Research On The Japanese Mongolian Machine Trans-Lation System And The Related Technology

In this paper, we study and implement a Japanese-Mongolian machine translation system(MTS) in light of theory and methods of the previous MTS research and the similar structures between Japanese and Mongolian languages. In the aspect of the system resources, we construct automatically a corpus which covers800reports on agriculture, forestry, animal husbandry, water conservancy from Japanese Every Day Newspaper in2002. Based on this corpus, we also compiled a Japanese-Mongolian stem dictionary. In this paper, we introduced a new Japanese grammar, that is, the derivational grammar and developed a Japanese syntactic parser based on the derivational grammar. In order to improve our MTS, we do some study and experiments as follows:(1) In order to make use of the similar features between the derivational grammar and Mongolian grammar, we change the results of Japanese morphological analyses based on traditional Japanese grammar to the additional components defined in the derivational grammar. Hence, we developed an analyzer based on the derivational grammar.(2) put forth the method of changing Japanese clause order based on Mongolian word order and solve the problems in verb clause translation. At first, we wrote word order rules, then, according to the rules, changed the orders and got the natural clauses of Mongolian language. At the same time, we also solve the problems of(?) nouns and (?) verbs. (3) We put forth some view point on the semantic analyses of Japanese function words and its Mongolian translation, as a result, solved the problems of function words translation.(4) We developed a Japanese syntactic parser based on the derivational grammar. We wrote some syntactic rules and the difficult problems in Japanese syntactic parser KNP, for example, the related problems of brackets, comma, full stop.Based on the above research works, we implement a Japanese-Mongolian translation system and did some evaluations. We translated the30newspaper reports and evaluated187corresponding translated sentences (the average words is27.1). As the result, the results of translations for the150sentences are correct, which means the accuracy is80.2%. In the evaluation of syntactic parsing, we analyzed1519sentences in the same310reports. There is not any wrong translation in the result.