[andolanra]

This research is built on some pretty shaky ground—including some very loaded picking-and-choosing of vocabulary—and because it doesn't have very strong predictive power, hasn't made much of a dent in the community in the last decade since it's been published. Here's a discussion from the blog Language Log that discusses a lot of the methodological problems that show up: https://languagelog.ldc.upenn.edu/nll/?p=4612

[mcswell]

Nuts, you beat me to it; I was going to provide that exact same link, together with an excerpt from a comment: "The LWED database reconstructions just aren't good enough to form the basis of a successful effort along those lines."

Here's another relevant post at Language Log: https://languagelog.ldc.upenn.edu/nll/?p=4634. (Before anyone gets too excited, notice the "filed under" terms.)

[ummonk]

Aside from the specific details, I feel like papers of this nature would be on much firmer ground if they included controls. In this case for example, that would mean showing that are greater word-correspondences in these eurasian language families than with other language families.

[topaz0]

I remember that languagelog post. Some great critique in there.