한국어 격조사 복원에 대한 실험적 고찰

Abstract

This thesis is an experimental study on case particle restoration in Korean. The case particles in Korean sentences are omitted frequently. The omitted particles cause ambiguity in syntactic attachment and decrease performance of syntactic analysis. In this thesis, we restore the omitted case particles using machine learning techniques and suggest the most proper features for case particle restoration. The system for restoring omitted particles can be one component in the parsing system and also can be used for indexing terms in information retrieval. We have done experiments on several experimental settings and have observed the results. For the experiments, we have used ETRI syntactic tree-tagged corpus. The correct restoration rate of the system is 81.11 in accuracy of omitted case particles. We have observed that nouns and verbs, themselves, are very important features for restoring case particles.