ALT-J/M a prototype Japanese-to-Malay translation system

MT Summit VII Sept. 1999 ALT-J/M A prototype Japanese-to-Malay Translation System Kentaro Ogura, Francis Bond and Yoshifumi Ooyama NTT Communication Science Laboratories, Kyoto, 619-0237, JAPAN {ogura,bond,ooyama}@cslab.kecl.ntt.co.jp Abstract In this report we introduce ALT-J/M — a prototype Japanese-to-Malay translation system. The system is a semantic transfer based system that uses the same transla- tion engine as ALT-J/E, a Japanese-to- English system. 1 Introduction In this report we introduce ALT-J/M — a prototype Japanese-to-Malay translation system. In Malaysia, there has been a lot of work on Malay- English machine translation systems, culminating in Zaharin et al.'s (1994) user-driven machine translation system SISTEP/JEMAH. This is a fully functional translators workbench, with the ability to perform on- line dictionary look-up, glossing and automatic trans- lation of phrases and clauses. It is, however, only available for Malay and English. In Japan, there has been much work on Japanese- English machine translation, but, none on Japanese- Malay so far. At the NTT Communication Science laboratories, we have been developing ALT-J/E, the Automatic Language Translator — Japanese to En- glish since 1987 (Ikehara et al. 1987; Ikehara et al. 1991). The aim is to produce a high quality machine translation system that can be used to facilitate com- munication via machine translation. We began work on a Japanese-to-Malay system for two reasons: (1) Malay is from a different language family to both Japanese and English, so it is an in- teresting test of our translation engine; (2) There is increasing trade between Japan and Malaysia so the need for such a system has been increasing. Both Malay and Japanese are the official language of over 100 million speakers (Crystal 1987:287). This report is organized as follows. In the next section, we describe the ALT-J/M system's archi- tecture. We then look at the lexicon, the backbone of any MT system, in Section 3. In Section 4 we look at some examples of how the system translates. We then look at future issues in Section 5. 2 Architecture We use the multi-level translation method in our MT system (ALT-J/M), as shown in Figure 1. The in- put is separated into tense and modal information (the 'subjective' part) and the kernel sentence (the 'objec- tive' part). The objective part is translated using the multi-level transfer method: wide ranging rules, such as direct parse tree transfer are applied first, followed by idiomatic expressions then patterns from the se- mantic valency dictionary, and finally the default gen- eral patterns. This allows transfers of varying granu- larity. The process of translation can be divided into seven parts. First, ALT-J/M splits the Japanese text into morphemes. Second, it analyses the sentence syn- tactically, often giving multiple possible interpreta- tions. Next, it rewrites complicated Japanese expres- sions into more easily translated ones. Fourth, ALT- J/M semantically evaluates the various interpreta- tions. Fifth, syntactic and semantic criteria are used to select the best interpretation. Sixth, this interpre- tation is used as input to generate Malay. Finally, the Malay sentence is adjusted to give the correct inflec- …

Content truncated to 500 words