Is Word Segmentation Necessary for Deep Learning of Chinese Representations?