Improvement of translation quality of English newspaper headlines by automatic preediting

MT Summit VII Sept. 1999 Improvement of Translation Quality of English Newspaper Headlines by Automatic Preediting Takehiko Yoshimi Ichiko Sata Software Business Development Center Software Business Development Center SHARP Corporation SHARP Corporation Abstract Since the headlines of English news arti- cles have a characteristic style, different from the styles which prevail in ordinary sentences, it is difficult for MT systems to generate high quality translations for headlines. We try to solve this problem by adding to an existing system a preedit- ing module which rewrites the headlines to ordinary expressions. Rewriting of head- lines makes it possible to generate better translations which would not otherwise be generated, with little or no changes to the existing parts of the system. Focusing on the absence of a form of the verb of 'be', we have described rewriting rules for putting properly the verb 'be' into the headlines. 1 Introduction As Japanese people have more opportunities to see English newspaper articles through WWW, it be- comes more important to translate correctly them into Japanese. The most essential information of articles is expressed by the headlines. They have character- istic styles different from those of ordinary sentences, so as to convey much information with as concise ex- pression as possible. The characteristic styles prevent a syntactic parser from generating appropriate parse trees for the headlines, thus decrease the quality of translations. There are at least two possible solutions for this kind of problem. One is to extend syntactic rules so that the parser may analyse characteristic expressions. The other is to add to the existing MT system a preed- iting module which rewrites characteristic expressions to ordinary ones. Possible problems of the former ap- proach include the difficulty of keeping the consistency and the portability of the syntactic rules. The latter approach (Shirai et al., 1993; Kim and Ehara, 1994) is preferable from the viewpoint of system design and maintenance. Adopting the latter approach, we propose a method of improving the quality of translations for the headlines. In this paper we focus on a conspicuous phenomenon in the headlines, the absence of an appro- priate form of the verb ‘be’, and formulate rewriting rules for inserting the verb ‘be’ in its proper place in the headlines, based on information obtained by mor- pholexical and rough syntactic analysis1. Rewriting of the headlines makes it possible to generate better translations, with little or no changes of the exist- ing parts of the systems. While most systems would not probably accept, for example, the headline “Sales up sharply in June”, they may be able to generate a satisfactory translation of the expression “Sales are up sharply in June” where “are” has been inserted. We have incorporated the proposed method into our English-to-Japanese MT system Power E/J, and car- ried out an experiment with 312 headlines as unknown data. Our method has satisfactorily marked 81.2% re- call and 92.0% precision. 2 Preediting Module 2.1 Framework of Automatic Preediting Figure 1 shows the flow of analysis in our experimental …

Content truncated to 500 words