Need advice for testing of hypotheses related to the self-citation method

[Jorge_Stolfi]

The only interesting question for me is: could a model of how a human would plausibly do it with the self-citation method replicate most or all the known properties of Voynichese?

I would rather not spend much time studying the "self-citation method" (SCM), but maybe I can still say something useful.  I don't believe the 'hoax' conclusion.  The main reason I will explain in my 10-minute talk at the conference. The second main reason is that it impossible to prove that some string "contains no message".  Or even to provide statistical evidence that would make such a conclusion more likely than not.

One can only prove that a text contains no message if there is a short and deterministic program that generates the whole text verbatim, without any input, without any coin-tossing.  (And, even then, the text will still contain about as many bits of information -- "meaning" -- as are needed to code that program in some concise language.)

But that is not what the SCM authors have shown.  The SCM does not do that. IIUC, at every output word it tosses a coin to decide whether to start a new "self-citation" or take one more word from the current one. In the first case, it needs a bunch of bits (~14 on average for a VMS-size text) to decide at which past word it should start the citation. In the second case it needs another bunch of random bits to decide whether and how to mutate the next word.  

Even if this summary is wrong in some details, the important point is that the method is a probabilistic algorithm that needs several coin tosses per output word, on average.  Now, a probabilistic program is equivalent to a deterministic one that reads a stream of bits from an external source and pretends that they are the outcomes of the coin tosses.  Thus the SCM is not really a generator but a transducer that maps a stream of bits to a stream of words.  That is, an encoding method that can be used to encode a message -- by feeding it as the stream of "coin tosses".

So even if there exists an input bit sequence (coin toss outcomes) that causes  SCM to output an exact copy of the VMS (which I doubt, but let's assume it does), in order to show that the VMS "has no meaning" one would have to show that this magical bit sequence is not a meaningful message.  But there is no test that will show that a bit sequence that cannot be generated deterministically is "meaningless"...

And this is a problem with any claim that the VMS is "meaningless", not just the SCM.

All the best, --jorge

PS. Whenever my prof of automata theory at the U of São Paulo needed an integer for some example, he always used 418.  He explained that, in his experience, that was the best random number for such purposes.  I respect his authority in the matter, and for the last 50 years I have always tried to use that number whenever I needed a random multiplier, hash table size, test input, etc.; and I recommend it to students and co-workers.  Even in those occasions when that turned out to be a bad choice, I had no reason to suspect that it was not mere bad luck...