+- The Voynich Ninja (https://www.voynich.ninja)
+-- Forum: Voynich Research (https://www.voynich.ninja/forum-27.html)
+--- Forum: Theories & Solutions (https://www.voynich.ninja/forum-58.html)
+--- Thread: Why and how the text could be Bavarian (/thread-5312.html)
RE: Why and how the text could be Bavarian - Grove - 20-09-2026
Am I wrong or does this suggest that no word beginning with P ends in y? And that all words beginning with ckh or cph end in y?
RE: Why and how the text could be Bavarian - JoJo_Jost - 20-09-2026
No, as stated, the table shows only the “entries” into the tokens, not the entire token. The base glyph is not the last letter of the token.
(In VBM, the first glyphs (in Reva the last) also form the vowel bridges, and in those cases it is often actually the last glyph in the core (without a vowel bridge), but of course not always)
RE: Why and how the text could be Bavarian - JoJo_Jost - 20-09-2026
But isn't this structure of the VMS in Reva incredibly uniform???
RE: Why and how the text could be Bavarian - Jorge_Stolfi - 21-09-2026
(19-09-2026, 05:14 PM)JoJo_Jost wrote: You are not allowed to view links. Register or Login to view.When writing the “sh” in “shedy,” the writer would already have to know that the next word begins with 51 % = “qo.” That is what is peculiar: a glyph in a word “knows” something about its successor. That is strange—a kind of long-range effect.
JoJo, forgive me for repeating the advice I gave on another thread: character-level statistical analyses (character and digraph frequencies, entropies, correlations, etc.) are very likely to be a waste of time. That includes these so-called "attractions" and "remote effects" between certain glyphs and digraphs, like between Sh and qo.
There are a few situations in a few languages where correlations are indeed associated to specific letters or digraphs, independently of the words that contain them. In Turkish there are "vowel harmony" rules that generally require all vowels of a word to be "front" or all of them to be "back". So that if the root uses "front" vowels, all suffixes attached to it must also use "front" vowels: "ev"="house", "evler"= "houses", but "araba"="car", "arabalar"="cars". Thus, in that case you could indeed say something like "front vowels attract front vowels".
But such character-based rules seem to be very rare. In English there is the rule that the article "a" changes to "an" when the following word begins with a vowel. But that rule is character-based only on the right side, because it only applies to the article "a", not to any word that ends in "a". And in fact it it is not quite character based on the right either, because the "a" may remain just "a" depending on how the initial letter of the next word is pronounced: "an umbrella", "an hour", "an ostrich", but "a university", "a house", "a one-way street".
Statistics of characters and digraphs are mainly determined by the statistics of the few most common words that contain them, and the latter are all mixed together. Thus, for example, consider the claim
- in English the digraph "th" at the beginning of a word is "attracted" to the letter "i" at the beginning of the next word
I haven't checked, but even if this claim is true it is pretty useless and maybe even misleading. It will not lead to any useful insight into the English grammar, and could lead one to infer rules that are actually not valid.
If that claim is true, it is probably due to the common pairs "this is", "that is", "there is", etc. Meaning that it is not an attraction between the characters "th" and "i", but between certain common words that happen to use those characters. Thus a more useful statement would be
- in English the words "this","that","there","he","she","it" are attracted to the words "is","isn't", and there is a similar attraction between "then","thus","therefore" and "it","its". But "the","their", on the contrary, are repelled by those four "i"-words.
Thus, in your case, what seems to be happening is that there are certain words that begin with sh that tend to be followed by certain words that begin qo.
However, judging from some quick checks I ran, there probably are some non-Sh-words that are attracted just as strongly to qo-words, and some Sh-words that are actually repelled by some qo-words; and the effects may be opposite for some other qo-words.
Identifying those attractions and repulsions between word pairs has a much better chance of providing useful insights than hunting for "remote effects" between characters.
