This thesis describes a system which generates Korean semantic networks from conceptual dependency representation. The conceptual dependency is a language-independent meaning representation mechanism and semantic network is a meaning representation mechanism more or less dependent on the syntactic structures of each language.
From a conceptual dependency representation, the system first selects a proper Korean verb through a discrimination network, a tree structure whose internal nodes represent branching decisions and terminal nodes represent Korean verbs. Then it generates a Korean semantic network by filling slots of the case frame for the verb.
In this thesis, we developed discrimination networks for two primitive action categories INGEST and MTRANS, case frames for the verbs in the discrimination networks, and slot filling methods for the verbs. Although the system was constructed and tested for only two primitive action categories, it was shown that the system is capable to catch delicate meanings hidden in conceptual dependency representation into Korean verbs. Based on this initial study, we may conclude that this approach is appropriate to use in Korean conceptual generation.
