A key to understand the VMS

[Torsten]

I think this is a good argument against the Natural Language hypothesizers.

There is much what can be said about this network of similar words. If you think about what it is necessary to generate even 7000 different word types where the types has to differ in only one glyph from each other! In my eyes it is obvious that you can only reach such a goal with some methodical approach. For the VMS some more observations are important. 

There is the shift from Currier A to Currier B which happens since words near "chol" are frequently used in Currier A and word types near "chedy" are frequently used in Currier B. This is an indication for a change of the word set during writing the manuscript. With other words this indicates that the words where generated during writing the manuscript.
(Another feature pointing in the the same direction is that the words used in the text  is depend in some way on there position within a page or line. Some words are for instance typical for a specific position within a line.)

Last but not least it is possible to draw such a graph for each page of the manuscript. This demonstrates that similar words do co-occur for each page. This is an indication for a local element in the method to generate the text. See for instance the pairs daiin/dain and chol/chor on page You are not allowed to view links. Register or Login to view. in You are not allowed to view links. Register or Login to view.:

   

... the selection of the different groups is fairly arbitrary.

The groups are selected by the most common words "daiin", "ol" and "chedy". If you take also the frequency for each type into account it makes sense. 

See for instance "daiin" and "ol":

aiiin( 41) daiin  ( 17)
aiin (469) daiin  (863)
ain  ( 89) dain   (211)
an   (  7) dan    ( 20)
aiir ( 23) daiir  ( 23)
air  ( 74) dair   (106) 
ar   (350) dar    (318) 
ail  (  5) dail   (  2)
al   (260) dal    (253)
am   ( 88) dam    ( 98)
os   ( 29) dos    (  1)
or   (363) dor    ( 73)
ol   (537) dol    (117)

The word types near "ol" are frequent and the word types near "daiin" are frequent. It is more as if the two groups of word types do overlap. 

I have uploaded the gephi project for the graph for the complete manuscript. You need to download the free tool gephi from You are not allowed to view links. Register or Login to view. to open it.

  VoynichComplete.gephi.zip (Size: 493.66 KB / Downloads: 43)