Representation Learning for Natural Language Processing