Effect of Verb Subdivision and Noun Incorporation on Dependency Parsing

Parsing based on tree bank is a central issue of current natural language processing. The machine learning method of SVM and the dependency tree bank of HIT-IR-CDT is adopted in this work. In order to increase the parsing accuracy by linguistic means, verb subdivision and noun incorporation is done. The result shows, after verb subdivision, the accuracy of unlabeled attachment score increases from 79.05% to 79.5%, and the accuracy of labeled attachment score increases from 76.38% to 76.81%. Noun incorporation has little effect on the accuracy of dependency analysis but it can reduce training time efficiently.