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

[quimqu]

It would be also interesting to see what @npcompl33t's serialization algorithm achieves on @quimqu's text: horseshoe or no horseshoe?

@Mauro I gave it a try. I rebuilt npcompl33t’s seriation from what he explains in his posts, so it’s my version, not his code: for each page, which glyph tends to follow which, then how different two pages are, and then MDS to draw them on a map.

Horseshoe: yes, the skribe makes one too. In the first figure the real text is on the left and the skribe on the right. Same arch, with the pharmaceutical pages at the top, herbal A on one side, herbal B, stars and biological on the other, in the same order. It’s the same in all five runs. And the pages of the same bifolium sit just as close together in the skribe as in the real text. The skribe has no chronology at all, only the habits of each hand and section and one “taste” per sheet, so that is enough to give the horseshoe.

       

In the second figure I checked if the herbal A pages drift along the map with the folio number. At first it looked like they do, but it’s only the herbal pages bound in the pharmaceutical quires (f87 to f96, the orange circles), which are written a bit differently and lean towards the pharmaceutical style. Without them there is no real trend, and the skribe varies about the same. The skribe writes those pages like any other herbal page, so that’s one small thing it misses.

Script and figures are on GitHub, in external_tests/npcompl33t. @npcompl33t, if you want to compare it with yours, I’d be happy to.