Around 1968–1970, Terry Winograd developed SHRDLU at MIT: a program that understood English commands about a simulated blocks world, moved pieces, and answered questions about state.
Its strength was integration: syntax, semantics, and domain knowledge locked inside a limited environment. It showed natural language “worked” when the world was small and known.
The ambivalent lesson—success in micro-worlds, difficulty leaving them—shaped decades of symbolic NLP until statistical and neural methods changed the scale.