All the best, --stolfi
RE: Why and how the text could be Bavarian - JoJo_Jost - 21-09-2026
@ Stolfi, I already replied to you in your chat, but sure, I’m happy to do it again here:
Your argument is: Character statistics are a byproduct of the most common words. “th” pulls “i” because “this is” and “that is” are common. That’s not wrong, but it doesn’t account for the effect of “sh” / “ch” (and others).
To demonstrate this, I’ve separated the occurrences by word and by the frequency of each word.
The bottom row shows the effect for words that occur more than 100 times. There, you might be right—those could be collocations: “shedy” before “qokeedy” and so on.
BUT: The top row consists of words that appear exactly once in the entire manuscript: 543 with “sh,” 1,263 with “ch,” and 730 with “k.” A word that appears only once cannot have a collocation. There’s no pattern like “this is” for a word that exists only once!
And lo and behold: Nevertheless, 20.8% of the one-time “sh” words are followed by a “qo” word, 10.8% of the one-time “ch” words are followed by a “qo” word, and 11.6% of the one-time “k” words are followed by a “qo” word.
The crazy thing is: The ratio of “sh” to “ch” remains nearly the same—between 1.7 and 2.0—regardless of the number of words, that is, in every row, from the one-time words all the way up to the most frequent ones. That is the question your model must answer: Which word pairs generate this, and the top row?
Next: Your quick checks are correct: “sho,” “shol,” and “shor” are below the baseline rate.
But look at which “sh” words are at the bottom. The five at the bottom are “sho” 9.6%, “shol” 12.1%, “shor” 12.8%, “shar” 12.9%, and ‘shy’ 14.8%, and none of them have an “e.”
Every “sh” word with an “e” is above the base rate: sheo 22.9, sheor 25.5, sheey 36.8, sheol 37.3, sheody 38.3, shey 42.3, shedy 49.0, sheedy 53.0. And your non-sh words that take qo all have an “e” before the end: olkeedy 51.2, cheedy 47.3, otchedy 45.5, qokeedy 41.9. So the rule is not only “sh takes qo.” It is: a sequence of “e” before the end of a word causes “qo,” and “sh” reinforces this effect compared to “ch,” “k,” and “t” (-shedy 49.5%, -chedy 39.3%, -kedy 33.2%, -tedy 31.0%).
And THAT is a rule about a word’s structure. It applies to words that occur only once just as much as to the most common ones. A word-pair model has no place here.
Second, since you say we should look at word pairs: here are some.
shor and chor: same ending, same length, different initial sound (excluding “e”). In 29% of cases, “shor” is followed by an “sh” word; in 8.4% of cases, “chor” is followed by an “sh” word. In 28,4% of cases, “chor” is followed by a “ch” word; in 9,2% of cases, “shor” is followed by a “ch” word. Here, too, we see this effect.
I calculated this relative to the line in which the word appears—that is, relative to the proportion of “sh” or “ch” words among the other words in the same line: “shor” → “sh” is 2.2 times the line expectation, “chor” → “sh” is 1.1 times; “chor” → “ch” is 1.5 times the line expectation, “shor” → “ch” is 0.7 times the line expectation. So this is not a line effect.
The same pattern holds for keey/sheey with l (20.5% vs. 3.0%), kain/dain with ch (37,9% vs. 21.3%), taiin/kaiin with o (47.1% vs. 30.9%), and keo/sheo with r (20.4 vs. 0.0%).
You might say these are just different words with different neighbors. But here, too, we see the sh/ch effect—how can all of this be explained using word pairs??? Moreover, the other examples also show that these effects extend even beyond sh and ch and qo...
The point is, to explain this using word pairs, you need a separate collocation history for each pair, and all of these histories would have to happen to coincide in such a way that the initial sound of the preceding word is the same. When you look at the big picture, that’s really quite far-fetched.
So the question really isn’t whether character statistics are a waste of time. The question is which word-pair model is supposed to generate such a structure—and as hapax, no less!
---
I'm not really concerned with sh and qo here. What matters to me is which rules are tied to a word's form and which to its frequency; the former say something about the cipher, the latter about the text.
