02-04-2026, 02:47 PM
New Frequency Comparison: VMS vs. MHD Reference Corpus
I mentioned the frequency comparisons before. Here's the updated table - I've continued working on it.
On the left you see the raw comparison, without any adjustments. On the right, a transformed version that accounts for the cipher's key characteristic: absorption of function words (explained below).
The correspondence is incredible
I want to be very clear about what we're looking at here. These are not random features. These are the fundamental structures of grammar -- specifically German grammar:
Finding this kind of match in a randomly generated text that also possesses internal logic would be virtually impossible. You simply can't argue that away.
And this is the second important point: there is no other language without a Germanic foundation that could produce these frequencies. This structure tells us that the text underlying the VMS is German. The only alternative would be some kind of code deliberately designed to mimic German grammar -- which makes no sense.
------
Method: Forward Transformation
The standard approach is to transform the MHD reference texts "forward" (applying absorption rules to the MHD), then compare the resulting frequencies to what we observe in the VMS. This avoids circularity - I dont decode the VMS and then check if it looks like MHD.
The MHD reference corpus:
- Ortloff von Baierland (46,565 words) -- medical handbook
- Breslauer Arzneibuch (92,347 words) -- pharmacopoeia
- Admonter Bartholomaeus (29,244 words) -- herbal/medical
- Kochrezeptsammlung Cod. germ. 1 (2,725 words) -- recipes
Ca: 170,881 words
All MHD texts are Unicode-normalized before counting (long-s, old-d, old-r, umlauts, combining diacritics stripped).
Why a transformation?
The problem is that the VMS text, when viewed as an absorption cipher, naturally has fewer words than the MHD text because they merge with the individual words.
So I added that as a point and called it transformation.
The following absorption rules were examined:
1. Every ARTICLE (der/die/das/ein... – 104 forms) that directly precedes a content word is integrated into that word as a prefix, thus reducing the number of tokens.
2. Every PREPOSITION (in/von/mit/an... – 78 forms) that precedes an article merges with the article to form the following word, also reducing the number of tokens.
3. Every VERB PREFIX (ge-, ver-, be-, er-, zer-, ent-) is integrated into the verb stem.
4. Every "and" (und/vnd/unde... – 10 forms) is integrated as a prefix or suffix into its neighbor.
After the transformation, the number of tokens in the MHD decreases from 170,881 to 129,647, because function words have been integrated into content words.
Why articles = o + gallows consonants?
In my interpretation, gallows consonants (EVA k, t, p, f) mark the beginning of nouns. Articles precede nouns. Therefore, the prefix of the absorbed article "o-" should specifically appear before gallows consonants.
Otherwise, I would have had the problem of having to introduce a boundary for words beginning with "o," which seemed too variable.
Words like "ol," "or," and "odaiin" begin with the vowel o—they are NOT absorbed articles. Only o directly followed by a gallows consonant—this makes sense in many respects.
What -dy encompasses
The VMS glyph "dy" is a single tokenization unit that likely originated historically in the CH cipher from d + y. It functions as an absorption suffix and encompasses three different MHD sources:
1. -ch (+ inflection): the primary meaning. It includes -ch, -cht, -che, -chen, -chet, -chte, -chten. All inflected forms merge into a single VMS form. Example: mach/machen/machet/macht all become "chedy".
2. -sch (+ inflection): It includes -sch, -schen, -schet. Words like mensch, waschen, mischen, fleisch.
3. d + y (dental + final -e): It includes -de, -te, -se, -ze. This is a graphical homography - it looks identical to the dy glyph but represents the two separate sounds (dental + final -e).
Each component has a unique phonetic source in MHD:
- MHD endings -ch: 5.81% of MHD tokens
- MHD endings -sch: 0.86% of MHD tokens
- MHD dental + final -e: 5.33% of MHD tokens
- Combined: ~12.0% (raw), ~11.0% (after transformation)
VMS frequency -dy: 17.5%, which corresponds to a ratio of 1.59x to the transformed MHD. This is somewhat elevated, but still within a reasonable range, as it also depends on the writer. Furthermore, some words with -dy can encode additional, as yet unidentified, endings.
What -ey encompasses
The suffix -ey absorbs MHD -er (with inflectional forms: -ern, -ere, -ert, -ers).
Examples: mehr -> chey (ch=m, ey=er), sehr -> shey, her -> tey, er -> ey.
VMS: 6.0% vs. MHD-transformed: 6.9% – ratio 0.86. Good agreement.
Conjunction "und" (and) = EVA s
"und" appears in three positions in the VMS:
- Standalone "s": 309 tokens (0.81%)
- Word-initial s (first glyph after tokenization): 976 tokens (2.55%)
- Word-final s (last glyph after tokenization): 1,081 tokens (2.83%)
Total: 2,366 tokens (6.19%)
MHD frequency of "und/vnd" in the four reference texts: 6.14%.
Ratio: 1.01 = perfect match – although that's more likely a coincidence. You can see it in the heavily overrepresented "und" (and) in the recipes.
Examples of absorbed "and":
- s + aiin = "and a" (119x)
- s + ar = "and he" (77x)
- cheo + s = "...and" (37x)
Note that over 50 % of s-initial and final VMS words also appear in the corpus without the "s" — exactly what you'd expect if s is a detachable prefix meaning "und".
----
The real question now is this: does the German structure reflect an actual German source text, or a code built on German linguistic structure but without recoverable sense? I do not know that yet.
