NYU Deep Learning SP21 - 12L – Low resource machine translation

Course website: http://bit.ly/DLSP21-web Playlist: http://bit.ly/DLSP21-YouTube Speaker: Marc'Aurelio Ranzato Slides: https://bit.ly/DLSP21-12L Chapters 00:00 – Welcome to class 01:56 – Machine translation 11:08 – Beam search 19:40 – How alignment works 24:56 – Translation with uncertainty 29:13 – Evaluation 31:45 – The long tail of languages 35:10 – Study case: Nepali ↔ English translation 43:45 – Low resource machine translation 49:35 – FLoRes evaluation benchmark and process 56:35 – ML perspective 58:13 – Supervised learning 1:05:15 – Self-supervised learning (DAE) 1:07:35 – Semi-supervised learning (ST) 1:15:23 – Semi-supervised learning (BT) 1:19:03 – Semi-supervised learning (ST + BT) 1:21:53 – Multi-task/-modal learning 1:24:01 – Domain adaptation 1:28:50 – Unsupervised MT 1:39:17 – FLoRes Ne-En 1:47:02 – Source target domain mismatch 1:56:39 – Final remarks