Yes! The text one entered could be *parsed*, so one could actually
compose more semantically-rich utterances than the "me tarzan, you
jane" level of discourse provided by Window/Icon/Menu/Pointer interfaces.
Instead of replacing the input text with the result, one might even imagine
interleaving input and output on a scrolling display: the upper part
of the screen would serve as a cache of most-recently-used interactions,
so one wouldn't need to use a back button or undo to review recent context.