SYSTRAN Purely Neural MT Engines for WMT2017
This work addresses machine translation quality for English-German news translation, but it is incremental as it applies existing methods like OpenNMT and adaptation techniques to a specific benchmark.
The paper describes SYSTRAN's neural machine translation systems for the WMT 2017 English-German task, built using OpenNMT with LSTM encoder/decoders and attention, and enhanced through back-translated monolingual data and hyper-specialization adaptation, achieving competitive results in the shared evaluation.
This paper describes SYSTRAN's systems submitted to the WMT 2017 shared news translation task for English-German, in both translation directions. Our systems are built using OpenNMT, an open-source neural machine translation system, implementing sequence-to-sequence models with LSTM encoder/decoders and attention. We experimented using monolingual data automatically back-translated. Our resulting models are further hyper-specialised with an adaptation technique that finely tunes models according to the evaluation test sentences.