqokol sol okShar Sheol cKhol dotokaiin ol
otaiin doty okaiin oteol y dsol chear ddar aiin cKhey choteain dotar
Sheey chaiin cKhy okaiin Shy cKhol aiin aiin seseaiin dsar daiin ol
Sheaiin otear ol Sheeol cTheain ol aiin choteol y sain dShar chchy cThol ain dotar
chaiin sar cKhy qoty okSheol daiin cKhaiin cKhey chchaiin cKhaiin chchy
Sheey Sheaiin cKhy aiin sol oteaiin ain okaiin ol chShar okShar Sheoteain dotar
I chose Italian for a particular reason, but it creates a problem from the fact that it has numerous doublets, or at least the Dante text has. Therefore, there are quite a few unusual cases of chch and dd , both of which can occur, but only quite rarely.
The most interesting part (for me) is that the inverse method, i.e. the one that brings back the original Dante from the above, is capable of converting 96% of genuine Voynich MS text.
The best part is that the entropy of this converted text is almost 'normal' (not quite).
Unfortunately, it does not convert it into readable Italian, but into meaningless strings.
This was also not expected, because I have not taken into account a lot of important things, mainly the unusual line-start behaviour.
There is also no explanation for the variation between Currier languages, and with this type of method, the biological text is almost inevitably nonsensical.
RE: Voynich text generation - geoffreycaveney - 17-03-2019
(17-03-2019, 04:50 PM)ReneZ Wrote: You are not allowed to view links. Register or Login to view.Indeed.
My experiments are also character based. I have a second iteration, below. The source text is, as before, from Dante. The method does not generate any of the following characters:
g m f k b
They are all considered to be variants of other characters to be used in special cases. (I consider the same to be true for n by the way.)
I believe you meant to say your method does not generate "g m f p b", because it does generate [k].
RE: Voynich text generation - -JKP- - 18-03-2019
My translation is not quite character-based, it's Voynichese pattern-based. It's not token-based like Don's method.
Since René chose Italian, I chose something else (mainly to make it interesting). If I have time when I get back, I'll try it with a couple of other languages to see what happens.
I did not make a distinction for m as a separate character. I treated it the way I suspect m is used in the VMS and the way certain symbols are used in scribal Latin. It's possible it is a separate character and yet still works the same way. I need to look at this more specifically to see if I can figure which way it leans in Voynichese. I've studied most of the glyph patterns pretty intensely, but I put m on the back-burner until I got some of the other issues sorted out.