I mentioned the frequency comparisons before. Here's the updated table - I've continued working on it.
On the left you see the raw comparison, without any adjustments. On the right, a transformed version that accounts for the cipher's key characteristic: absorption of function words (explained below).
The correspondence is incredible
I want to be very clear about what we're looking at here. These are not random features. These are the fundamental structures of grammar -- specifically German grammar:
- Articles
- Prepositions
- Verb prefixes and suffixes
- Conjunctions
Finding this kind of match in a randomly generated text that also possesses internal logic would be virtually impossible. You simply can't argue that away.
And this is the second important point: there is no other language without a Germanic foundation that could produce these frequencies. This structure tells us that the text underlying the VMS is German. The only alternative would be some kind of code deliberately designed to mimic German grammar -- which makes no sense.
------
Method: Forward Transformation
The standard approach is to transform the MHD reference texts "forward" (applying absorption rules to the MHD), then compare the resulting frequencies to what we observe in the VMS. This avoids circularity - I dont decode the VMS and then check if it looks like MHD.
The MHD reference corpus:
- Ortloff von Baierland (46,565 words) -- medical handbook
- Breslauer Arzneibuch (92,347 words) -- pharmacopoeia
- Admonter Bartholomaeus (29,244 words) -- herbal/medical
- Kochrezeptsammlung Cod. germ. 1 (2,725 words) -- recipes
Ca: 170,881 words
All MHD texts are Unicode-normalized before counting (long-s, old-d, old-r, umlauts, combining diacritics stripped).
Why a transformation?
The problem is that the VMS text, when viewed as an absorption cipher, naturally has fewer words than the MHD text because they merge with the individual words.
So I added that as a point and called it transformation.
The following absorption rules were examined:
1. Every ARTICLE (der/die/das/ein... – 104 forms) that directly precedes a content word is integrated into that word as a prefix, thus reducing the number of tokens.
2. Every PREPOSITION (in/von/mit/an... – 78 forms) that precedes an article merges with the article to form the following word, also reducing the number of tokens.
3. Every VERB PREFIX (ge-, ver-, be-, er-, zer-, ent-) is integrated into the verb stem.
4. Every "and" (und/vnd/unde... – 10 forms) is integrated as a prefix or suffix into its neighbor.
After the transformation, the number of tokens in the MHD decreases from 170,881 to 129,647, because function words have been integrated into content words.
Why articles = o + gallows consonants?
In my interpretation, gallows consonants (EVA k, t, p, f) mark the beginning of nouns. Articles precede nouns. Therefore, the prefix of the absorbed article "o-" should specifically appear before gallows consonants.
Otherwise, I would have had the problem of having to introduce a boundary for words beginning with "o," which seemed too variable.
Words like "ol," "or," and "odaiin" begin with the vowel o—they are NOT absorbed articles. Only o directly followed by a gallows consonant—this makes sense in many respects.
What -dy encompasses
The VMS glyph "dy" is a single tokenization unit that likely originated historically in the CH cipher from d + y. It functions as an absorption suffix and encompasses three different MHD sources:
1. -ch (+ inflection): the primary meaning. It includes -ch, -cht, -che, -chen, -chet, -chte, -chten. All inflected forms merge into a single VMS form. Example: mach/machen/machet/macht all become "chedy".
2. -sch (+ inflection): It includes -sch, -schen, -schet. Words like mensch, waschen, mischen, fleisch.
3. d + y (dental + final -e): It includes -de, -te, -se, -ze. This is a graphical homography - it looks identical to the dy glyph but represents the two separate sounds (dental + final -e).
Each component has a unique phonetic source in MHD:
- MHD endings -ch: 5.81% of MHD tokens
- MHD endings -sch: 0.86% of MHD tokens
- MHD dental + final -e: 5.33% of MHD tokens
- Combined: ~12.0% (raw), ~11.0% (after transformation)
VMS frequency -dy: 17.5%, which corresponds to a ratio of 1.59x to the transformed MHD. This is somewhat elevated, but still within a reasonable range, as it also depends on the writer. Furthermore, some words with -dy can encode additional, as yet unidentified, endings.
What -ey encompasses
The suffix -ey absorbs MHD -er (with inflectional forms: -ern, -ere, -ert, -ers).
Examples: mehr -> chey (ch=m, ey=er), sehr -> shey, her -> tey, er -> ey.
VMS: 6.0% vs. MHD-transformed: 6.9% – ratio 0.86. Good agreement.
Conjunction "und" (and) = EVA s
"und" appears in three positions in the VMS:
- Standalone "s": 309 tokens (0.81%)
- Word-initial s (first glyph after tokenization): 976 tokens (2.55%)
- Word-final s (last glyph after tokenization): 1,081 tokens (2.83%)
Total: 2,366 tokens (6.19%)
MHD frequency of "und/vnd" in the four reference texts: 6.14%.
Ratio: 1.01 = perfect match – although that's more likely a coincidence. You can see it in the heavily overrepresented "und" (and) in the recipes.
Examples of absorbed "and":
- s + aiin = "and a" (119x)
- s + ar = "and he" (77x)
- cheo + s = "...and" (37x)
Note that over 50 % of s-initial and final VMS words also appear in the corpus without the "s" — exactly what you'd expect if s is a detachable prefix meaning "und".
----
The real question now is this: does the German structure reflect an actual German source text, or a code built on German linguistic structure but without recoverable sense? I do not know that yet.
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)