As I said, I wanted to take a closer look at the VMS cores. I ran into a few problems—no surprise there. But then I had a major breakthrough - (major? 
)
Actually, this was already hinted at in my cipher; the “y” was treated differently depending on its position. But so was the “o.” The problem was that I hadn’t fully grasped the extent of it myself.
Some, if not all, of the VMS glyphs are used differently depending on their position.
And that’s exactly what creates some of the strange structures of the VMS! It’s highly interesting.
In short: VMS is a position-dependent absorption cipher.
I’m still working on a few things, but here are the initial results:
Because of the different positions, it became clear to me that the vowels and “aiin” must be used in a position-dependent manner!
In this respect, the problem is that you can’t achieve consistency if you compare all vowels from the frequency data with the MHD corpus. So you have to
focus on the medial part and leave out the initial and final vowels.
For a VMS token with N glyphs, “middle” refers to glyphs 2 through N−1. Tokens with fewer than 3 glyphs have no middle position and were excluded!
The digraph “ee” (two adjacent “e” glyphs) is also treated as a single unit, since the hypothesis under examination is: “ee = MHD u”.
If you split it into two separate “e”s, you increase the number of medial “e”s by about 9,600 and get a completely different picture. This is the single largest cause of discrepancies between counts by different people, which is why I am pointing it out right at the start.
Everything else—including ai, dy, ey—is counted as two separate characters. The aiin sequence is handled separately (see below).
Concrete example: “chedy” is tokenized as [ch]-[e]-[d]-[y], so its middle glyphs are [e, d]. “okeedy” is tokenized as [o]-[k]-[ee]-[d]-[y], middle glyphs = [k, ee, d].
The “aiin” Problem
There is a well-known complication
. The sequence "aiin" (and its variants "ain," "air," "aiir") occurs in about 15% of all VMS lemmas, almost exclusively in final position (85–91% of occurrences). It behaves like a positional unit (probably an ending like -en/-em) and not like a string of independent vowels. But the "a" and the "i" within it are located in the medial position and are counted, which significantly inflates both the medial a and the medial i in VMS.
Raw medial counts including aiin: VMS a = 13.48%, VMS i = 12.75%. This is far too high for any plausible vowel pairing. As soon as one recognizes "aiin" as a functional unit and removes its components from the medial count (5,902 a, 9,670 i, 247 n/r), the picture changes dramatically.
The numbers
VMS source: EVA transcription v3b, 38,227 tokens.
MHD reference: Ortloff von Baierland, Breslauer Arzneibuch, Admonter Bartholomäus, recipe collection Cod.germ.1 – total of 170,892 words. (Normalized)
Medial glyphs before removal of aiin: 93,606.
Medial glyphs after removal of aiin: 77,787.
The first three pairings land within half a percentage point. For o = e and a = a, the delta is 0.06% each — on a base of 77,787 medial glyphs vs 374,665 medial letters. But to be honest, this is a very specific analysis. If you make minor adjustments or take a closer look at the whole thing, you get different results—results that are still good, but don’t yield such precise matches. I'd say that these perfect matches are partly due to chance. But it’s also more about the general direction...
What this means
Four medial vowel pairings look solid:
EVA o in word-internal position behaves like MHD e — the most frequent vowel in both systems.
EVA a stays a.
EVA ee (the digraph) behaves like MHD u.
EVA e = MHD i is plausible (13.53% vs 11.15%) but not as tight.
EVA i collapses to 2.91% after aiin removal, which means almost all medial i in the VMS belongs to aiin sequences rather than appearing as an independent vowel.
Important!
This is a very fascinating correlation in the frequency of vowels in the medial position.But it is still a long way from being deciphered.
However, viewed once again from a different perspective, it demonstrates that the VMS possesses an internal vowel structure that corresponds, at the medial level, to medical texts from the Middle High German period, with specific and verifiable letter assignments.
(Note: The data was, of course, calculated by Claude but intensiv cross-checked by ChatGPT, with only minimal deviations, which are likely due to the cleaning of the Middle High German texts.)
