If you use the word scientific that way I think you lose a quite valuable word. I consider NLP to be extrapolated from evidence. I even have seen it tested directly a variety of times. At the same time I don't consider it to be scientific in the popular usage of 'scientific'.

For discussion on LW I think Keith Stanovich criteria's for science are good:

Three of the most important [criteria of science] are that (1) science employs methods of systematic empiricism; (2) it aims for knowledge that is publicly verifiable; and (3) it seeks problems that are empirically solvable and that yield testable theories.