Lipogrammatic text

[geoffreycaveney]

I agree. The text in the actual examples is perfectly grammatical and of course it behaves like language. The lipogram exercise is only interesting if you produce a correct text; anything else would be trivial and doesn't even qualify as a lipogram.

A kind of statistics that is obviously affected is the list of the most frequent words. I compared You are not allowed to view links. Register or Login to view. with an extract by Dumas of a similar length. Four of the 10 top words by Dumas are in the top 20 by Perec. The exceptions are:

• 5 words that contain 'e'
  
• 'vous'
  

The difference between Currier A and B is much more dramatic (only daiin is consistenlty present among the high-ranking words).

Here I have plotted conditional entropy vs MATTR200 for the files in Brian Cham's corpus. I also included Currier A and B from the VMS (ZL transliteration) and the Perec and Dumas files. This shows that lipograms have no significant effect on these two measures.

Marco, thank you for this analysis. I would be curious to see what effect my "extreme lipogram" English text of 953 words with no use of the letters A, B, and C has on such statistics as the most frequent words, conditional entropy, etc. I know it will not have the statistical structure of the Voynich ms text. But I am just wondering how different it will be from normal English, and in which ways, and how such things show up in various statistical measures. As far as I know, it may be the only example of an "extreme lipogram" text with multiple letters excluded from the same text in a lipogrammatic style.