I had this idea to use regular expressions for term extraction but you know: I'm a busy man.
Atril has released a new build of Déjà Vu and they are showing what can be done with CafeTran too in an Atril video.
I think that it can be done in CafeTran more elegantly.
Perhaps Dominique can make a video about it, since his screencasts are higher rated than mine :). Perhaps it's the French accent?