Lojban morphology
for some people complaining about it.
Jay Kominek notices that there is a body of literature describing algorithms which can produce finite automata, given positive and negative examples of the strings which the automata are supposed to recognize. Presumably someone with enough spare time could write some software which takes a big list of examples, and generates an automaton. (And then, whenever you find things that should be, but aren't, or are, but shouldn't be recognized, then can just be added to the appropriate list, and the automata regenerated, rather than the current system used by la jbofi'e. Probably not the best way of generating a human-interpretable description of the morphology, but it sure would be convenient from the stand point of automated recognition.)