Machine Translation

Machine Translation is the task of automatically translating text or speech from one language to another.

Early systems were rule-based, followed by statistical approaches that learned translation probabilities from large parallel corpora. Neural machine translation took over in the mid-2010s, first with RNN Encoder-Decoder models, then with the Attention Mechanism introduced in Neural Machine Translation by Jointly Learning to Align and Translate (Sep 2014), and then with the Transformer from Attention Is All You Need.

Translation quality is commonly measured with the BLEU Score. Backtranslation is a common trick for generating extra training data.