18-09-2020, 11:31 AM
D.O.,
should the search space be limited to: "taking some previous word and applying some changes to it in order to create the present word to be written"?
The above is a generic formulation of Torsten's approach, and I am not aware of any more precise one.
A forward approach (simulating it) will never generate anything like the Voynich MS text because, if this is how it was done, it was stochastic (though governed by some human behaviour).
A reverse approach (optimising an algorithm with some parameters that best fits what we observe) is what I think you are suggesting, but its validation will be limited by the same problem as for the forward approach.
Furthermore, such an algorithm must also be constrained by the well noted special features of the Voynich MS text, like word patterns, paragraph top line behaviour, line-ending behaviour etc, that necessarily makes the algorithm for 'arbitrarily' changing a previous word to the present one *very* complex.
I believe, from reading a comment at Nick Pelling's blog, that the algorithm in Torsten's app is already quite complex, but it does not yet model these aspects.
I have not checked the Java code, and I will not, but I believe that Nick has.
I wonder if this addresses your question adequately....
should the search space be limited to: "taking some previous word and applying some changes to it in order to create the present word to be written"?
The above is a generic formulation of Torsten's approach, and I am not aware of any more precise one.
A forward approach (simulating it) will never generate anything like the Voynich MS text because, if this is how it was done, it was stochastic (though governed by some human behaviour).
A reverse approach (optimising an algorithm with some parameters that best fits what we observe) is what I think you are suggesting, but its validation will be limited by the same problem as for the forward approach.
Furthermore, such an algorithm must also be constrained by the well noted special features of the Voynich MS text, like word patterns, paragraph top line behaviour, line-ending behaviour etc, that necessarily makes the algorithm for 'arbitrarily' changing a previous word to the present one *very* complex.
I believe, from reading a comment at Nick Pelling's blog, that the algorithm in Torsten's app is already quite complex, but it does not yet model these aspects.
I have not checked the Java code, and I will not, but I believe that Nick has.
I wonder if this addresses your question adequately....