A Study of Performance Evaluation for GA-ILMT Using Travel English

A STUDY OF PERFORMANCE EVALUATION FOR GA-ILMT USING TRAVEL ENGLISH Hiroshi Echizen-yat,Kenji Arakitt, Y oshio Momouchitand Koji Tochinaitt t Department of Electronics and Information Engineering, Hokkai-Gakuen University S 26-Jo, W 11-Chome, Chuo-ku, Sapporo, 064-0926 Japan Email:fechi,momouchil@eli.hokkai-s-u.ac.jp ttDivision of Electronics and Information Engineering, Hokkaido University N 13-Jo, W 8-Chome, Kita-ku, Sapporo, 060-8628 Japan Email: { araki,to chinai} @media. eng .hokudai. ac. jp Abstract Recently, many machine translation systems have been developed. However, for translation of conversation, correct translation rates and quality of translation are particularly low. This is due to machine translation systems not being able to generate translation results which fit the context of the conversation . We previously proposed a method of Machine Translation Using Inductive Learning with Genetic Algorithms(GA-ILMT). We compare this system's results to two others that use rule-based translation method, and evaluate the results of experiments done with GA-ILMT, measuring it's performance when applied to travel English. As a result of the evaluation experiments, we confirmed that GA-ILMT can generate translation results which are more appropriate to the context of the conversation. 1 INTRODUCTION Recently, the internet has come to be used by many people around the world. As a result, it is necessary to communicate correctly and quickly, information that is expressed in different languages. Machine translation is a very effective method of responding this need. Therefore, substantial research has been carried out in the machine translation field, and many machine translation systems have been developed. However, their correct translation rates and quality of translation are not sufficient to date. In the translation of conversation, the correct translation rate and the quality of translation are particularly low. This is due to machine translation systems not being able to generate translation results which fit the context of the conversation. Machine translation systems must be able to translate well to generate translation results which will also fit the context of the conversation taking place.. To date, machine translation systems have been rule-based machine translation[1]. However, this method has several problems. Rule-based machine translation cannot deal adequately with various linguistic phenomena due to its use of limited rules. Moreover, it also has difficulty in dealing with words that it does not recognize due to the use of it's non-dynamic dictionary. It is difficult for the rule-based machine translation method to translate conversational sentences which have a certain context. To resolve these problems, an example-based machine translation method is being researched [2, 3, 4]. This method can improve the correct translation rate and the quality of translation by incorporating any new translation examples that it is given. It becomes possible to translate conversational sentences through having many translation examples. However, serious obstacles remain before the realization of a practical translation system, as the example-based machine translation method requires both a large base of translation examples and a lot of CPU time. As a result, it is difficult to produce a machine translation system based on this method. -285- Source Sentence Translation Process Translation Result� (Proofreading) - Proofread Translation Result …

Content truncated to 500 words