You’re right: a single finding of this kind proves nothing. But multiple findings pointing in the same direction are all we have, and that’s why I weigh them against my own assumption just as I would against any other.
And that’s how one decipher texts, too...
RE: Why and how the text could be Bavarian - Aga Tentakulus - 21-09-2026
Have you ever compared ‘ch’ with ‘sh’?
You’re comparing gold nuggets with pebbles.
‘Sh’ occurs just as frequently as ‘cr/chr’ in the Bavarian text.
Take a close look at what’s happening – there are even two instances of ‘r’
RE: Why and how the text could be Bavarian - JoJo_Jost - 21-09-2026
@ Aga ???
And what does that have to do with what Stofli and I were talking about?
Und was hast das mit dem zu tun, worüber Stofli und ich geredet haben?
RE: Why and how the text could be Bavarian - Aga Tentakulus - 21-09-2026
(21-09-2026, 09:54 PM)JoJo_Jost wrote: You are not allowed to view links. Register or Login to view.And what does that have to do with what Stofli and I were talking about?
Then have a look at the pictures. Can you spot the difference? “ch/sh”?
Stolfi’s quite right – that’s not how it works.
RE: Why and how the text could be Bavarian - Jorge_Stolfi - 21-09-2026
(21-09-2026, 07:54 PM)JoJo_Jost wrote: You are not allowed to view links. Register or Login to view.@ Stolfi, I already replied to you in your chat
Sorry. I saw your post there, but I though you had not really replied to my point. Maybe I did not read it with due care.
Quote:Quote:Me: Character statistics are a byproduct of the most common words.... The top row consists of words that appear exactly once in the entire manuscript: 543 with “sh,” 1,263 with “ch,” and 730 with “k.” A word that appears only once cannot have a collocation. There’s no pattern like “this is” for a word that exists only once! ... A word-pair model has no place here.
First, many hapax words in the VMS are probably malformed, truncated, or joined versions of words that appear multiple times. So, for example, if "shedy qokedy" is a common word pair, you should see also many pairs like "shoedy qokedy", "shddy qokedy", "shshedy qokedy", and so on -- where the first words of those pairs are hapaxes.
You could try to investigate this possibility by separating "good" words that fit one of the various word models out there from "bad" words that do not fit. IIRC, when I tested my own word model, I found that something like 5-7% of the tokens failed it in some way.
Second, valid hapax words too definitely can have attraction or repulsion for qo-words and contribute to the apparent sh-for-qo attraction.
For example, in English, depending on the topic, it may well happen that the majority of the hapax words are nouns ("chamomille", "shingles", "baldness", etc.), while the verb words are few and mostly occur many times ("cures", "drink", "boil", etc). I am guessing that verbs will rarely be followed by "is", while that will be often the case for nouns. If it happens that the most common verbs do not begin with with (say) "p", the result will be an apparent attraction of "p"-words to "i"-words, including among the hapaxes -- which is actually due to repulsion of a few common non-hapax non-"p" words by "i"-words.
Quote:The crazy thing is: The ratio of “sh” to “ch” remains nearly the same—between 1.7 and 2.0—regardless of the number of words, that is, in every row, from the one-time words all the way up to the most frequent ones. That is the question your model must answer: Which word pairs generate this, and the top row?
Well, I think that it is you who should answer that question.
But seriously: the first thing I would do is to plot the frequency of each shXXX word against the frequency of the corresponding chXXX word, for each EVA string XXX, and see whether those dots -- at least those with higher frequencies -- fall close to a straight line. (A log-log plot would be better, I think)
If they mostly do, it would be evidence (not proof, of course) that the ch-words are equivalent variants of the corresponding sh-words. That is, there is a certain probability that a sh-word will lose the plume and become a ch-word, or vice-versa. Various processes could have this effect: confusion of similar sounds, sandhi, transcription errors, ink fading, poor handwriting on the draft...
