Session 9: SEMANTIC RESOLUTION A GENERAL-PURPOSE LANGUAGE TRANSLATION PROGRAM FOR THE IBM 650 COMPUTER

I will discuss the application of redundancy-sorting techniques to the resolution of semantic problems. I must leave it to you to decide whether or not these techniques actually deal with semantic problems. I believe they do, even though this paper describes a manual of operation for a computer program. I should perhaps explain that the term "redundancy sorting" refers to a battery of data-processing techniques that operate directly on the sequences of language to obtain information which is useful for the purpose of translation. Because these operations have been designed for application to sequences in general, the adequacy, or inadequacy of their results is independent of the particular language processed; for the same reason, the sequences processed need not be those of language, but might represent any sort of non-verbal behavior, such as chess playing, etc. In applying these procedures to machine translation, one thinks of the sequences processed as a set of examples which represent, more or less adequately, the relative distribution of the various linguistic entities contained in the set. To this set of source-language examples there corresponds a second set each element of which is a target-language equivalent of one of the source-language examples. In a typical situation, the input to the translation procedure will be a sequence constituted of source-language entities, but not contained in the set of examples. Upon the presentation of such an input, the procedure is to compare each source-language example with the expression to be translated, and to compute for each example a similarity rating located within a range whose maximum value indicates identity of the two sequences, and whose minimum value is zero. The result of each computation is then assigned to the target-language equivalent of the source example just processed, on the assumption that each target-language example resembles the desired translation as much as its source-language counterpart resembles the input.