Previous Experiments in Mechanical Translation

Disregarding experiments in simulated machine translation, which in place of a computer employs a person with no knowledge of the language to be translated acting in accordance with a set of given rules, only three actual attempts have been made, to my knowledge, to translate in this way. These were by the Institute of Languages and Linguistics of Georgetown University in association with the International Business Machines Corporation (1954, Russian to English); by the Computation Laboratory of Birkbeck College, London, under the direction of Dr. A. D. Booth (1955, French to English); and by the Institutes of Scientific Information and of Precise Mechanics and Computing Technique at the U. S. S. R. Academy of Sciences, Moscow (1955/6, English to Russian). Little can be said about the IBM experiment, because little definite is known. An account of it is given by Dr. L. E. Dostert in Machine translation of Languages, chap. 8. pp. 124 ff., but he has unfortunately allowed himself to be sidetracked into discussing problems arising out of the experiment rather than the experiment itself, and the reader is consequently left to guess at the exact details of the latter. From what the author does say, however, it may be gathered that the experiment was designed to translate a number of specified Russian sentences into English. The lexicon required consisted of 250 terms, each with one or, at most, two English equivalents, but whether this figure included the different (inflected) forms of nouns, verbs etc. or not, is not stated. Nor, in fact, is anything said on the subject of morphology, except that with a “complex” Russian word (i. e. one divisible into a stem and ending) the two parts were coded and stored separately. For dealing with syntax and the choice of the correct English equivalent, where more than one was given, a set of operations “limited in number but widely significant in scope” was provided.