15-08-2020, 09:53 PM
It is interesting to see how language groups flowed out fairly naturally from the data, in groupings that one might expect. There are almost always ways to tweak an algorithm to improve it, but this is a promising start.

