That was one thing that came to my mind. I also wonder what it'd look like by minimizing LCS score of anagram pairs. Too lazy to code it right now though =/
Finding ones where the two words have some relation surfaced some interesting ones. I tried using Wordnet to do that.
Unfortunately, even after stemming them, most of the list wasn't in Wordnet, so the list is pretty short. The more interesting words are ones with a score lower than 1, so scroll down a bit: https://gist.github.com/anonymous/e69f448fa08481603d1ca77ea7...
