Under what circumstances do apple's dictation models “learn”?

Is there any scope for specifically training or teaching it to match individual voice and usage patterns? I would be happy to do whatever is needed to help voice control dictation become more accurate to my voice and vocabulary patterns, but I've been unable to find clear enough indication of under what circumstances it will learn (I'm aware of "add to vocabulary ", I mean any learning it may be doing beyond that - audio, vocabulary signatures, context. In writing this I'm even starting to wonder whether it actually does try to learn your audio, or usage patterns; by which I mean which words you're likely/unlikely to be using, perhaps in any particular context; maybe it doesn't?). If I know what kind of student it is, I can be a better teacher. For example, if I dictate and it gets it wrong, then I correct the incorrect text *using the keyboard*, does it learn from that? Alternatively, if I dictate and it gets it wrong and then *I say "correct that"*, is that what’s needed for it to be taking notice and update its models? I haven't stumbled upon evidence either way as yet. It might be nice to try to design an experiment to find out, but I'm not sure how to proceed on that. Does anybody know if the models Apple are using are open source or proprietary? If they’re open source we might be able to find out a bit more. On the other hand I suppose personally asking an academic expert in the subject might be the best way to proceed... ? It's a pity documentation is so sparse! ---- As an example of this, I don't know whether I should be trying to speak in a certain kind of special accent, in the hope that it would recognise my speech better (as we would actually do for many humans - speaking clearly, adaptive to the listener). It's as if you have an assistant taking dictation, but they're a complete black box and you're totally unable to ask them how they work. I don't feel I have a way to "get to know" this assistant. I wish there was a way to do that. Of course for general speech this isn't really an issue, it works pretty well. The problem arises when you start to talk about technical subjects (which may often actually 'peek in' to "general speech"). One specific frustration is that for some unusual words, the assistant will repeatedly assume you are talking about something else, and there doesn't seem a way to untrain it. For example it often mis-recognises the word "Cortana", when there is no situation ever I would want to use that word, it's not in my vocabulary! This happens because that word is similar to a technical word I want to use (and even with that word added to vocabulary, it still mis-recognises to the word I don't want). So this issue about how it learns does come up if you use it for technical subjects... Voice assistant is an assistant, after all...