The Voynich Ninja

(01-02-2026, 03:48 PM)Koen G Wrote: You are not allowed to view links. Register or Login to view.Character entropy is a major problem if you can't come up with a spelling system that would account for it.

Indeed, for a language proposal to be taken seriously, one would have to propose also its spelling system.

But there is no value of the character entropy that makes that task impossible.  For instance, oonnee ccaann rreedduuccee tthhee cchhaarraacctteerr eenntttrrooppyy ooff aannyy llaanngguuaaggee in half simply by repeating every characater.   And oNE cAn IncREmENt tHE CHaRactEr EntROpy BY onE simply by alternating between upper and lower case at random. Or increment it by ~2 by ixnasfewrctjidnqgo a random letter after each letter. 

The per-character entropy of German would increase noticeably if one replaced "sch" by "x", other "ch" by "q", "ck" by k, "ie" by "i" etc. 

If Bavarian indeed is close to monosyllabic, and its entropy per word is typical of other languages, then its per-character entropy must be similar to that of Voynichese -- because, very roughly, the sum of the entropy of all characters of a word must be the word entropy.

All the best, --stolfi

(01-02-2026, 08:45 AM)Aga Tentakulus Wrote: You are not allowed to view links. Register or Login to view.3. The proof of the pudding is in the eating. Rubbish in, rubbish out until you get a result.


That's a VM word ‘4o89’. How many possibilities are there?

@Aga Tentakulus,

… hmmm! "You are the Goddess of your Country"! How Interesting! That sounds like saying in a nutshell what the Voynich manuscript is about!

[attachment=13883]

@ Koen 


Actually I’m not claiming that Bavarian by itself explains the low character entropy.
My point is that Bavarian lowers the amount of entropy the cipher must eliminate in order to reach a Voynich-like Style.

This works because Bavarian (especially in a strongly reduced phonetic spelling) has very short word stems, has many monosyllabic function and content words,
already shows strong vowel reduction, and, in addition, frequent consonant clusters, which can be captured by "class markers" such as Gallows.

As a result, the information per word is concentrated in very few segments, allowing a historically plausible substitution system to collapse the character variability without resorting to highly artificial and anachronistic cipher structures.

Quote:My point is that Bavarian lowers the amount of entropy the cipher must eliminate in order to reach a Voynich-like Style.

Correct me if I am wrong but I feel it is the opposite.

High entropy = disorder, randomness, low entropy = predictable patterns.

If I understand correctly Bavarian is a kind of shortened German. It eats some vowels, I suppose "bringen" may become "bringn". So it works a bit similar to Latin abbreviations.

I believe it was shown here on these forums for Latin abbreviations that they actually make the entropy higher.

It could be the same for Bavarian.
"Bringen" has "en" in the end. "N" after "E" is quite predictable -> lower entropy
"Bringn" has "gn" in the end. "N" after "G" is not so predictable -> higher entropy

---------------
By the way, I know only what I can Google but it feels to me that Bavarian is not so monosylabic.
There is actually a Bavarian Wikipedia 
You are not allowed to view links. Register or Login to view.

Let's take the featured article: You are not allowed to view links. Register or Login to view.

Da Fosanegl oda Fasenickl is a boarische Fosnochtsfigua. Ea drogt a ästhetische Hoizmaskn und aufwendige, barocke Gwanda. 
Ea is a majestätische, prochtvolle Gschtoit, de wo a weng an an venezianischn Karnevai erinnad.
"Nigl/Negl" (dt. Nickel) moant im Boarischn an Koboid (vgl. Bosnigl), "fosln" (faseln) hoaßt unvaständle bzw. sinnlos dahearedn.

De Hoizmaskn vom Fasenickl muaß aus an Lindnhoizblock gloddlarvig aussagschniazt wean. Da Gsichtsausdruck is weda lustig no grauslich, eha a weng obghom. Obwoi ra an Kuglschnuaboart hot, san de Gsichtszig a weng feminin: hoche Stian, kloane Schtubsnosn, kloana Mund und millifoarbane Haut.


It is not really monosylabic. Would medieval Bavarian be different?

(01-02-2026, 04:56 PM)MHTamdgidi_(Behrooz) Wrote: You are not allowed to view links. Register or Login to view.That sounds like saying in a nutshell what the Voynich manuscript is about!

Google translate is able to hallucinate as well as chatbots. Aga knows but he likes it.

@ Rafael

Two points on this:

1. Your example describes something using many words that are, how shall I put it, rather modern, some of which are borrowed from other languages, such as: aesthetic, Fosnachsfigua, baroque, majestic, Venetian, carnival. (lat: carne levare)

The Bavarian dialect of the Middle Ages was much less wordy and was spoken differently. You should therefore use a text that corresponds to the period, as I did with "Ortloff von Baierland".

But let's take a text from Wikipedia that fits the period. For example, the description of a plant.

