The Voynich Ninja

(26-09-2026, 05:20 PM)RadioFM Wrote: You are not allowed to view links. Register or Login to view.The fact of the matter is there's no apparent grammar in Voynichese, the prefix-suffixes rules are haphazardly ignored every now and then, and AFAIK it's the only manuscript of its time (that we know of) that would've used such an unusual/anachronistic cipher to encode a whole book. Matching plants and labels have yielded basically nothing in terms of decipherment.

Although it could be a cipher or a conlang, there's kind of enough evidence that points to gibberish as well.

In my opinion, these are just speculations without a clear and coherent conclusion.

I think it’s the "Voynich curse" — the manuscript could well be anything you can imagine.

(25-09-2026, 11:59 PM)npcompl33t Wrote: You are not allowed to view links. Register or Login to view.it is looking to me like Pharma represents more of an in-between point between A and B. 

...

So the order would be Herbal A -> Pharma -> Herbal B

That's consistent with You are not allowed to view links. Register or Login to view. of bigram frequencies, where PCA1 (X axis) basically encodes A vs B and Pharma (orange/yellow points) falls at the center (X~=0).


[attachment=17914]

(25-09-2026, 11:59 PM)npcompl33t Wrote: You are not allowed to view links. Register or Login to view.chy, while common in a, mostly vanishes in Pharma, and Pharma seems to inject e before o, but it has not fully adopted the common -edy form in other sections.

That's also Torsten's point of view. E.g. You are not allowed to view links. Register or Login to view..

torsten Wrote:The table demonstrates a shift from 'chol/chor' via 'cheol/cheor', 'cheo/sheo', 'chey/shey' to 'chedy/shedy'.

(26-09-2026, 07:16 AM)JoJo_Jost Wrote: You are not allowed to view links. Register or Login to view.Here's another chart on this topic; as you can see, the pharmaceutical sector aligns very well with A...

Thanks for the clear chart!  

But are you including labels in the statistics?  What about parag-initial lines?

Changes in character and digraph statistics may be due to several causes, including different 

1. languages or dialects,
   
2. spelling rules, 
   
3. scribal quirks ,
   
4. overall prose styles, 
   
5. or subject matter. 
   

Or some mix of them.

None of these causes would result in a simple set of character-based mapping between languages A and B.  

Even for very similar languages (#1) the mapping is likely to be very irregular.  Here are the counts and frequencies of occurrence (in %) of the 20 most frequent words in the Portuguese novel Dom Casmurro and in a (human) translation of the same novel into Spanish:
 2681  4.1 que     3001  4.3 de
 2490  3.8 a       2549  3.7 la
 2187  3.4 e       2509  3.6 que
 1971  3.0 de      2086  3.0 y
 1673  2.6 o       1607  2.3 a
 1530  2.3 não     1558  2.2 el
  775  1.2 um      1508  2.2 no
  664  1.0 os      1309  1.9 en
  663  1.0 é       1093  1.6 me
  628  1.0 da       881  1.3 los
  609  0.9 mas      800  1.1 un
  593  0.9 do       766  1.1 se
  571  0.9 se       703  1.0 con
  555  0.9 era      670  1.0 las
  543  0.8 para     631  0.9 lo
  537  0.8 as       602  0.9 mi
  536  0.8 com      583  0.8 pero
  533  0.8 eu       544  0.8 por
  490  0.8 me       517  0.7 le
  461  0.7 em       511  0.7 una
The approximate mapping for the words on the left would be
  que    que
  a      la | a
  e      y
  de     de
  o      el | lo
  não    no
  um     uno
  os     los
  é      es
  da     de la
  mas    pero
  do     de lo
  se     si | se
  era    era
  para   para
  as     las
  com    con
  eu     yo
  me     mi | me
  em     en
with complications like oblique pronouns being separate words in Portuguese but often joined to the verb in Spanish.

Even a spelling rule change (#2) will probably be more complicated than "every instance of ed is now written de".  It could be like "the plural  of a noun ending in consonant plus -y shall henceforth end in -ies instead of -ys, so "babys" is to be spelled "babies"; unless the letter before the -y is a vowel "u" pronounced as  "w",  like "soliloquy", or the noun is a compound of non-noun words like "flyby".  This change does not apply to verbs in the 3d person like "stays", nor to possessives like "baby's"".

