Korean Word Sense Disambiguation using Parallel Corpus as Additional Resource

Abstract

Most previous research on Korean Word Sense Disambiguation (WSD) were focusing on unsupervised corpus-based or knowledge-based approach because they suffered from lack of sense-tagged Korean corpora. Recently, along with great effort of constructing sense-tagged Korean corpus by government and researchers, finding appropriate features for supervised learning approach and improving its prediction accuracy became an issue. To achieve higher word-sense prediction accuracy, this paper aimed to find most appropriate features for Korean WSD based on Conditional Random Field (CRF) approach. Also, we utilized Korean-Japanese parallel corpus to enlarge size of sense-tagged corpus, and improved prediction accuracy with it. Experimental result reveals that our method can achieve 95.67\% of prediction accuracy.