The 'Chinese' Theory: For and Against

[Jorge_Stolfi]

Claire Bowen and Luke Lindemann's work on entropy was a pretty big step I think, to add to the list. 

Rene and others had investigated the entropy of words and characters some time between 2000 and 2005.  IIRC, they observed that the n-order entropy per character of Voynichese is lower than that of English or Latin, but unlike these two it remains somewhat constant along each "word", so that in the end the entropy per word is roughly the same, ~10 bits.  

From a quick reading of You are not allowed to view links. Register or Login to view., I gather that they did a much more thorough investigation of the second-order character entropy h2 of Voynichese and many other language samples.  They again noted that the values for Voynichese are significantly lower than those for all the other samples.   

But I don't see whether they got any significant insight from those numbers, besides what Rene and others had already concluded.  The low value of h2, by itself, does not say much.  As I pointed out before, that number is a property of the spelling system, not of the language.  One can cut the h2 entropy of English by half bbyy ddoouubblliinngg eevveerryy letter.  Or bcyz fgoplmlmopwxijnogh efvwefrsyz lmeftutuefrs wxijtuhi tuhief noefxytu lmeftutuefrs opfg tuhief ablmpqhiabbceftu.   It is almost expected that the spelling system of Voynichese, having been recently invented, would be much less efficient (in terms of bits of information per character) than any spelling system that has been used and evolved for centuries.

The variation of h2 depending on position along a word does imply that Voynichese is not a simple substitution cipher or phonetic rendition of a Western language.  But that had already been observed before, and I did not see this point discussed in the paper.  Was it?

The entropy per word is more interesting, because it should not be much affected by the spelling system (other than the splitting or joining of compounds) or even by any encryption scheme that maps each lexeme (word type) always to the same string, preserving word spaces. Including a codebook cipher.   And, in this metric, Voynichese was already found to be fairly similar to other natural languages.  Did Bowern and Lindemann investigate this point?

All the best, --stolfi