Machine translation: An American perspective

Translation of human languages was one of the first applications considered for digital computers. Indeed, the idea of mechanizing translation predated the invention of such machines, but it was only after World War II, when digital computers became generally available, that Machine Translation (MT) was taken up seriously [4]. This paper explores the field or MT from the viewpoint of an American researcher/developer, and is intended for nonspecialists. We will not dwell on historical details; see [4] for an introduction. Nor will we present a comprehensive survey of MT systems, nor a long bibliography, since [10], [11] provide these. Instead, we will concentrate on aspects of R & D that, in the author's opinion, are of critical importance in developing any large-scale Natural Language Processing (NLP) application, which MT exemplifies. Case studies will be drawn from our experience in developing an MT system that is now being marketed. For context, surrounding sections of this paper will sketch the history of MT, then outline the nature of the translation problem, present various methodological approaches to solving the problem, and, following the discussion of R & D issues, consider future directions for MT research.