As for scribal quirks (#3), I am all but certain that the glyph m is an abbreviation of iin (and maybe of other groups) that the Scribe could use when space was tight (like at the end of a line) or when he just felt lazy.  And I suspect that many ee may  be instances of Ch whose ligature faded away or was omitted by the Scribe; or, conversely, that Ch is basically an ee with a ligature added by to avoid ambiguity.  And I also strongly suspect that the Scribe often misread the Author's -iin as -ir, and his d as k or l or other glyphs. Thus some of the A/B differences may be changes in the frequency of these quirks -- e. g. if the Author's handwriting improved or got worse between his drafting or Herbals A and B.

As for (#4) and (#5), compare the frequencies (%) of the 20 most frequent words in Culpeper's herbal, Marco Ponzi's English translation of the Alchemists Herbal,  Well's War of the Worlds, and Agatha Christie's Affair at Styles:
  8.0 the    6.2 the      4.7 the    8.0 the     
  5.3 and    6.0 and      3.0 i      4.2 and     
  3.6 of     4.0 it       2.4 to     3.8 of     
  2.5 in     3.0 this     2.2 of     2.7 a       
  1.9 to     3.0 of       2.1 a      2.1 i       
  1.8 it     2.8 herb     2.0 and    2.0 to     
  1.7 or     2.4 for      1.8 it     1.7 in     
  1.6 a      2.1 in       1.7 that   1.4 was     
  1.6 is     2.0 will     1.6 was    1.3 that     
  1.0 with   1.9 take     1.6 you    1.1 it       
  0.8 that   1.7 a        1.5 in     1.0 had     
  0.8 as     1.6 be       1.3 he     0.8 my       
  0.8 are    1.5 to       1.0 is     0.7 as       
  0.8 for    1.4 with     0.9 not    0.7 with     
  0.6 them   1.3 is       0.9 had    0.7 at       
  0.6 be     1.3 if       0.9 she    0.7 he       
  0.5 by     1.3 them     0.8 her    0.6 on       
  0.5 also   1.2 or       0.8 but    0.6 were     
  0.5 which  1.1 on       0.7 his    0.6 for     
  0.5 but    1.0 healed   0.7 at     0.5 from
The first half dozen words in these lists have similar frequencies.  But note the word "I" in the last two texts, which hardly occurs in the first two.  That is an example of general style difference (#4) -- namely, the last two books are first-person narratives, while the first two are more impersonal.  Note also the anomalously high frequency of "this", "herb", "take", "will"m and "healed" in the second column, which is a consequence of its highly formulaic style ("herb XXX: for YYY, take this herb and they will be healed").  This is an example of differences due to both style (#4) and contents (#5).

All the best, --stolfi

(26-09-2026, 07:16 AM)JoJo_Jost Wrote: You are not allowed to view links. Register or Login to view.I agree with Stolfi.
Here's another chart on this topic; as you can see, the pharmaceutical sector aligns very well with A...

Your charts just arent graphing the correct things. Pharma has almost no words that end in -chy, and contains far more -eo- in the middle of words than the other sections. It is very apparent it maps into a different space when looking at the PCA, rene also agrees here in the work he published on his currier expansion, as MarcoP said.

Im working on a webapp to allow people to explore the rules i've created as well as ones they want to create, I have to go out of town tomorrow, but i'm hoping ill be able to share it. 

One of the issues is that -or and -ol cannot all be turned into -edy, there is too much unexplained residual mass left over. Chol and Chor collectively occur at like 2-3x chedy's rate each, so there has to be something else capturing at least a portion of that. The yellow portions of the bars in the chart below represent missing mass when trying to map the chor/chol family into chedy.

[attachment=17926]

I think nablator is right and to make all of this work likely requires some sort of contextual merging, maybe like a type of proclitic, particularly involving -or, ol, and possibly the ain/aiin families, either into individual or merged forms with other words. So maybe some of chol / chor gets absorbed as 'or' or 'ol' as standalone words or mapped as prefixes. 

The web-app i'm creating has a 'translate' function that applies rules to a given page, for example, f2r, and tries to 'translate' it into Herbal B, or pharma. 

On reading this, it actually starts to sound quite similar to bio, words that would have had different endings all start to sound similar. See the line below, line 7 from f9r, so maybe something like this is actually occuring, and possibly could explain why bio in particular seems so repetitive.  The -edy ending is taking different word forms, maybe something like qokol, qokchor, etc and turning them into qokedy.

