Leveraging Neural Caption Translation with Visually Grounded Paraphrase Augmentation