This thesis describes an example-based transfer method in a Korean-English machine translation system. The basic idea of example-based transfer is that translations is a matter of finding similar examples, remembering how a particular source language expression has been translated before. An example-based transfer utilizes translation examples as transfer rules. Main problems of example-based transfer are to construct a database of translated examples, which have information for transfer, and to retrieve similar examples from the database. Previous models can not be well applicable to the mapping of features, which are important in translation, like modal, tense, and so on.
This thesis propose a new distance function and three types of mapping. The distance function calculates similarity scores between the examples and the input sentence to select the best matching examples from the database. The three types of mapping are lexical, feature and structural mapping and are combined in order to express the manifold correspondences between examples of both languages. Using this proposed method, we implement Korean-English machine translation system with Prolog.