It cannot be a language - in reference to Koen's latest video

[zamolxe]

"It just can't be a language, under the assumptions that were made".

This is a very important point. My video was in the first place a reaction to the way most Voynich theories are formed: pick a language you like, find some words that match, immediately get into trouble so start inventing rules that eventually give you a ton of freedom to produce what you like. My point was very much that we can't just wildly start slapping Latin or Greek on it - first, the system must be discovered.

My conviction that the text must contain meaning has decreased over the years, but I still give it over 50% chance. Just not a simple substitution cipher.

On the other hand, I'm glad to see that you understood my point very well: even forced vowelless systems won't work, for a variety of reasons. I think most people overestimate how legible an unknown text in vowelless English would be, for example. And it still wouldn't match Voynichese in any way.

Most importantly, I believe the glyph set and the system are chosen with a certain elegance. This was premeditated and cannot have come about as a coincidental effect of doing something like omitting vowels from a language.

Just a comment from a non professional: if the VM is not a plain text or a simple substitution cipher (1:1), the analysis of Voynich letters statistics have a limited use and applicability. Even if some VM letters are in fact abbreviation signs will affect the statistical analysis.

About the alphabet. An 24 letter alphabet can be written using 13 signs, 12 letters and one "joker"/shifter. If the joker is attached to a letter, it will change/shift it to another letter (second meaning). So the alphabet will be represented both by single signs and bigrams. By the way, I suppose this will decrease the first order entropy, isn't it? 
Similar other systems cam be constructed increasing the number of "jokers" (shifters): 8+2, 6+3 etc. 
In extremis, just 3 signs can be used (I,V,X) to represent all the alphabet by writing each letter as a roman numeral. This will be easy to detect. To make the system more difficult to be detected and the text harder for deciphering attempts, multiple notations for each basic numeral can be used: for example "I" can be represented as "i", "c", "a" etc. Even the rules for constructing the numbers (subtraction, addition) can be changed.

In VM context, can for example "r", "or", and "ar" be 3 different letters? How to differentiate the letters represented by 2 signs from real language bigrams?

Without any proof apart of it's length, I tend to think that the VM was written to be easy readable for someone who knows the system/cipher without performing complicated operations to get the plain text.