Ruby's Greek Thread

[MarcoP]

Simple substitutions failed over and over again with several European languages, Greek included, and we know why. As you say, if this simple method worked, the text would have been deciphered long ago.

Simple substitutions of European languages are therefore excluded. Research can follow three other directions:

• ...
  
• ...
  
• A simple substitution / phonetic rendering of an artificial language or possibly an exotic language (e.g. a monosyllabic language, as proposed by Stolfi).
  

I don't quite understand your logic, Marco.

The logic is not complex. Simple substitutions preserve most of the statistical features of the source language, for instance character conditional entropy, the binomial distribution of word lengths, consecutive repetition of words. The source language must therefore match the values of these properties observed in the VMS. European languages don't match these properties and are therefore ruled out for a simple substitution.

On the other hand, an artificial language is purely hypothetical and we cannot exclude that one was invented that showed these features. The case for exotic languages is less clear: You are not allowed to view links. Register or Login to view. discussed how some Eastern languages have a similar word-length distribution as the VMS. You are not allowed to view links. Register or Login to view. posted an example of the amazing rate of apparent repetition in Chinese (when rendered with the Latin alphabet). A modern writing system for Vietnamese (VIQR) mentioned by Stolfi results in character entropy values comparable with those of the VMS.

The following plot shows conditional character entropy (X) and distance (RMSE) of token lengths from the binomial distribution (Y). Viet VIQR is the blue diamond that appears within the cloud of pink Voynich samples.
Green samples are Greek texts and yellow are Latin and Romance languages. Three-letters labels correspond to samples from You are not allowed to view links. Register or Login to view..

   

These three plots compare the distribution of token lengths in Cham's Greek sample ("History of Animals" by Aristotle), VIQR Vietnamese and CUVA-encoded Herbal B. As can be seen, Greek has a peak for 3 character long words and a "fat" tail made of longer words. Viet and Voynichese peak at 4 characters and are nearly symmetrical, with fewer short words and longer words than Greek (EVA results in a similar shape, but with a peak at 5 characters instead of 4). Other European languages show distributions similar to Greek.

   

It is well possible that Eastern languages could be dismissed by some other quantitative feature: there are few Voynich researchers from East Asia and we collectively lack knowledge about these languages. But the case for an artificial language is so speculative that I cannot think of any way to dismiss it.