Automatic extraction of subcorpora based on subcategorization frames from a part-of-speech tagged corpus