In this thesis, we present a design and implementation of Korean parser using the syntactic dependency relations. To achieve high efficiency, the parser uses dependency relations between word-phrases rather than morphems. To reduce invalid dependency relations generated in the process of parsing, we devised a fine grained category system for word-phrases, according to their functions and part of speech. Also, we propose four principles, on which particular dependencies can be detected. These principles are derived form some properties of Korean phrase sturcture. The empirical results show that these principles are effective to reduce invalid dependency relations when tested with sample sentences.
The experimental system has been implemented in "C" and "PROLOG" programming languages on the UNIX.