Controlled Korean for Korean-English MT

Controlled Korean for Korean-English MT Munpyo Hong a, Chang-Hyun Kimb a Department of German Lang. & Lit. Sungkyunkwan University 53 Myeongnyun-dong Seoul, Korea skkhmp@skku.edu b Natural Language Processing Team ETRI Gajeong-dong 16 Taejeon, Korea chkim@etri.re.kr Abstract. This paper addresses the issues in designing the so-called ‘Controlled Korean’ for a Korean-English MT system. Controlled Language is a sublanguage of a natural language which is supposed to improve the readability and the translatability of a text. Much effort has been made to design a controlled language for major international languages such as English, German, Spanish and etc. However, little effort has been made yet to design a Controlled Korean in the context of machine translation. In this paper we introduce the concept of the Controlled Korean we have developed for a Korean-English MT system and compare the Controlled Korean with the Controlled English and Controlled German from the perspective of the translatability. The result of our experiments shows that in designing a Controlled Language, not only the linguistic characteristics of the language but also the characteristics of an MT-System must be taken into account. Keywords: Controlled Language, Controlled Korean, Machine Translation, Translatability, Negative Translatability Indicators 1. Introduction Controlled Language is a sublanguage of a natural language designed to improve the readability and the translatability of a text. Originally, the concept of a controlled language was introduced in the field of technical documentations to prevent the misunderstanding of texts. 1 In the last couple of years the necessity of a controlled language has increased significantly, especially in the technical documentation domain, as the number of pages to be translated has increased enormously. The increase of the volume of the documents to be translated made it necessary to employ a full automatic translation system like an MT or a machine-aided translation system like translation memory. In Korea, however, MT systems have not been widely welcomed by the experts in the localization business, because the quality of the translations failed to match their expectations. The recent researches on the controlled languages show that the use of a controlled language can generally lead to the improvement of machine translation quality, thus reducing the cost of 1 cf. Esselink(2000), Göpferich(1995) 391 22nd Pacific Asia Conference on Language, Information and Computation, pages 391–396 post-editing. The general assumption behind the idea of the controlled language is that the cost of the use of a controlled language that can be paraphrased as ‘pre-editing’ is lower than that of ‘post-editing’ in using an MT system for localization. Previous works on the controlled language in the context of MT have mainly focused on the impact of the controlled language on MT. Aikawa et al. (2007), for example, addressed the impact of the controlled English for MSR-MT. Lehrndorfer (1996), Lehrndorfer&Schachtl (1998) dealt with the controlled German for German-English MT. In this paper, we will introduce the concept of the ‘Controlled Korean’ for Korean-English MT. We will not only show the impact of the Controlled Korean on Korean-English MT, but also share our experience …

Content truncated to 500 words