Visual Representation of Natural Language Scene Descriptions