Writing a fake Voynich by rules: copies, habits and the link between words

[Mauro]

Mauro, just to be clear: I’m not saying I don’t see what npcompl33t sees. It’s the opposite. I see the same changes and the same horseshoe, and my skribe makes them too. 

Yes I understood that, and indeed that is just what I find to be unexpected. I try to be clearer too.

@npcompl33t ran his algorithm on the VMS and found the horseshoe. He attributes the horseshoe to the existence of a gradient along the manuscript at the bifolios levels (with bigger jumps between sections). The question (raised mostly by @Stolfi) is: is this gradient along bifolios actually a feature of the manuscript, or is it just a happenstance, like in the example of a bookshelf ordered by books height?

I have checked some of @npcompl33t's links and it looks to me the literature supports his position. Then I checked the sequence retrieved by @npcompl33t against the Q13 and Q20 sequences found by Lisa and, to my surprise, they agree, with raw odds of this being due to chance of one in a few thousands. So I considered this matter settled: there is a gradient along VMS bifolios and this gradient can be retrieved with a reasonable margin of certainty (at least in Q13, Q20 and in the overall manuscript structure).

But now comes @quimqu: he builds an ingenious writer, who knows about sections but who has no gradient along bifolios. You set the parameters of each bifolio (page, actually) randomly around the section baseline (if I understood correctly), so there's no gradient in your bifolios (more precisely: the gradient inside each section is flat, but noisy). Your writer should be akin to @Stolfi's bookshelf, where books have a random heigth around a mean and their ordering by height has no meaning.

But nonetheless, using the same means as @npcompl33t, you find the horseshoe, which should mean a gradient instead exists and is significant, and this is what puzzles me.

The same goes for Lisa's paper: your Skribe with her algorithm gets similar results, in the sense that you find an ordering of bifolios which greatly increases the LSA score, just as Lisa found for the VMS. But your Skribe has no such ordering to start with, so the question is open: what's actually LSA finding in the VMS, a real gradient or just a happenstance?

To sum it up, we have: two different methods which find a gradient at the bifolios level in the VMS, to the point to retrieve essentialy the same sequence in two quires. And then a gradient-less Skribe, where nonetheless the same methods find a gradient which does not exist and, accordingly, should not be found. How can this be?

You also (if I'm not mistaken) re-ran @npcompl33t's and Lisa's algorithms on the VMS on your own, but then you did not find the same sequences (not sure for Lisa's, did your LSA analysis find the same Q13 and Q20 sequence as hers?). Which is also weird.


You guys help me sleep better tonight, I hate contradictions