The missing subroutines all seem to want to pick random words from a dictionary file, as per Ode to a random number. The improvement I haven't yet figured out (without a lot of manual typing in) is to have a modified dictionary file with a part of speech field alongside the word. i.e. in your example you want your insult to call somone an adjective noun or an adverb adjective noun and the filtering should be data-driven rather than by using different subroutines.