Could the Voynich Author Be a "Ramanujan-style" Anomaly?

[Jorge_Stolfi]

Oh, you’re using my favorite examples against me…  

Rather, I am just pointing out that those two examples are not good analogs for the VMS.

For example, nulls can somewhat distort the statistics of ciphertext. Perhaps there are some other unaccounted factors?

As I wrote in a previous post, nulls will replace each word type by several different word types.  That would ruin the Zipf plot (and other statistics expected from natural languages) unless the nulls are chosen with specific non-uniform probabilities. 

Said another way, if the encoding preserves the Zipf plot and other "natural language" statistics without complicated and specific tricks, it probably is such that each word type of the plaintext is mapped to a single word type of the ciphertext.  That pretty much excludes Vigenère-like and polyalphabetic ciphers, and encodings with random nulls.  

Codebook ciphers have that property, or any character-based encoding algorithm that is deterministic and is reset at each token.  

However, an encoding that is one-to-one on word types can be cracked if the plaintext is running prose in a known (or correctly guessed) language.  One can look for patterns of words that are created by the language's grammar -- like articles before nouns in German and Romance languages, verb-at-the-end for German, etc.  But all attempts to do this with Voynichese have failed...

All the best,--stolfi