Machine translation system PENSÉE: system design and implementation

MT Summit VII Sept. 1999 Machine Translation System PENSÉE: System Design and Implementation Sayori Shimohata Toshiki Murata Atsushi Ikeno Tsuyoshi Fukui Hideki Yamamoto Oki Electric Industry Co., Ltd. Research & Development Group, Kansai Laboratories 1-2-27 Shiromi, Chuo-ku, Osaka 540-6025 Japan {shimohata245,murata656,ikeno546,fukui556,yamamoto436}@oki.co.jp Abstract This paper describes a new version of our machine translation system PENSÉE. In the light of its past systems, new PENSÉE is designed to improve portability from de- velopers' point of view and translation qual- ity from users' point of view. The features of new PENSÉE are: 1) Java implemen- tation and 2) pattern-based transfer ap- proach. In addition, new PENSÉE places a great importance on user interface espe- cially in building user dictionaries. We will discuss why and how we resolve the exist- ing MT problems and present dictionary building tools to support user customiza- tion. 1 Introduction We have been researching and developing a se- ries of machine translation (MT) systems “'PENSÉE”. PENSÉE appeared on the market first as a Japanese to English MT system in 1986 and then as an English to Japanese one in 1988. In the early stage. PEN- SÉE was a rule-based transfer MT system running on a workstation. With the improvement in the com- puter performance and the increase of personal com- puter(PC) users in 1990's. PENSÉE has changed its platform from a workstation to a PC. The number of MT users is growing year by year. And accordingly, a more comfortable user environ- ment and a higher-quality translation will be required. To meet the demand of those users, we are launching a new version of PENSÉE which features 1) Java imple- mentation and 2) pattern-based approach. Java im- plementation ensures portability and security on net- work computing. Pattern-based approach improves the quality of translation and makes customization easier. In addition, new PENSEE places a great im- portance on user interface especially in building user dictionaries. In this paper, we will describe the fea- tures of new PENSEE and present dictionary building tools to support user customization. 2 Java Implementation Figure 1 illustrates an overview of new PENSÉE (we call it “PENSÉE”' here after). PENSÉE is written entirely on Java platform. Java is an object-oriented programming language developed by Sun Microsys- tems. Inc [Gosling et al. 96]. Under the slogan “Write once, run anywhere”, Java has been designed for use on networks from the beginning. The application writ- ten in Java works on any kind of compatible devices that support the Java platform. This Java platform's ideal fits the trend of the com- puter world. Under the circumstances that users can choose and change their platforms to their tastes, the applications are expected to work on as many plat- forms as possible. Java enables us to develop PEN- SÉE independent of platforms and saves the time and expense of multiple ports. We are developing PEN- SÉE on a network computing environment and confir- m that PENSÉE will run on the different platforms, such as Unix. Windows, MacOS, and Java OS, with- …

Content truncated to 500 words