Here is the description of ‘salat’ in Bavarian “salod” english "salad"
You are not allowed to view links. Register or Login to view.

Salodpflanzna nennt ma olle Pflanzna, aus dena a koids, unkochts Essn gmocht wead, des wo ma Salod nennt. Des san oiso koa botanische Gruppn vo Pflanzna sondan a reine Zuaordnung vo da Kuchl.

And here too the language is even more modern, such as: botanical, in the past probably more ‘"nach der natur" – in the past, people would not have said "salodpflanza", but only salod. ‘Zuardnung’ is also rather modern.

2. In addition, I clearly stated that it was written phonetically, as was not uncommon at the time.

And then it would have come out as medieval (botanical was replaced, Salodpflanzna= Salod, Zuordnung stays).

Salod nent ma all Pflanz, aus dena kouds unkocht essn gmocht wead, des woma salod nennt. Des san oiso koa Gruppn nach der natur, sondan a rain Zuaordnung vonda kuchl.

And here it fits reasonably well again.

But people at that time would probably have expressed it differently. That is why texts from the Middle Ages should be used for comparison. 

(01-02-2026, 07:48 PM)Rafal Wrote: You are not allowed to view links. Register or Login to view.I believe it was shown here on these forums for Latin abbreviations that they actually make the entropy higher.

That should be the case for the character entropy if the bits that are abbreviated are predictable and the abbreviations are systematic and invertible. Like, when one always replaces "-us" by "-9".  That's because (roughly) the word entropy will not change but the average chars per word will decrease, so the average entropy per char will increase.

Otherwise abbreviations may lower the entropy. E.g. if one abbreviates all of "-able" -"eble" "-ible" "-oble" and "-uble" and "-ably" etc. by "-bl" one is actually deleting ~3 bits of information from those tokens, that the reader would have to supply from context or subtle grammatical/lexical info.  The word entropy would be lower; whether the average character entropy is lower or not depends on some arithmetic...

But, again, the character (digraph etc) entropy is not a very useful statistic, because it depends on the spelling system.  The word entropy is more useful, as it is independent of the spelling (provided the same word is spelled the same way most of the time).

And, again, there is no such thing as the "statistics of language X", whether character, word, or anything else.  Such statistics are properties of texts.  

Someone once wrote an entire novel in English (or French, not sure) without using the letter "e" even once.  And AFAIK readers who are not told about it do not notice that big hole in the statistics...

All the best, --stolfi

Stolfi: when people here talk about entropy, it's alnosy always about character entropy. And this is incredibly relevant, because as your own examples show, regular spelling variation cannot account for Voynichese statistics. You are venturing into verbose cipher and obfuscation territory.

(01-02-2026, 01:03 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.

(01-02-2026, 01:15 AM)ReneZ Wrote: You are not allowed to view links. Register or Login to view.The low conditional entropy (or the bigram entropy) of the Voynich text is a consequence of this word structure

The character entropy is not significant because it depends on the spelling system, which we do not know.  The low entropy does not imply "it is not European".  It could be simply the result of many phonemes being encoded with 2 or 3 EVA characters (like German using "sch" for the "ʃ" sound).  


All the best, --stolfi

As Koen also said, it is significant as it tells us something about what the characters could or could not be.
Indeed, as I tried to point out earlier in reply to JoJo_Jost, this does not say anthing about the source language.
Whatever was done to it, obfuscates the differences between such different source languages.

A verbose encryption is a possible explanation. Or perhaps potential is the better word, as we don't know if it is really possible.

(01-02-2026, 11:46 PM)Koen G Wrote: You are not allowed to view links. Register or Login to view.as your own examples show, regular spelling variation cannot account for Voynichese statistics

I gave those extreme examples only to show how spelling affects the character entropy.

But here is what I think are the actual characters of the Voynichese script:
    {q} {o} {a} {y} {d} {r} {l} {s}   
    {ch} {che} {sh} {she} {ee} {eee}
    {k} {ke} {t} {te} {p} {pe} {f} {fe}
    {ckh} {ckhh} {ckhe} {ckhhe}
    {cth} {cthh} {cthe} {cthhe}
    {cph} {cphh} {cphe} {cphhe}
    {cfh} {cfhh} {cfhe} {cfhhe}
    {n} {in} {iin} {iiin}
    {m} {im} {iim} {iiim}
        {ir} {iir} {iiir}

Except that the circles {o} {y} and maybe {a} may be modifiers for the above, so instead of just {ch} one may have also {cho} and maybe {chy}, and instead of just {che} one have also {cheo} and {chey}, and instead of just {d} one may have also {do} or {dy}, etc.

If one computes the character (or digraph) entropy of Voynichese by counting each of those elements as "one character", one will get a much higher entropy per "character" than if one computes it using EVA letters as the "characters".  Possibly twice as much, or more.

All the best, --stolfi