Comparing Trajectory and Vision Modalities for Verb Representation