Modifications have produced a markov that produces variable-length, randomly-started chains. This isn't ideal (ideal would be flagging start-prefixes, and picking randomly from one of them, which I suspect I could actually do) but for the moment it's interesting.
Of course, at this point it's taking a (sort of) regular text file as input. I need to provide log-strippin' and some restructuring to avoid common bad sentence constructions in IRC (seeing as how the Markov's going to mangle it anyway, and then the Engrish will mangle it further).
