This thesis describes a design and implementation of transfer phase for an English-Korean machine translation system with transfer approach. The most important roles of the transfer phase are structure transformation and selection of translation equivalents. This thesis explains the details of these two roles, and we implement the transfer system to fulfil these roles.
For two languages belonging to different language family such as English and Korean, the transfer system must adjust the differences between two languages so that it generates translated sentences of good quality. Therefore we construct a transfer system with three phases-FET(Front-End Transfer), MAT(MAin Transfer), and RET(Rear-End Transfer). MAT transfers an English dependency structure to a corresponding Korean dependency structure by selection of translation equivalents. Both FET and RET adjust the differences of language pairs by transforming the dependency structures.
In general, an English word is corresponding to multiful Korean words, so it is important to select correct one among them. For this purpose, we propose a filter mechanism in order to select correct one. Currently, we have designed six filters-DR(Domain Restriction), CG(Case Government), DC(Deep Case), SM(Semantic Marker), CO(COllocation), and SY(SYnatactic) filter. The transfer system selects a correct translation equivalent by using combination of appropriate filters.
These three phase transfer system and filter mechanism are so flexible that they allow incremental improvements. Because linguistic contents in machine translation is incomplete, it is very useful and important to improve them incrementally.