The present state of research on mechanical translation

Abstract

One of the interesting examples of the influence which a newly invented tool may exert in opening up fresh lines in theoretical research and in advancing new techniques for the solution of old problems is provided by the rise of electronic computers. They were originally designed to solve certain mathematical problems quicker than the human brain could but it soon turned out that their components, in which elementary logical and computational operations could be carried out at extremely high speed, might well be recombined to yield similar results in noncomputational fields. When electronic computers were still in their infancy, in 1945, the question was raised whether a computer-like machine could not be designed that would automatically translate from one language to another. The story of this idea is told by the man who apparently conceived it first, Dr. Warren Weaver, Director of the Natural Sciences Division of the Rockefeller Foundation, in a memorandum of July 15, 1949. This memorandum aroused considerable interest, followed by some active research. Some of the first steps towards a solution of the problem of mechanical translation are described in another memorandum by Dr. Weaver, dated March 6, 1951. The present paper summarizes the results achieved up to the end of 1951. They involve a clearer understanding of the aims of machine translation, of various possible divisions of labor between man and machine in a translation partnership, and of the preliminary steps that have to be taken before the final solution of the problem can be found. Some of these steps seem to have independent value and especially the task of providing for an operational syntax (see ref. 4) is a challenge that should appeal to structurally-minded linguists and give a new twist to their investigations. AIMS Interest in mechanical translation (MT) may arise through sheer intellectual curiosity concerning a problem whose solution, perhaps even attempted solutions, will in all probability provide valuable insights into the functioning of linguistic communication. Interest may also arise from many practical standpoints. One of these is the urgency of having foreign language publications, mainly in the fields of science, finance, and diplomacy, translated with high accuracy and reasonable speed: the scarcity of expert bilinguals is causing a log jam in scientific translation which is costing research an amount that can hardly be estimated but might well run into millions of dollars yearly, due to the fact that important scientific methods and results are not made available in time or perhaps not at all to research workers. Another is the need of highspeed, though perhaps low-accuracy, scanning through the huge printed output [of actual or potential enemies,] in newspapers, journals, propaganda leaflets, etc. These two aims are only partly overlapping, but a good method of achieving one of them would probably be of great help in attacking the other, since we apparently have here another case of a wellknown situation where accuracy may be traded for speed, and vice versa.