[attachment=17922]

While i think it is possible to track how certain masses of words become other words, it seems the rules are a bit more complex than simple bigram replacement and may involve context from adjacent words, which is a known voynich trait anyways with respect to q and y.

actually interpreting why certain words become others, especially when multiple options exist (tense, abbreviations, or any of the other items from Solfi's list) is a more difficult task, that I could definitely use help with.

Here is the current rule set i am experimenting with (which has expanded a bit from the initial 2), applied to the same chart Jorge_Stolfi shared. It may not be perfect yet, there may be certain contextual triggers i haven't identified.

[attachment=17924]

(26-09-2026, 05:20 PM)RadioFM Wrote: You are not allowed to view links. Register or Login to view.The fact of the matter is there's no apparent grammar in Voynichese, ... there's kind of enough evidence that points to gibberish as well.

There is a language out there that has none of the inflections seen in "European" languages: not for gender, number, case, mood, tense, for augmentatives and diminutives, for superlatives and comparatives, etc.  It has no articles or demonstrative pronouns, and rarely uses prepositions and conjunctions, including "and".  It does not even have a copula (like the verb "to be" in English).  And, unlike Indo-European languages, it does not divide its lexicon into disjoint classes of nouns, adjectives, verbs, and adverbs. And it has no suffixes or prefixes to make derived words (like "writer" from "write"), or to change their syntactic class (like "slowly" from "slow").

The Jesuits who set out to write tutorials for that language in the 1600s even claimed that "it had no grammar."  Which is not true -- but the grammar is subtle, not visible to the sort of statistics that can reveal it in "European" languages.

All the best, --stolfi

I'm seeing quite a lot of suggestive evidence that Herbal A's -ol and -or material is preserved in Herbal B through some type of splitting, compounding, or merging.

For example, the words containing *aiin interact with chol and chor, and -or/-ol, in unexpected ways.

• In herbal A, all cho forms are more likely to appear before an aiin form and less likely to appear after it. Chol is the most likely word to proceed *aiin.
  

• In herbal B, or specifically occurs at 2.4x the base rate in front of *aiin. All other possible variants, ol, or, ar, and even 4 of the 6 chors are more likely to appear AFTER an *aiin word.
  

This seems to suggest that some material is actually being reorganized across words, not merely transformed. For example, this material could be getting reorganized in B in the following ways: 

• chol daiin → olaiin
  
• chor daiin → or aiin 
  
• chor daiin → aiin chor
  
• chol daiin → chdy olaiin
  
• chor daiin → chedaiin or
  
• okol daiin → okaiin dol
  

My working theory to explain the missing -ol and -or mass is that not every word can be 'translated' into a different section with a 1:1 transformation. For at least some contexts, the content from multiple words in the source is reorganized across multiple words in the destination.

I've already been able to improve my existing rule-set by extending it to consider preceding and following words. 

• With respect to determining whether a word should begin with da- (like daiin) or a- (like aiin, air, etc), the preceding ending strongly conditions its form. r+s endings on the previous word make the a- 22x more likely, l makes it 0.5x as likely, y makes it 0.15x as likely. 
  
• this means that technically the following word determines whether -al or -edy appears. an a initial (like aiin) makes the -al ending on the prior word more likely, d- (like daiin) makes the -edy form more likely.
  

There are also indicators that words beyond the immediate adjacent two influence form realization, and before anyone yells 'self-citation', these gains cannot all be explained by an exact or similar copy of a word on the same line. 

In Herbal B, edy is more likely than -al to appear on lines whose other, nonadjacent words have fewer o-initial and l/r-final forms, and more y-final forms. I see predictive gains around .076 bits/token, which persists across several controls. This does not seem to affect all spelling choices, for instance bare gallows. o-, and qo- fronts only see gains of .006 bits/token. These results are suggestive but the actual mechanism is still unresolved.

I'm also seeing indicators that other words possessing an initial q can influence whether another word on the same line takes an -edy form. The actual position within the line of this other q- word matters. Words containing q- to the left of the target word, seems to make -edy forms more likely to occur, and this survives withholding the target word stem, the physical sheet, and removing repeated stem family words.

A concrete example of this can be seen in the Herbal B line:

• qokar okar [okedy] dar ykchey kaiin okas chedy okar a ralos
  

Moving the sole q from the first qokar to the later identical okar preserves every core, both immediate neighbors and total q- words, yet the held-out model’s edy probability falls .805 → .544.

