Close to real time. There’s a lag that’s socially acceptable but it’s really short.
I had a bad stutter as a kid. Too many words wanted to pour our at once. Had to slow down vocal muscles and brain. Speech therapist put strawberry paste on the back of my throat for some reason and showed me how to create a little ‘void’ in my head to sort words out quickly before speaking them.
The shortest speed of the world is 13ms. Anything shorter than that is instant. Mutliplayer gaming programmers and network engineers try to work with that 13ms as a golden goal to reach when they can.
Now, 50ms lag is the same as 20 frames per second – and 20 hz.
That’s fast and it’s enough to create an illusion of continuity. Notes lower than 20hz sound like beats (some people hear down to 16 hz but the same rule applies), and video running at 20 frames per second is fast enough for an illusion of motion, although it’s possible to notice the difference between 20 frames per second, 30 frames per seconds, 60 frames per second, etc.
Anyway, there’s processing time available. The grammar systems, where you can mentally form words to respond in complete sentences such as “Oh no, I don’t like this” is at the p400 mark – that’s 400 ms. Less than the tick of a clock but an eternity within one’s mind when focused.
Lag is generally not noticeable until about 150ms though. That’s still pre-verbal but you’ll know it in your head or “gut’.