Beyond Triplet: Leveraging the Most Data for Multimodal Machine Translation