Choosing which to use? A study of distributional models for nominal lexical semantic classification