Machine Translation of Sentences with Fixed Expressions

Machine Translation of Sentences with Fixed Expressions Naoto Katoh 1 Teruaki Aizawa NHK Science and Technical Research Laboratories 1-10-11, Kinuta, Setagaya-ku, Tokyo 157, Japan { katonao, aizaw a } @ strl.nhk, or.jp Abstract This paper presents a practical machine translation system based on sentence types for economic news stories. Conventional English-to-Japanese machine translation (MT) systems which are rule-based approaches, are diffi- cult to translate certain types of Associated Press (AP) wire service news stories, such as economics and sports, because these topics include many fixed expressions (such as compound words or collocations) which are difficult to be processed by conventional syntactic analysis and/or word selection methods. The proposed MT system, an economic-news stories machine translation system (ENTS), can translate eco- nomic news sentences with fixed expressions. The system consists of three processes, to handle different types of sentences, fixed type, economics-specific type and general type. This paper focuses mainly on the translation method for fixed-type sentences, which is a kind of example-based approach. In this translation method, fixed sentence trans- lation (STRA) data plays a key role. The STRA data is a set of bilingual templates, which is built automatically from fixed English sentences and their Japanese transla- tion equivalents. The fixed English sentences are extracted automatically from the AP corpus. A series of experiments to evaluate ENTS using eco- nomic news in the AP news stories showed the translation accuracy was about 50 % higher than with our conven- tional rule-based MT method. 1 Introduction We are developing an English-to-Japanese machine trans- lation (MT) system to produce real-time rough translations for Associated Press (AP) wire service news stories. With some news topics, troubles with fixed expressions lows the translation accuracy of the MT system. Economic news stories in particular are difficult to translate by con- ventional rule-based methods, because they contain many fixed expressions sharing two major characteristics: cl) The fixed expressions produce economics-specific syntactic structure. c2) Equivalents of the fixed expressions require Japa- nese economic jargons. These characteristics respectively cause two major bot- tlenecks for the conventional rule-based MT system: bl) General-purpose grammatical rules are not suffi- cient to yield correct analysis of economic news stories. (Simple addition of grammatical rules increases syntactic ambiguities.) b2) It is difficult to select the appropriate Japanese words for the translation. Actually, these problems reduce the translation accu- racy of our rule-based MT system to only 20%, which is too low for practical use. This paper presents a new English-to-Japanese MT sys- tem for economic news stories, which is called ENTS (Economic News stories machine Translation System), to process fixed expressions effectively. ENTS consists of three sequential processes (as shown in Fig. 1), based on the three basic types of economic news sentence. Process 1 is a kind of example-based approach, while Processes 2 and 3 are rule-based ones that differ in grammatical rules. This paper focuses mainly on Process 1, which is com- posed of fixed sentence translation, compound word trans- lation, fixed sentence translation data production and fixed sentence extraction, …

Content truncated to 500 words