In this paper, we present two deterministic Korean dependency parsing algorithms using support vector machine. We modify the forward algorithm proposed first by Kudo(2002) and propose the new backward algorithm. It's well-known fact that backward pars...
In this paper, we present two deterministic Korean dependency parsing algorithms using support vector machine. We modify the forward algorithm proposed first by Kudo(2002) and propose the new backward algorithm. It's well-known fact that backward parsing is more effective in head-final languages. We use proper features to train parsing models for Korean that has a complex word form (eojeol in Korean), an ellipsis possibility of constituents, and so on. In the experiment for Korean Language Information Base System (KIBS) corpus, we achieve 88.25% dependency accuracy that is state-of-the-art performance. It means that proposed methods outperform previous statistical parsing methods and deterministic parsing algorithm is also suitable to Korean parsing.