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

[quimqu]

Compared to the manuscript there is a greater frequency of gallows words in your generated text.

Also the manuscript has the property that a gallows word follows another gallows word more frequently than would be expected if words were shuffled randomly. Your generated text doesn't have this to the same degree.

Thank you, I'll check this rule and add it to the generator.

Please remember I don't believe much on this generators... but it is interesting.

@dashstofsk, I checked it as promised.

I wasn’t sure what you count as a “gallows word”, so I tried both:

• Any word containing a gallows: I don’t see either effect. In quire M the scribe has slightly fewer of them (50% vs 53%), and they cluster about the same as in the manuscript.
  
• A word starting with a gallows (t, k, p, f): both of your observations hold. The scribe has more of them (5.8% vs 4.6%) and no clustering at all (0.94 vs 1.48).
  

So I took the second one. Is that what you meant?

The interesting part is where the clustering comes from. Leaving out the first word of each line (mostly paragraph starts, which the scribe already gets right), the real text gives 1.45. If I shuffle the words inside each line it still gives 1.46. So it isn’t one gallows word calling the next one. It’s the lines: some lines are full of words starting with a gallows and others have almost none. Shuffling within the page brings it down to 1.16, and shuffling the whole book to 1.0.

I added it to the scribe as a line habit: each line, and a little each page, is more or less inclined to start words with a gallows. Now it gives 1.42 (real 1.45), and these words go from 4.2% to 4.8% (real 5.1%). Nothing else in the battery got worse, and it holds when the scribe learns only from the odd folios and writes the even ones.