Voynichese is a numeric cipher?

[Jorge_Stolfi]

Purely theoretically, it’s possible to use it. The question is how one can even create a book cipher for a book of that size, and make it work flawlessly. So I don’t believe in it either.

The most practical and effective way I can think of for creating a codebook cipher is to write each pair (plainword,code) on a card, and keep the cards standing up in a box, sorted by the plainword, with index tabs sticking out for each initial letter or combination, like "BA-BL", "BM-BR", "X-Z", etc.  To encode a word, the author looks it up in that pile of cards; if it is missing, he assigns it the next unassigned number, writes the pair in a new card, and inserts it into the pile, in the proper order.

To decode the message, the reader re-sorts the cards in order of code, then proceeds as in the encoding (except that if the card is not found, instead of creating it the reader should recite out loud some unfounded allegations about the writer's mother's profession. Unless he happens to be the same as the reader, of course.). 

To share the code with someone else, one would have to pay a famished student to copy all the cards, and send the copy by trusted camel to the other guy.  But updates may have to be sent by either party if additional messages require new words.

One can optimize that method by writing also the 50 (say) most common pairs on a "quick reference" sheet of paper. Per Zipf's law, that cheat sheet should save a significant fraction of the time by bypassing the search the card box.

One can start the encoding with an empty card box.  Then the most common words will tend to get lower (hence shorter) codes -- which is another important statistical feature of natural languages, independent of Zipf's law.

But even with this method a codebook cipher would be quite cumbersome to write and read.  I don't give it much probability.

All the best, --stolfi