MACHINE the reasonable TRANSLATION ▬ middle ground FACT OR FANCY ?

few years ago, a number of news stories appeared here and there announcing "fundamental breakthroughs" in the development of machine translation from Russian into English and declaring that workable machine translation systems, if not already a reality, were just around the corner. Since then, one or two government agencies have begun to operate automatic translation facilities with which some users were partly satisfied, some not at all, and none completely. More recently, a committee constituted by the National Academy of Sciences, National Research Council—the Automatic Language Processing Advisory Committee (ALPAC)—conducted a two-year study of the field of machine translation and came to the conclusion that "without recourse to human translation or editing . . none is in immediate prospect." Where, then, do we stand in machine translation? Were the claims justified that were made in the earlier days, or is ALPAC correct in concluding that there is no prospect for its achievement in the foreseeable future? In my opinion, both are wrong. To substantiate my view, let me give a brief survey of the state of machine translation. First, let me make clear that the field of machine translation is (with one glaring exception—the photoscopic disc) not primarily concerned with the design of a special translation machine, but with the design of translation programs to be run on large general-purpose computers. Let me add that the major effort so far in this country has been directed toward the machine translation of Russian into English, although some experimental work has also been done on other languages (such as Chinese and German). Two extreme approaches have been taken to the field, and one which I consider a reasonable middle ground.