25-06-2025, 03:19 PM
(25-06-2025, 02:53 PM)RobGea wrote: You are not allowed to view links. Register or Login to view.color-coding the data
When generating the matrices I had to draw a line somewhere which words to include. I included only those whose initials and finals occurred at least 20 times, and calculated the counts and ratios for every initial-final merger in the matrix.
As you move away from the top corner words start to become rare. As words starts to get very rare the ratios start to get meaningless. Any one more or less occurrence will have a big effect on the ratio. This is where there are many of the 'reds'.
If you were to weigh each ratio by the word frequency then I believe the global ratio would be good.
Because we can only guess what method the authors used I think it is okay for some of the words to fall outside of the ratio range [0.5:2]. In fact it would be unbelievable if the ratios were all very close to parity. It would hint at something too regular, improbable. If the ratio values were widely scattered then it would hint at inconsistency. The middle ground feels just right. Just enough variability to account for mistakes, choices and individual style of the author.