Machine Translation from Japanese into English zyxw MAKOTO NAGAO, JUN-ICHI TSUJII, AND JUN-ICHI NAKAMURA Invited Paper zyxwvutsrqpo This paper describes the outline of our lapanese to English machine translation system, which is supported by the Agency of Science and Technology of the Japanese Government. Many new methodologies are introduced to obtain high-quality translation results. The analysis is based on case grammar, which is suitable for a word-order-free language such as Japanese. The dictionary is rich enough to handle many specific expressions. It contains not only case frame information, but also semantic information, idiomatic expressions, and many others. In the transfer phase, the system applies many structural transformations, so that the structural dif- ference ofthe same contents in lapanese and English can be relieved. In the generation phase, many structural transformations are again applied so that the ellipsis problems can be avoided, and that better stylistic expressions can be obtained. The system is running mainly for the abstracts of scientific and technical papers. The evaluation method of the translated results is also discussed, with many example translat;ons. zyxwvutsrqpon I. INTRODUCTION There is a theory of language translation which was advocated by Y. Nida, that the translation is a mapping of the sentential meaning in one language to another. That is, the utterance must be interpreted not only in the context of utterance and situation, but also by the culture of the language and the society. Therefore, the surface sentential structuremaynot necessarilybepreserved, butthetrans- lated (or created) sentence must give the same mental effects to the receivers of the translated language as to the receivers of the source language. That would be a wonder- fultranslation. This theory,however, presupposes the ex- istenceofthe same culturalbackgroundindifferentlan- guage societies. If such does not exist, in country pairs such as Japan and the United States, we have to have some other translation theories. Many western literary works translated into Japanese may not stir the Japanese readers with the same mental effects as those affecting the US. readers, but the fundamental appreciation of that literature by the Japanese readers will not be hampered by such incomplete translation. We can Manuscript received August 27, 1985; revised January 27, 1986. Theauthors are with the Department of Electrical Engineering, Kyoto University, Sakyo, Kyoto 606, Japan. accept literary translation and can imagine that the way of thinking of western people in such and such. Foreign con- cepts and customs in different countries can be appreciated through imperfect translation. Language has theabilityof explaining totally complex concepts foreign to the reader so that he can understand them as precisely as he wishes. We can get pseudo-experience of foreign culture through lan- guage, which basically must be obtained byreal experience. There are manykindsof research on language compre- hension and some trials to apply it to machine translation, even though there is no precise definition and consensus aboutwhat language understanding by computer means. We think this kind of language understanding framework is quite doubtful and inapplicable to a practical machine translation system because we have no way at presentof memorizing in the computer the cultural background of different languages and nations which are essential by the Nida’s translation theory. We think that the standpoint of literal translation is inevitable between languages of differ- ent families. The language understanding ability of a human being will cover the gap of incomplete translation. When thedocumentsto be translated are limitedto a specific domain of science and technology, there is no difference of cultural background between the languages. The problems can be purely limited to the syntactic, semantic, and con- textual information plus culture-independent, domain- specific knowledge. Within this scope we can escape the seriousproblemswhichNidapointedoutforthe transla- tion. The machine translation system we are developing is limited to this scopeofthe same knowledge domain be- tween languages. The levels of translation may be classified into the follow- ing four: 1) Free andcreativetranslation which aims at the same mentalreactionsbythe readers of source and target lan- guages. 2) Sentence-by-sentence translation in the free sentential style. Language particularities are fully considered in trans- lation. 3) Literal translation. The sentential structures of the source language remain strongly in the target language, PROCEEDINGS OF THE IEEE, VOL 74, NO zyxwvutsrqponml 7. JULY 1986 0018-9219/86/0700-0993$01.00 01986 IEEE 993