On the successful language model training for one of our clients
We’re happy to share a recent excellent example of engine customization with you: the preparation of a machine translation language model for one of our key clients. During preparation for the training, we gathered a large high-quality translation base for this client, which undoubtedly played a significant role during the training process.
The evaluation of the post-training test translation has shown that the engine handled the translation of the client’s marketing materials 10% better than other popular stock engines (based on hLEPOR metrics that determine the number of identically translated segments in the machine and the etalon (translated by a human and well-edited) translations):
- Google – 51 %
- Amazon – 51 %
- ModernMT – 49 %
- Trained ModernMT model – 60 %
Such a high-quality metric of the machine translation in this topic is truly successful for a new trained model.
We’re happy to note that many production departments of Janus Worldwide have been working with our MT team R&D for quite a while now, which helps us earn the trust of our clients who have just started their journey into the machine translation area. Every new client earned is a big victory over prejudice for us. These victories grow in value each time we manage to train a language model from scratch and get machine translation of great quality.
Leave a Reply
Want to join the discussion?Feel free to contribute!