The 'Chinese' Theory: For and Against

[Jorge_Stolfi]

the statistics of Voynichese are ... quite INcompatible with any European language, any plausible encryption scheme, and any plausible gibberish generation method

If Voynich is not consistent with known encryption methods, this does not mean that Voynichese is definitively Chinese. This could be an unknown encryption method, couldn’t it? Maybe it’s a special way to generate gibberish, isn’t it? you’re saying more “I don’t know of anything similar,” rather than “there’s nothing similar"

Yes, one can always construct an encryption method that produces a ciphertext that matches Voynichese under any given set of statistical tests.   And the output an order-3 word-based Markov generator would probably pass those tests too.

But my claim is stronger than just "I don't know".  Note that I wrote "plausible".  It is very unlikely that an encryption schema that the VMS Author could have chosen -- like, say, polyalphabetic substitution with nulls, or an Alberti or Vigenère cipher -- would happen to produce an output with the properties of a monosyllabic natural language, including Zipf's law, vocabulary size, word structure, word correlations, etc.  

The encryption method would have to be specifically designed to get that result.  For instance, if every token contains some null letters -- even if just one, in a fixed position -- each word type of the original will become any of N different word types in the cyphertext, for each of the N possible choices of nulls.  That multiplication of word types would be plainly visible in the Zipf plot -- unless the possible null choices are made with different, carefully tuned probabilities.  

And the same "plausibility" objection applies to gibberish generators. In the copy-and-mutate method, for instance,  the seed text must already be statistically similar to Voynichese (which already creates a chicken-and-egg problem) and the mutate procedure must be complex and fine-tuned to preserve the statistics.

All the best, --stolfi