(A unique English text that should be used more often in comparisons is the diary of the Lewis and Clark Expedition. Unlike most books and medieval manuscripts, it was written by two Army officers who had gone through the Academy but obviously were not much into literary virtues. So the text has tons of spelling errors and inconsistencies, not only on words of Native origin (I think I saw four different spellings of "buffalo", some in the same parag) but even in very common words like "to-day" and "behaveing" and "prisnair". Some of the errors seem to be stochastic but systematic -- that is, endings like "-ance" and "-ence", "-city" and "-sity" seem to have been often swapped independently of the word.)
If the ch-words are mostly variant spellings of the sh-words, then the two variants should show similar attraction or repulsion for each qo-word (or any other word) that follows it.
On the other hand, if the points on that ch/sh plot do not lie on a straight line, it would be evidence that a chXXX word is, at least some of the time, not a variant of the corresponding shXXX word. Then I would expect that the two will have different attraction or repulsion numbers. If they behave the same way in that regard, then I would have to scratch my head harder...
Quote:shor and chor: same ending, same length, different initial sound (excluding “e”). In 29% of cases, “shor” is followed by an “sh” word; in 8.4% of cases, “chor” is followed by an “sh” word. In 28,4% of cases, “chor” is followed by a “ch” word; in 9,2% of cases, “shor” is followed by a “ch” word.
These numbers seem to be consistent with the second possibility above.
Quote:to explain this using word pairs, you need a separate collocation history for each pair, and all of these histories would have to happen to coincide in such a way that the initial sound of the preceding word is the same.
The word pairs do not have to all agree. The "attraction" that you see is the sum of individual word attractions, so it is dominated by a few high-frequency words.
Quote:I'm not really concerned with sh and qo here. What matters to me is which rules are tied to a word's form and which to its frequency; the former say something about the cipher, the latter about the text.
Well, again, I am still not convinced that those numbers are due to attraction between characters or digraphs, rather than words.
Quote:You’re right: a single finding of this kind proves nothing.
It is not that, but rather that this kind of result is unlikely to lead to any useful insight into the Voynichese "grammar".
As I wrote before, trying to decipher a text by computing character and digraph statistics is like trying to understand the ecology of a forest by counting how many animals have tails, how many are brown, and how many have antennas. Each count would mix antelopes and armadillos, bears and beetles, cats and caterpillars, ..., weasels and woodpeckers, ... Those counts may perhaps suggest that the Black Forest is somehow different from the Amazon, and that something changes between summer and winder; but they would never lead to any real understanding of the ecology.
All the best, --stolfi
RE: Why and how the text could be Bavarian - JoJo_Jost - 22-09-2026
@ Stolfi: The argument about the errors is NOT evidence against it; it’s a theoretical possibility—it could be, or it might not be.
The problem, however, is that this approach allows you to retroactively label any “disturbing” form as an error. In this way, almost any finding can be dismissed. But the sh/ch structure is too deeply rooted—as Aga’s image impressively demonstrates—to be dismissed as a mistake.
But I do realize that these sh/ch structures I’m pointing out here at least have the theoretical potential to call your Chinese theory into question. That’s why I wanted to discuss the phenomenon in your thread, and so i started it there.
That’s why I wanted to discuss the phenomenon in your thread and presented it there as well, and I think we should continue the discussion in your chat: because the point is that I wanted an explanation from you as to how this fits in with the Chinese theory. And to do that, of course, we also have to talk about the Chinese theory, and that doesn’t fit thematically here. The board rule is: one topic per thread!
For VBM, especially in the REVA analysis (Eva reverse), this phenomenon isn’t a problem at all—because it doesn’t occur at the token level here, but within words and when read backward, it sort of dissolves. So this discussion here is of rather secondary importance. But if we reach a conclusion—which, unfortunately, I doubt—I'll post it here in my thread.
So if you agree, I’ll take your answer—since it does address my posts there —and respond to it in your thread on the Chinese theory; I think that’s the most reasonable approach.
PS. But I'm out and about today, so I probably won't be able to reply until tomorrow at the earliest
