This paper describes the transformation of Conceptual Structure into Korean Semantic Structure in the machine translation under PIVOT approach. The Conceptual Structure as PIVOT is language-independent, but Korean Semantic Structure contains some stylistic information peculiar to Korean. This transformation is mainly due to two problems.
The one is Stylistic differences on sentence structure between languages. The characteristics of Korean style are checked by use of Cross Translation Test, whose examples are BE-type sentence, Avoidance of inanimate actor and Nominal predicator, and so on Some of these style conversion cause the transformation of Structure or change of semantic information.
And the other problem is that surface realization of Deep Cases using only functional words causes ambiguity of meaning sometimes. So the deep case is replaced with the corresponding content word with same meaning.
Consequently, the use of stylistic information and replacement of some deep cases guarantee more natural generation of Korean sentence.
