Automatic Translation of Natural Languages
Martin Kay · 2016
The history of man's attempt to build a translating machine for natural languages has not been illustrious. There has probably been no other scientific enterprise in which so much money has been spent on so many projects that promised so little. In the late fifties and early sixties, numerous people obtained, from one agency or another of the United States government, appreciable sums of money, in return for which they promised to deliver, in a very few years, a computer program or even an actual machine that could produce high-quality translations automat ically. The events that brought these euphoric days to a sudden end are, by now, well known even to people who have no other knowledge of work in machine translation. Stimulated partly by the displeasure of some high-ranking civil servants and military officers at having received less than the best value for their money, the National Academy of Sciences in 1962 established the Automatic Language Processing Advisory Com mittee (ALPAC) and ordered it to investigate the entire matter of the federal sponsorship of research on machine translation. In its report, delivered in 1963,1 ALPAC was as kind to the designers of automatic-translation machines as it could possibly be. It concluded that there was no possibility of producing a satisfactory translating machine in the foreseeable future and recommended that no further funds be spent on contracts that had such development in view. The committee did not, however, see the development of such machines as forever beyond the wit of man and, in fact, expressed support of the funding of research that aimed at hastening the day when it would be reasonable to let such a development contract. Reactions to the report were predictable. For almost ten years, any application for financial support for a project involving language and computers, however modest or sound, could expect a swift and categorical refusal. None of the positive recommendations of the ALPAC report were acted upon, and a disservice may thereby have been done to many serious and inventive research workers as well as to the country. Nevertheless, although the number of research projects in computational linguistics has diminished, the discipline has attained far greater maturity. It required dedication to stay in a field that no longer had a ready source of money 217