30-03-2018, 09:54 PM
No that's not correct.
The logic should be that you should estimate the probability of the word to be "daiin", provided that the immediately preceding biword (is "biword" a valid English word?) is "daiin X". Now, the probability of the biword "daiin X" to occur is not the same as the probability of "daiin" to occur, since it should account for the exclusion of "daiin daiin" (as it does not fit the pattern).
Now, in estimating this, one could perform actual counts across the VMS corpus, but then the "expected" values would be the same as observed. Alternatively, one could assume some random word distribution, e.g. Zipf's law, for the "expectation".
When calculating this for King's Bible etc, the expectations can be based on more broad language statistics, but for the Voynichese we don't have anything beyond the VMS itself.
The logic should be that you should estimate the probability of the word to be "daiin", provided that the immediately preceding biword (is "biword" a valid English word?) is "daiin X". Now, the probability of the biword "daiin X" to occur is not the same as the probability of "daiin" to occur, since it should account for the exclusion of "daiin daiin" (as it does not fit the pattern).
Now, in estimating this, one could perform actual counts across the VMS corpus, but then the "expected" values would be the same as observed. Alternatively, one could assume some random word distribution, e.g. Zipf's law, for the "expectation".
When calculating this for King's Bible etc, the expectations can be based on more broad language statistics, but for the Voynichese we don't have anything beyond the VMS itself.