)Actually, this was already hinted at in my cipher; the “y” was treated differently depending on its position. But so was the “o.” The problem was that I hadn’t fully grasped the extent of it myself.
Some, if not all, of the VMS glyphs are used differently depending on their position.
And that’s exactly what creates some of the strange structures of the VMS! It’s highly interesting.
In short: VMS is a position-dependent absorption cipher.
I’m still working on a few things, but here are the initial results:
Because of the different positions, it became clear to me that the vowels and “aiin” must be used in a position-dependent manner!
In this respect, the problem is that you can’t achieve consistency if you compare all vowels from the frequency data with the MHD corpus. So you have to
focus on the medial part and leave out the initial and final vowels. For a VMS token with N glyphs, “middle” refers to glyphs 2 through N−1. Tokens with fewer than 3 glyphs have no middle position and were excluded!
The digraph “ee” (two adjacent “e” glyphs) is also treated as a single unit, since the hypothesis under examination is: “ee = MHD u”.
If you split it into two separate “e”s, you increase the number of medial “e”s by about 9,600 and get a completely different picture. This is the single largest cause of discrepancies between counts by different people, which is why I am pointing it out right at the start.
Everything else—including ai, dy, ey—is counted as two separate characters. The aiin sequence is handled separately (see below).
Concrete example: “chedy” is tokenized as [ch]-[e]-[d]-[y], so its middle glyphs are [e, d]. “okeedy” is tokenized as [o]-[k]-[ee]-[d]-[y], middle glyphs = [k, ee, d].
The “aiin” Problem
There is a well-known complication
. The sequence "aiin" (and its variants "ain," "air," "aiir") occurs in about 15% of all VMS lemmas, almost exclusively in final position (85–91% of occurrences). It behaves like a positional unit (probably an ending like -en/-em) and not like a string of independent vowels. But the "a" and the "i" within it are located in the medial position and are counted, which significantly inflates both the medial a and the medial i in VMS.Raw medial counts including aiin: VMS a = 13.48%, VMS i = 12.75%. This is far too high for any plausible vowel pairing. As soon as one recognizes "aiin" as a functional unit and removes its components from the medial count (5,902 a, 9,670 i, 247 n/r), the picture changes dramatically.
The numbers
VMS source: EVA transcription v3b, 38,227 tokens.
MHD reference: Ortloff von Baierland, Breslauer Arzneibuch, Admonter Bartholomäus, recipe collection Cod.germ.1 – total of 170,892 words. (Normalized)
Medial glyphs before removal of aiin: 93,606.
Medial glyphs after removal of aiin: 77,787.
The first three pairings land within half a percentage point. For o = e and a = a, the delta is 0.06%
each — on a base of 77,787 medial glyphs vs 374,665 medial letters. But to be honest, this is a very specific analysis. If you make minor adjustments or take a closer look at the whole thing, you get different results—results that are still good, but don’t yield such precise matches. I'd say that these perfect matches are partly due to chance. But it’s also more about the general direction...What this means
Four medial vowel pairings look solid:
EVA o in word-internal position behaves like MHD e — the most frequent vowel in both systems.
EVA a stays a.
EVA ee (the digraph) behaves like MHD u.
EVA e = MHD i is plausible (13.53% vs 11.15%) but not as tight.
EVA i collapses to 2.91% after aiin removal, which means almost all medial i in the VMS belongs to aiin sequences rather than appearing as an independent vowel.
Important!
This is a very fascinating correlation in the frequency of vowels in the medial position.But it is still a long way from being deciphered.
However, viewed once again from a different perspective, it demonstrates that the VMS possesses an internal vowel structure that corresponds, at the medial level, to medical texts from the Middle High German period, with specific and verifiable letter assignments.
(Note: The data was, of course, calculated by Claude but intensiv cross-checked by ChatGPT, with only minimal deviations, which are likely due to the cleaning of the Middle High German texts.)
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)