Based on neural networks
NMT is the latest and most effective translation technology. Using artificial neural networks to automatically translate and localize text, NMT is able to model entire sentences in a single procedure.
In the same way a human brain pieces together clues to overall meaning as it absorbs a piece of text one word at a time, NMT is able to recognise the context of individual words and phrases when determining the most accurate translated version.
Our own research demonstrated that NMT outperforms systems based on translation memories, with output comparable to, but with higher fluency than, domain-trained Statistical Machine Translation (SMT) systems.
Adding human expertise to NMT
While the benefits of NMT have made it an important part of translation management, human intervention is still required to review and edit the machine-translated text. This is where the Exfluency subject matter experts (SMEs) enter the process.
Neural networks learn from large volumes of data. The quality and quantity of the data determines the effectiveness of the translation.
By collating edits made by SMEs and feeding what has been learnt back into the NMT system, the neural network can use this knowledge to enhance future translations.
