25-10-2019, 08:16 AM
Thanks Torsten, that it quite helpful.
The pseudo-code suggests that all paragraph-initial lines are generated from earlier paragraph-initial lines and all other lines are generated from earlier text on the same page.
This would make the proposed test: how many words actually comply with this, quite feasible to execute.
One could check how much edit distance is needed to explain the Voynich MS text under the auto-copy hypothesis.
Of course, one has to apply reason to interpret such results. If only 5% of words don't fit, for the case of a small edit distance, then this should not be considered a show-stopper. Hoever, we have not seen such an analysis yet.
The 'evidence' of the network of words does not really support the auto-copy hypothesis, because the existence of such a network is a property of the vocabulary, not of the text.
The existence of word patterns (which helps to explain the network of words) is also a property of the vocabulary, not of the text.
The difference between vocabulary and text is critical.
It is two-fold:
1) the ordering of the words
2) the frequency distribution of the words
If one takes a text and shuffles all words around arbitrarily, the first point is changed, the second is not changed, the text becomes meaningless but the vocabulary stays the same.
The network of words also stays the same. The existence of this network of words is therefore completely independent of the question whether a text is meaningful or not.
However, the existence of such a network is a necessary condition for the auto-copy with small edit distances to work. Since English does not have a network with small edit distances, an English text will not look like the product of an auto-copying process.
However, if one changes the English text by mapping words 1-to-1 with Voynichese words, it *might* look like one.
By re-organising this mapping, it is possible to make it into a text that looks like auto-copying. this was already shown before.
The pseudo-code suggests that all paragraph-initial lines are generated from earlier paragraph-initial lines and all other lines are generated from earlier text on the same page.
This would make the proposed test: how many words actually comply with this, quite feasible to execute.
One could check how much edit distance is needed to explain the Voynich MS text under the auto-copy hypothesis.
Of course, one has to apply reason to interpret such results. If only 5% of words don't fit, for the case of a small edit distance, then this should not be considered a show-stopper. Hoever, we have not seen such an analysis yet.
The 'evidence' of the network of words does not really support the auto-copy hypothesis, because the existence of such a network is a property of the vocabulary, not of the text.
The existence of word patterns (which helps to explain the network of words) is also a property of the vocabulary, not of the text.
The difference between vocabulary and text is critical.
It is two-fold:
1) the ordering of the words
2) the frequency distribution of the words
If one takes a text and shuffles all words around arbitrarily, the first point is changed, the second is not changed, the text becomes meaningless but the vocabulary stays the same.
The network of words also stays the same. The existence of this network of words is therefore completely independent of the question whether a text is meaningful or not.
However, the existence of such a network is a necessary condition for the auto-copy with small edit distances to work. Since English does not have a network with small edit distances, an English text will not look like the product of an auto-copying process.
However, if one changes the English text by mapping words 1-to-1 with Voynichese words, it *might* look like one.
By re-organising this mapping, it is possible to make it into a text that looks like auto-copying. this was already shown before.