From To K/E: a Korean-English machine translation system based on idiom recognition and fail softening

MT Summit VII Sept. 1999 FromTo K/E: A Korean-English Machine Translation System based on Idiom Recognition and Fail Softening Byong-Rae Ryu, Youngkil Kim, Sanghwa Yuh & Sangkyu Park Knowledge Information Department Computer • Software Technology Laboratory Electronics and Telecommunications Research Institute (ETRI) 161 Kajong-Dong, Yusong-Gu, Taejon, 305-350, South Korea {ryu,skpark}@computer.etri.re.kr {kimyk,shyuh}@etri.re.kr Abstract In this paper we describe and experimen- tally evaluate FromTo K/E, a rule-based Korean-English machine translation sys- tem adapting transfer methodology. In ac- cordance with the view that a successful Korean-English machine translation sys- tem presumes a highly efficient robust Korean parser, we develop a parser re- inforced with "Fail Softening", i.e. the long sentence segmentation and the re- covery of failed parse trees. To over- come the language-typological differences between Korean and English, we adopt a powerful module for processing Korean multi-word lexemes and Korean idiomatic expressions. Prior to parsing Korean sen- tences, furthermore, we try to resolve the ambiguity of words with unknown gram- matical functions on the basis of the collo- cation and subcategorization information. The results of the experimental evaluation show that the degree of understandability for sample 2000 sentences amounts to 2.67, indicating that the meaning of the trans- lated English sentences is almost clear to users, but the sentences still include minor grammatical or stylistic errors up to max. 30% of the whole words. 1 Introduction We present in this paper FromTo K/E, a rule-based Korean-English machine translation system adapting transfer methodology, which has been developed by Electronics and Telecommunications Research Insti- tute (ETRI) [10]. The project for developing the Korean- English machine translation system, FromTo K/E, was carried out 1996-1998 by Computer • Software Tech- nology Laboratory of ETRI, former Systems Engineer- ing Research Institute (SERI), partly in collaboration with Seoul National University (SNU) and Korea Re- search & Development Information Center (KORDIC) [11]. The project was completed in 1998. The ini- tial purpose of this system is to provide Korean infor- mation to the foreigners who do not understand the Korean language [5]. FromTo K/E is a rule-based machine translation system adapting the transfer methodology, into which recent illuminating approaches to the various tricky ambiguity problems in MT have been integrated. They particularly aimed for minimizing the parsing com- plexity for Korean parser. Three major refinements can be pointed out: The first one is HMM-based Ko- rean tagger, which resolves the morphological ambigu- ity generated by a morphological analyzer. The sec- ond one is Korean dependency parser with idiomatic expression recognizer to reduce the syntactic ambigu- ities and to produce natural English translation. The last one is the resolver of the correct case roles and grammatical functions of Korean sentences based on the collocation. Collocation-based Korean-to-English transfer selects the best English target words for am- biguous Korean words. Those collocations are ex- tracted from the 400,000 word Korean corpus. The results of the experimental evaluation show that the degree of understandability for sample 2000 sentences amounts to 2.67, indicating that the mean- ing of the translated English sentences is by …

Content truncated to 500 words