Báo cáo khoa học: Machine Translation Development at the University of Washington

MACHINE TRANSLATION development at the University of Washington is a joint enterprise of the Department of Far Eastern & Slavic Languages & Literature and the Electrical Engineering Department. MT research at our University began in November 1949.
Nội dung trích xuất từ tài liệu:
Báo cáo khoa học: "Machine Translation Development at the University of Washington" [Mechanical Translation, vol.3, no.2, November 1956; pp. 33,41] 33Machine Translation Developmentat the University of WashingtonErwin Reifler, Far Eastern Department, University of Washington, SeattleMACHINE TRANSLATION development at the Supported by two grants from the GraduateUniversity of Washington is a joint enterprise School of our University, Dr. Micklesen carriedof the Department of Far Eastern & Slavic Lan- out two studies. In one he investigated the pro-guages & Literature and the Electrical Engineer- cess of compounding in the Russian languageing Department. and elaborated proposals for the economical MT research at our University began in dissection of compounds by machine. The otherNovember 1949. We realized very early the developed into an exhaustive analysis of MTimportance of a close cooperation between lin- form classes of the Russian language, the pre-guist and engineer and the advantages of work- requisite for the mechanical determination ofing jointly for a definite project with well de- intended grammatical and non-grammaticalfined linguistic and engineering conditions and meaning. He also worked out a complete tabu-limitations. The result was the planning of an lation of all subclasses of Russian paradigmaticMT Pilot Model by Dr. Thomas M. Stout, then form classes and determined the number of dis-of our Electrical Engineering Department, and tinctive forms in each paradigmatic set. Theseits construction under the supervision of Prof. classes are purely formal, representing theHill. most economical (structural) breakdown into During my research, I developed linguistic Stems and endings.solutions for the identification by machine of Dr. Micklesen has also been very much inter-grammatical categories, of both predictable ested in the theoretical aspects of the linguisticand unpredictable compound words whose con- problems of MT. As a structural linguist, hestituents occur in the machine memory, and for has been especially concerned with fitting thethe automatic recognition and transfer to the results of MT research into the general frame-output of words which, both graphically and in work of present-day linguistic thought. He re-meaning, are shared by the two languages con- cently contributed a chapter entitled FORMcerned in the machine translation process. It CLASSES—STRUCTURAL LINGUISTICS ANDis for the purpose of testing the fundamental MECHANICAL TRANSLATION to For Romanengineering feasibility of these linguistic solu- Jacobson (Mouton & Co, The Hague, 1956).tions that the pilot model was planned. Professor Hill has given much of his time to Along with these researches went a steady de- the study of the engineering aspects of a pro-velopment of an adequate terminology by the gram for machine translation using a high capa-linguists and engineers of our group working in city store. The recent development of large-close cooperation. capacity, rapid-access storage systems permits At present, I am continuing research in all adopting a point of view different from that pre-categories of words which can be omitted from viously employed. It is no longer necessary tothe machine memory without any loss in the in- reduce the number of entries by dissection oftelligibility and accuracy of the output text. I stems and endings or by the use of ideoglossa-am also studying the problem of how to deal ries. In fact, the vocabulary can be expandedwith proper and geographical names, which are to include idiomatic sequences as well as singlealso members of the general vocabulary of a words.language but should be left untranslated. From the machine standpoint even a whole My research has been supported by two grants string of words which for reasons of source-from the Rockefeller Foundation. target semantics has to be handled as an entity While my research, though primarily based can be entered in the store and given an idiomaon German language material, took into consi- tic translation. Such strings of words are thederation the identical or analogous phenomena longest representatives of what we call seman-of a variety of language ...