Machine translation for the next century

MT Summit VII Sept. 1999 Machine Translation for the Next Century Jun-ichi TSUJII Department of Information Science Faculty of Science University of Tokyo, Tokyo JAPAN Center for Computational Linguistics Department of Language Engineering UMIST, Manchester ENGLAND Abstract The panel intends to pick up some of the issues discussed in the Summit and dis- cuss them further in the final session from broader perspectives. Since the Summit has not even started yet, I will just enu- merate in this paper a list of possible per- spectives on MT that I hope are relevant to our discussion. 1 Nature of Translation and Limits of MT Technology Translation is a difficult enterprise. Even a human translator with the highest qualification cannot always provide perfect translation that satisfies both readers with different purposes and backgrounds, and the au- thors of original texts. Although translation seems essentially linguistic work and good command of the two languages seems sufficient, it actually requires the translator to under- stand source texts, which in turn requires her/him/it to have good knowledge of the subject field the source text is concerned with. More precisely, a good trans- lator has to be able to use knowledge of the subject field to infer pieces of information lacking in the source text but nonetheless necessary for composing the tar- get text. Furthermore, a good human translator should be a good author of the target language as well. This may imply that she/he/it has to take into considerations what the readers would like to know by reading the texts and what background knowledge they already have. In short, an ideal translator should be able to use diverse types of knowledge and information intelli- gently, not only knowledge of two languages but also that of the subject field of the text, and not only types of information explicit in the source texts but also those implicit. MT is one of the ultimate goals of Ar- tificial Intelligence. Since we have not yet succeeded in AI in any serious sense, MT will remain far away from being perfect. 2 Nature of Translation and Potentials of MT Technology Translation is a tedious work. If you have had a proper training as technical translator, translation of techni- cal manuals becomes a tedious work after translating several manuals of a similar kind. Although one has to acquire a set of skills and should be given a set of terminology of two languages in advance, you can produce fairly good translations without actually un- derstanding the text. Translation becomes a repetitive work, which only requires table look-up of fixed patterns and terminol- ogy. Such a work can be carried out far better by a computer system than a human translator. A MT system can produce much more consistent translations than a human translator. In translation of similar languages (eg. Korean and Japanese), the translation speed of human translation is limited only by the speed of typing. Although a translator has to stop from time to time and tries to understand, …

Content truncated to 500 words