A key to understand the VMS

[ReneZ]

I see in the (interesting) work of Torsten four aspects. In the discussion these are occasionally mixed, or different people prefer to concentrate on different aspects. It may therefore be useful to identify the four clearly.

1) There is the observation that words in the Voynich MS tend to be very similar. The large majority of words form a big family where all words in this family can be converted to other words by a single change (addition, deletion or modification of 1 character). The number of changes needed to go from one word to any other is called the edit distance, defined using the Eva transcription alphabet.

2) There is another observation that words with an edit distance of 1 tend to cluster together in the MS.

3) There is the question whether the above features are the result of an intentional design, or only a by-product of the manner in which the text was generated.

Here we have a minor ambiguity. Points 1 and 2 concern different aspects of the text, and point 3 may refer to one or the other, or to both. The 'auto-copying hypothesis', as far as I understand it, argues that both 1 and 2 are intentional.

4) There is the question whether the text is meaningful or not.

The auto-copying hypothesis would also seem to argue that the text is meaningless.

I think that it is important to separate this highly contentious question from the observations on which it is based. This tentative conclusion may be wrong, but there is certainly something of value in the observations. So let's look at them again.

Point 1 concerns the vocabulary of the MS. There seems to be a rule how words can be formed.

Point 2 concerns the way these words are combined in the text. Basically a rule about syntax.

Both 1 and 2 are observations, that can be verified by statistics. Points 3 and 4 are quite different, in that they represent and interpretation, or possible explanation of either or both.

Point 1 is not too controversial. The interesting plots (and the modified version by Marco) basically show this 'family' of words in a big cloud. At the same time, the argument of Emma presented above is important. The changes related to edit distance 1 are not arbitrary. They follow relatively strict rules.

Point 2 is already more tricky. Anyone who has looked at the text intensively (especially anyone who has transcribed the text) will have noticed this, so it is there. But how much? To what extent? To the extent that it is unusual?
Torsten has shown some clear examples, so again, it is there.
But what if one takes an arbitrary page. Can one establish some kind of 'score' for the use of similar words?

The following thought experiment may illustrate why points 1 and 2 are very different.
Just imagine that someone has a plain text with less than 10,000 different words. He sorts these words according to decreasing frequency, and represents the most frequent word by 1, the second most frequent by 2, etc. to 9999.
If this plain text is then converted word by word to these numbers, we end up with a new text that exactly fits the point 1. All words can be converted to other words by a single change (edit distance 1).

However, the objection of Emma does not apply in this case. Changes can be made arbitrarily, at any point. To solve that, let's now assume that he does not write 1,2,3, .... 9999 but uses roman numerals instead: I, II, III ... Again, the edit distance 1 is preserved throughout the entire vocabulary, but in addition, changes are subject to very specific rules.

However, for both cases, there is nothing at all that would explain the point 2: collocation of similar words. 

What this means is:
Whether point 1 is a correct observation or not does not say anything about whether the text is meaningful or not.
The critical point for this question is only point 2.
The reason why points 1 and 2 should occur (if they do) is not yet clear, but worth exploring further.