These are mostly exploratory and suggestive findings. Held out models can boil down information in complex ways, and this definitely needs more in depth analysis and testing to validate and pin down what is happening mechanically. 

I will unfortunately not be able to work on this much over the coming week, but very excited about some of these findings and their potential to explain more correspondence between A/B.

You are not allowed to view links. Register or Login to view., Koen found that MATTR 500 is another feature that places Pharma (Small Plants) higher than Herbal A and lower than Herbal B. As he discusses, the most striking result is the behavior of Quire 13.

[attachment=17938]

In case this hasn’t been mentioned yet, this is You are not allowed to view links. Register or Login to view. were the mapping between Currier languages was discussed. Here the analysis is based on the whole A/B Corpora rather than Herbal A vs B only, so they are not immediately comparable.

Even older (2013) research by Julian Bunn You are not allowed to view links. Register or Login to view..
[attachment=17939]

I computed some basic stats for Herbal A Early and Herbal B, based on the ZL transliteration, ignoring uncertain spaces.
Benches and benched gallows were treated as single characters.
The colored column shows if the item is more frequent in A (red) or B (green).
As always, it’s possible I made errors along the way.


Characters:
[attachment=18006]
This shows that the five most frequent glyphs vary considerably.

• 'y' is the most stable of the five.
  
• 'o' and 'ch' are considerably more frequent in HA.
  
• 'a' and 'd' are considerably more frequent in HB. For 'd', the thing can easily be understood on the basis of the frequent -edy suffix in B.
  
• B:a increases but B:n decreases, so the additional 'a's do not occur as -ain -aiin.
  

Given that the three "circles" o,a,y are often regarded as related, one can imagine that there is some kind of re-distribution inside this class of glyphs (A:o turning into B:a,y).

It's interesting that the frequency of 'q' doesn't change: the class of q-words could be a relatively approachable subset for an A/B mapping.

Bigrams:
[attachment=18005]
The increase of ‘dy’ in B is again fully expected.
I highlighted a few bigrams whose changes I speculate that might be connected:

• The frequent ol/or bigrams in A might “become” al/ar bigrams in B (see char stats above).
  
• sho/cho in A might partially “become” she/che in B. But cho drops by 5% while che only increases by 1.5% and a lot of ‘ch’ disappear in B (see char stats), so there could be cases where ‘ch’ is dropped or changed into something else.
  

Words:
[attachment=18007]
I find word results hard to read: word stats appear to be entirely different in the two text samples. Honestly, this table makes me think that the task of an HA/HB mapping is terribly hard.

Something I want to point out is that both ‘chor’ and ‘shor’ decrease from A to B, while both ‘or’ and ‘ar’ increase, so it could be that the initial bench is sometimes dropped and initial o- sometimes turns into a-.
The two most frequent words in A, daiin and chol, have dramatic drops. A classical and plausible idea is that aiin in B partly corresponds to daiin in A, but the difference for daiin is -3% of all words, while the increase in aiin is only +1%, so aiin can only explain a minor part of what happens with daiin.
The idea that A:qokchy might become B:qokedy (put forward in post #1) appears to be a plausible one and explains some of the loss of A:ch.

A note on the context rules mentioned by npcompl33t:[url=https://www.voynich.ninja/user-3795.html][/url]
The a-/da- effect is the word break combination described by Currier and Smith & Ponzi. My research suggests this has to do with continous writing, strong within a line, 1/6 across a drawing that interrupts the line and disappear across lines. 

The -edy vs -al effect looks like a paragraph ending mode. My research suggest -r/-l is set per paragraph and changes between them more than can be explained by drift. Holding the paragraph and the exact position in the line fixed should show wether anything is left.

MarcoP: 
on q being stable: The line initial glyphs follow a cycle (see You are not allowed to view links. Register or Login to view.) and q is the stable glyph.

(26-09-2026, 06:18 PM)MarcoP Wrote: You are not allowed to view links. Register or Login to view.A classical and plausible idea is that aiin in B partly corresponds to daiin in A, but the difference for daiin is -3% of all words, while the increase in aiin is only +1%, so aiin can only explain a minor part of what happens with daiin.

My theory is that daiin and aiin are the same word in both sections, the d is just added in certain contexts, line initial, or based on the preceding word.

