

Yes, and that’s a good thing if you don’t want it to start killing processes. You have that extra time/space to deal with the out-of-memory condition yourself.
Or you can ignore that condition and continue using the system in a degraded state, with swap as “disk RAM”.
No, but it’s a hell of a lot easier to put huge language datasets into the machine learning blender and get a model out, instead of manually programming every conceivable linguistic construction.