Updated PCA analysis of various languages

[DonaldFisk]

Thank you, Donald! I think the PCA diagrams provide a useful tool to compare different scripts.
You write that the diagrams can be "rotated or reflected" and mean the same: I guess that a 90 degrees rotation would swap the first and second components, and I am not sure that would be desirable. For instance, in Italian it is the X axis (1st component, I guess) that separates vowels from consonants. But mirroring one or both axis so that the most frequent letter (excluding *) is in a fixed quadrant (e.g. bottom-right) could help the visual comparison.

In your analysis, your write that some of the plots "vaguely resemble the Voynichese plot". It would be great to have a quantitative similarity measure as an indicator of how much two plots are close or different.

Mirroring PCA plots is very easy for me to do and I'll ensure that comparable plots are aligned the same way in future.

Comparing PCA plots is more problematic.   There are standard algorithms for matching points in different 3D scenes but these aren't suitable here.   It's better to identify plausible languages by eye first (see my other post) and then match up the glyphs by trial and error, while ensuring the output texts share properties (ideally, words).   The main concern here is to avoid outputting gibberish.   This is something cryptographers will know more about than me, but I'm interested in adding useful algorithms to my text-processing library.