Customizing NMT Engines - Automated In-domain Adaptation

Track: TAUS | S3 |
Wednesday, April 4, 2018, 2:45pm – 3:30pm
Held in: S
Gábor Bessenyei - Globalese 
Riki Izawa - Kawamura International
Host: Anne-Maj van der Meer

One of the challenges in neural machine translation (NMT) is its lack of customizability — the ability to configure the engines with glossary and style guides. Globalese 3.1, an NMT system developed by Morphologic Localisation, introduces the new feature augmented in-domain adaptation where a user can use a smaller amount of core corpora and larger auxiliary corpora to build an NMT engine which is trained for a specific domain. In this presentation, we will explain this new feature and present some case studies in Asian and European languages.

Takeaways: Attendees will learn the latest development in research on the customization of NMT engines and tips on building an effective domain-specific NMT engine.