Loading timeline…
20142010s
Neural Machine Translation (NMT)
End-to-end translation models replacing complex pipelined statistical translation methods.
Why It Was Important
Using massive sequence-to-sequence LSTMs (and later Transformers), engines like Google Translate fully swapped to NMT around 2016. Because the model learned the entire mapping from source to target sentence holistically in dense vector space, translation accuracy for complex phrasing skyrocketed immediately.
Who Invented It
Ilya Sutskever / Dzmitry Bahdanau
Engineers pioneering sequence-to-sequence capabilities.
Applications
- Google Translate Retrofit
- Real-time Voice Translation
- Multilingual parsing
Key Papers
- Sequence to Sequence Learning with Neural Networks
Ilya Sutskever, Oriol Vinyals, Quoc V. Le · NeurIPS 2014