Improvement in Customizability Using Translation Templates

Machine Translation IMPROVEMENT IN CUSTOMIZABII,ITY USING TRANSI~ATION TEMPLATES Satoshi KINOSHITA, Akira KUMANO, ltideki IllRAKAWA R & D Center, Toshiba Corporation. Abstract This paper outlines customization of a machine translation system using translation templates, which enable usm~ to represent the bilingual knowledge needed for complex translation. To evaluate their effectiveness, we analyzed a bilingual text to estimate tire improvement in eustomizability. The result shows that about 60% of mistranslated sentences can be translated as nrodel translations by combining the proposed fra,nework with the conventional customizing functions. 1. Introduction The ability of natural language processing (NI.P) systems is limited due to the knowledge they have, not their framework. This is reflected by recent intensive research on acquisition of linguistic knowledge from a corpus[2][6][91. Machine translation (MT) systems are no exception. Conrpared with monolingual knowledge, knowledge needed for translation is difficult to collect. Knowledge acquisition from a bilingual or parallel corpus is considered to be a promising way to reduce tile painstaking task[l ][10]. Without customization, no general-purpose MT system can output satisfactory translations; therefore it is essential to tune the system by developing a user- oriented lexicon or by registering al)propriate target words. The kind of customization needed depends on how the system is used. If a user translates a document to skim it, he can judge the ability of his MT system by semantic invariance: what percentage of the content of the source text is preserved in its t,anslation. If, on tile other hand, he requires translation o1" publication qtmlity, semantically correct tnmslation is not sufficient; that is, translations should be wcllq'ormed so as to conform to a documcntational style• To minimize post-editing, more elaborate customizing functions than in the former ease are required. In this paper, we will describe a customizing framework which uses 'translation templates.' This enables users to represent bilingual knowledge for complex translation where a drastic change in lingt, istic structures occu,s to generate natural translations. Then we will discuss the effectiveness of this framework by comparing it witl~ the practically used customizing functions based on the analysis of a bilingual text• 2. Machine Translation Using Translation Templates 2.1 Aim of Translation Templates If a user wants publication-quality translations, stylistic wcll-fonnedness is as important as semantic invariance. Consider translating the Japanese sentence (1). Although its Iranslation (2), which is the result of our current MT system, is correct, (3) sounds more natural than (2); in (.3), the verb phrase "using these detectors" is nominalized to function as a subject to represent tile cause of the 'reduce' event. If tire user prefelx (3) to (2) as a translation of (1), (2) needs to be post-edited. (1) korera-no kenshutsuki-wo tsukau kotoniyori these detectors-OP;J use by kakaku-ga teigen-shita price-SUBJ reduce-PAST (2) The price dropped by using these deteetm,'s. (3) Use of these detectors reduced the price. As the above example illustrates, when source and target languages have a significant difference in their linguistic features, linguistic structures of source sentences are drastically changed to generate natural translations. In this paper, we …

Content truncated to 500 words