Data used during training for a standard MT system (left), and for an unsupervised MT system (right). The former is a large collection of translations, while the latter a (much larger) collection of unrelated documents in each language.
![datasets.001 (1)](https://engineering.fb.com/wp-content/uploads/2018/08/datasets.001-1.jpeg)
Data used during training for a standard MT system (left), and for an unsupervised MT system (right). The former is a large collection of translations, while the latter a (much larger) collection of unrelated documents in each language.