Effectiveness of Automatic Expansion of Training Data for Japanese Word Sense Disambiguation

Abstract

In this paper, we propose a method to expand training data automatically, using example sentences of a dictionary, other sensebank and unlabelled corpus. We tested on the data of SemEval-2010 Japanese WSD task and achieved 79.5% accuracy in our experiments. Then, we limited the number of used training data and achieved 80.0% accuracy.