In this thesis we propose a method that creates sense tagged data using machine readable dictionaries for ambiguous word sense tagging. Our method introduces a bootstrapping method to create training data additively and learn gradually. In this bootst...
In this thesis we propose a method that creates sense tagged data using machine readable dictionaries for ambiguous word sense tagging. Our method introduces a bootstrapping method to create training data additively and learn gradually. In this bootstrapping routine, this method applies penalty to a word that leads to wrong sense tagging. So it makes the right sense tagging in the next bootstrapping turn. The proposed method creates the meaning group that consists of distinguishing surrounding words near target words. This method introduces concentration concept into word weight formula for selecting important word and enhances the tagging accuracy. The tagging result on the SENSEVAL-2 data shows that the proposed method improves about 7% of accuracy over the base tagging system.