Two letters per ciphertext word
I began to think. As Robert had obviously thought, the numbers of 23 and 21 groups, respectively, were too tantalizingly close to the number of letters in the latin alphabet to be rejected. (Especially if one considers that not all the renaissance alphabets included seperate characters for “j”, “y”, “x”, or “w” and thus had less than 26 letters.) What if each of the odd and even groups was one enciphered plaintext letter each?
That would make sense, the VM entropy is approximately 10 bits per ciphertext word. That’s enough to encode 2 character at 5 bits per character — not quite enough for ASCII, but enough for 32 different combinations, ie the latin alphabet plus a few extras.
But what are the two different alphabets with their “overlap”?
How to do the Voynich, Mark II:
Assume the encipherer wants to encipher the word “abdomen” again:
He breaks down the plaintext into two-letter groups:
ab do me n
Now, he converts it into a kind of “CamelCase”, by capitalizing the first letter of each word:
Ab Do Me N
And from here on, he continues as in Mark I.
And what’s the result?
Now, by adding this extra steps, the encipherer might think that he strengthened his code (but actually he weakened it — see below).
Furthermore, he did add structure to the VM words, since every ciphertext word begins with an enciphered capital letter, and some penstrokes only occur in capital letters. “abdomen” is a good example for this, since the first ciphertext word would incorporate “Ab”, and the very first enciphered stroke should be the upwards slash of the capital “A”, namely “/”. And this is one stroke which never occurs in minor letters. Depending on how “V”, “W” and “X” would be decomposed into strokes, the “A” might very well be the only occassion where the “/” ever appears in the whole alphabet — and thus, it would always be ciphertext word-initial!
In a similar manner, several more of the VM text features can be explained by assuming a “Mark II encipherment”. In a nutshell, in Mark II the word is an encipherment unit and thereby follows certain rules, which was not the case with Mark I.