(26-09-2026, 06:18 PM)MarcoP Wrote: You are not allowed to view links. Register or Login to view.Something I want to point out is that both ‘chor’ and ‘shor’ decrease from A to B, while both ‘or’ and ‘ar’ increase, so it could be that the initial bench is sometimes dropped and initial o- sometimes turns into a-.

I agree, and i think another important insight is that or/ol as prefixes of words ALSO increases, so it could be that the initial bench is dropped, and the word is appended directly to the front of other words. I'm working on turning these into additional rules, but dropping benches is destructive, and there are a large number of possibilities, so it is taking time to try to go through them all.

(26-09-2026, 06:18 PM)MarcoP Wrote: You are not allowed to view links. Register or Login to view.Characters:

This shows that the five most frequent glyphs vary considerably.

• 'y' is the most stable of the five.
  
• 'o' and 'ch' are considerably more frequent in HA.
  
• 'a' and 'd' are considerably more frequent in HB. For 'd', the thing can easily be understood on the basis of the frequent -edy suffix in B.
  
• B:a increases but B:n decreases, so the additional 'a's do not occur as -ain -aiin.
  

Given that the three "circles" o,a,y are often regarded as related, one can imagine that there is some kind of re-distribution inside this class of glyphs (A:o turning into B:a,y).

It's interesting that the frequency of 'q' doesn't change: the class of q-words could be a relatively approachable subset for an A/B mapping.

This matches with my own analysis. I agree that some o's seem to be turning into a's, the extreme e discrepancy also means that somehow E's need to be created.

However, in my opinion, the most interesting statistic in these tables is the inverted T/K. After doing a bit of analysis on this, I can't help but feel there is a massive voynich clue hiding some where in this inversion.

Most word families containing k/t, exhibit inverted frequencies between early Herbal A and late herbal A/Pharma. For example, the relative frequency of something like otol/okol are swapped between early  HA and late HA. Early HA virtually universally prefers 't', while late HA/pharma almost universally prefer 'k'.

Some of this is hidden by other word transformations: after pooling words affected by the rules I mentioned (for example, okol->okeol->okedy, and okchy->okedy, pooling both okchy and okol), applying a dumb blanket swap K->T significantly repairs disjoint frequencies between word families in early HA vs late-HA, and this gets better when additional contextual constraints are applied. 

In particular, families that exist in early Herbal A, that contain a gallows followed by an e, seem to already be frequency corrected with their counterparts in other sections. This seems similar to a theory Stolfi mentions in his 2012 paper, that a single e following a character changes it (if you read this Stolfi I'd love to know if this is still your belief). 

There is more: consider 'Davis expansion' -- LFD's theory, discussed in her 2022 keynote, that P = ke and F = te,. Her argument was convincing to me, but the tables MarcoP shared show a massive issue. Her argument that P = ke is based on frequencies of k/t, and she notes this is a counterintuitive mapping, because P and T both have two loops, while F and K both have 1. 

This completely falls apart when analyzed within the context of early Herbal A. 

• P occurs at roughly the same rate between all sections. Yet ke and te do not, they are far less common in early Herbal A. It
  
• The sequence pch is also extremely common, yet tech/kech is basically nonexistent. However tch / kch is common in early Herbal A compared to other sections, nearly 5x.
  

IF:

•  P = 'early herbal A t', and F = 'early herbal a K' . 
  
• and you assume K and T were swapped between early Herbal A and later sections
  
• and P/F still retain the early Herbal A mappings
  

These problems mostly vanish, (although ke and te still do occur in early Herbal A so there may still be some interaction there). This also fixes the counterintuitive mapping of P->K and F->T based on numbers of loops. 

There are several possible theories to explain this, but all would require pretty significant reinterpretation about voynich, language, and early Herbal A vs later Herbal A.

Immediately I'm reminded of something like English knife vs knives. English in general underwent a sound change from 'v' to 'f', but certain forms still retain the archaic sounds.

P, F, K , and T could be recording some similar shift in voynich, some kind of a swap/inversion of K/T gallows, possibly related to the introduction of more 'e' words, where P and F preserve a more archaic version of voynich grammar. 


I have also put my tools on You are not allowed to view links. Register or Login to view., as well as a You are not allowed to view links. Register or Login to view.. Note this is something i developed for myself, so not everything may be intuititve.