Discussion of "A possible generating algorithm of the Voynich manuscript"

[MarcoP]

In general about edit distance: if two three-letter words are completely different, do they have an edit distance of three?

Hi Rene,
yes, Levenshtein distance works like that. A related measure (Levenshtein ratio) computes a similarity coefficient in the 0-1 range that takes length into account.

This is the graph for the new experiments. The Pliny files are those you linked.

   

I encoded Machiavelli as suggested by you and Koen, by mapping words according to number of tokens, so the mapping is:
  e  -> daiin 
  che  -> ol 
  di  -> chedy 
  non  -> aiin 
  a  -> shedy 
  ...
The most frequent Italian words are mapped into the most frequent Voynichese words.

This mapping, as what you did with Pliny, does not impact distance=0, which still is complementary to the original TTR. Pliny has too many types per N tokens to be a good match, but it seems clear that both methods of encoding produce a number of similar tokens that can also appear next to each other.

These methods do not generate exact reduplication (which, as you explained above, should be present in the original text, in order to appear in the ciphered output).
They also do not generate sequences made of several occurrences of two similar word-types, like the famous 5-tokens / 2-types
<f75r.38,+P0>    qokeedy.qokeedy.qokedy.qokedy.qokeedy
These are the number of sequences with length 3 or more (since the files have different lengths, I include the number of K-words). 
   

The only other occurrence with the exception of the VMS and Timm's text is in the original Machiavelli: "ne se ne".
9 of the 45 occurrences in the VMS do not involve exact reduplication, e.g.:

<f76r.34,+P0>    cheor.shey.qoolkal.shedy.shedy.shey.shedy.ollchy.shlches.shcthy.sain.oly
<f107v.39,+P0>  okain.cheor.olkaiin.oain.cheary.raiin.okaiin.odaiin.okaiin.y
<f108r.32,+P0>  ykedar.chedy.qokey.lkeedy.otedy.okedy.otedy.otal.shol.alodar.or.alold