How soon can we program computers to understand pragmatics behind natural language?
This is bigger than just pragmatics behind natural language. This involves the whole issue of embodiment in robotics, as well as the limitations on game AI of working outside the box. Pragmatics is something like reading between the lines. Pragmatics is next level, on the way to AGI, somewhere between discourse and cognition. Pragmatics is a kind of semantics based on non-literal inferences of conceptual cues external to text. Such cues may come from so-called common sense knowledge abstracted about the world, such asPresupposition. Think of it as a sort of advanced sentiment analysis for concepts.
IMHO, whenever all the tasks of NLP which are being done manually or are semi-automated today become fully automated, only then can we move to the next level and begin to build manual or semi-automated systems on top of a foundation of fully automated building blocks. There is of course that ever present bugaboo of Technological singularity to take into consideration, when automated systems will hypothetically be able to build automated systems; however, the one thing I’m confident about is that I won’t see it in my lifetime, and I seriously doubt in yours as well. ;^)
See also my quick and dirty webpage: