A Comparative Study of Japanese and English Sublanguage Patterns
Virginia Teller, Michiko Kosaka, Ralph Grishman · 1988
As part of a project to develop a Japanese-English machine translation system for technical texts within a limited domain, we conducted a study to investigate the roles that sublanguage techniques (Harris, 1968) and operatorargument grammar (Harris, 1982) would play in the analysis and transfer stages of the system. The data consisted of fifty sentences from the Japanese and English versions of the FOCUS Query Language Primer, which were decomposed into elementary sentence patterns. A total of 187 pattern instances were found for Japanese and 191 for English. When the elements of these elementary sentences were classified and compared with their counterparts in the other language, we identified 43 word classes in Japanese and 43 corresponding English word classes. These word classes formed 32 sublanguage patterns in each language, 29 of which corresponded to patterns in the other language. This paper examines in detail these correspondences as well as the mismatches between sublanguage patterns in Japanese and English. The high level of agreement found between sublanguage categories and patterns in Japanese and English suggests that these categories and patterns can facilitate analysis and transfer. Moreover, the use of operator-argument grammar, which incorporates operator trees as an intermediate representation, substantially reduces the amount of structural transfer needed in the system. A pilot implementation is underway.