Generated word tokens from chars ( pairs )

[bi3mw]

....
(I am sorry for the cryptic text, but at this point in your example, I am also a bit lost finding your direction of reasoning in the method)

This is probably because I have not sufficiently described the idea behind the method, sorry for that. The initial question was:

Is it possible to generate a text from word types of a plain text that shows similar characteristics as the VMS and from which these words can be decrypted ?

Basic conditions that must be met are:
The sum of the words in the generated text should not deviate too much from the source text. Also the ratio of word types to word tokens should not change significantly. The plaintext consists of 25014 words, the generated text of 23264. In the VMS , the ratio word types / word tokens" is 8114 to 37919. So the ratio is 1: 4,6732807493221. In the encrypted comparison text, the ratio is 5653 to 23002. Starting from the word types, one would expect a total text of 26418 words (at best). In my opinion, the deviation to the VMS is acceptable.

The generated text must be highly repetitive. In the encrypted text, 5391 word types face 17873 word tokens. All 17873 are generated from the 262 new word types. This condition is certainly fulfilled.

The average word length must be comparable. It is probably 5,6 in the VMS. The encrypted comparison text has a word length of 5,8.

Just to avoid misunderstanding, I do not claim that the manuscript was made this way.  But it seems to be possible.

About filling with nulls, I had used that in another try. But far too long words were generated.


Finally, just for fun, this is what the text looks like (not a single thought of the charset)