TXR looks rather like the CRM114 language that's been used to implement some rather amazingly accurate text classifiers (some better than most people on their own mail), though a bit less bizarre and I think more accessible:
http://crm114.sourceforge.net/docs/INTRO.txt
CRM114 too treats pattern matching as the fundamental construct and has blazing performance for it and certain kinds of number crunching (it has to), but I don't think it's nearly as useful for the average hacker trying to munge a couple of text files. Still, worth a look both to users and possibly to language implementors. I'm definit