ModernMT

ModernMT is an adaptive neural machine translation engine developed by Translated with the Fondazione Bruno Kessler, the University of Edinburgh, and TAUS, funded by the European Union. Its defining trait is real-time learning: instead of translating from a static model, it pairs a generic background model with a foreground model that updates the moment a translator corrects a segment, so suggestions adapt to the user's terminology, style, and domain without any retraining cycle or setup. It also uses the context of the whole document rather than translating sentences in isolation.

The engine is designed as a translator's assistant rather than a replacement: it integrates with the open-source CAT tool Matecat (where its corrections feed the adaptive model) and with Crowdin, and it includes glossary controls so brands can lock in preferred terminology and override model suggestions. Translated's T-LM service pairs ModernMT with GPT-4 to bring generative AI performance to 200 languages at English-equivalent token cost (provider-described). Access typically runs through Translated's services, Matecat, or its API; details are on modernmt.com.

1. Enable the ModernMT plugin in Matecat or Crowdin for adaptive suggestions.
2. Point it at your existing translation memories so it adapts instantly.
3. Correct segments as you work; the model updates in real time.
4. Add glossary files to lock in brand terminology.