Exploring the Necessity of Visual Modality in Multimodal Machine Translation using Authentic Datasets | Synapse