Just last week, CDS Professor Kyunghyun Cho and an international group of his colleagues* released Nematus, an exciting toolkit for Neural Machine Translation (NMT).
Funded by the European Union’s Horizon 2020 research and innovation programme, Nematus performs neural machine translation using an encoder-decoder model, an approach that replaces traditional ‘phrase based’ translation and has become increasingly popular in the field.
The toolkit “has recently established itself as a new state-of-the-art in machine translation,” explains the team, and is already being used by the World Intellectual Property Organization (WIPO), a special branch of the United Nations, to train WIPO Translate, their machine translation system. WIPO can now perform highly accurate Chinese to English translations on their patent documents thanks to the high accuracy rate of Nematus, and plans to extend their reach into French language patent documents soon.
Implemented in Python, Nematus is open source and available for anyone to use as long as it has been cited. Check it out here.
*Alexandra Birch (University of Edinburgh), Marcin Junczys-Dowmunt (University of Edinburgh) , Orhan Firat (Middle East Technical University), Barry Haddow (University of Edinburgh), Julian Hitschler (Heidelberg University), Samuel Läubli (University of Zurich), Jozef Mokry (University of Edinburgh), Maria Nadejde (University of Edinburgh), Rico Sennrich (University of Edinburgh), and Antonio Valerio Miceli Barone (University of Edinburgh).
by Cherrie Kwok