Today I scripted a very small proof of concept of an adaptive “gag”. The idea is that instead of using a fixed blacklist of words you cannot use, this one works the other way and you can only use words that are whitelisted.
Then things got slightly out of control and I made a very basic “learning” system, where the object listens on local chat and if a word is repeated enough times, it is automatically added to the known word list.
That’s pretty much like learning to talk I suppose, excepted that your vocabulary is entirely contained in a single 64k script, which makes it extremely limited, this is waaaay below “bimbo” level.
I’m running a few tests, trying to tweak the learning mechanism to avoid clogging it with garbage… Having it start from zero does suck quite a bit, it makes it impossible to be understood.