So, what exactly is NMT?

NMT (Neural Machine Translation) is the latest trend in automated translation, a field that remains a niche compared to its human counterpart, but is experiencing strong growth around the world. NMT uses neural networks to automatically translate source texts (for example, English) into other languages, without human intervention. The system can then be improved and adjusted to achieve even better results over time. It surpasses previous MT engines (rule-based MT, statistical MT) and its level of fluency is close to human translation in some language pairs.

Thus, neural algorithms are the foundation of this new technology. In this model, sentences are considered as a single unit; this differs from other MT models where the unit is the word. Result? Significant improvement in grammar, style, fluency and consistency. Based on machine learning, the neural translation engine can learn continuously and thus further improve the quality of the translations produced.

NMT: a versatile solution

Can neural machine translation be integrated into your existing translation solution? Absolutely! In reality, the NMT is only part of a professional multilingual translation solution for the 21st century. Whether it’s CAT tools, translation memories, terminology databases or company-specific translation standards, the neural MT engine easily harmonizes with the entire existing translation ecosystem and significantly increases productivity. Indeed, according to researchers in the field of neural translation, significant productivity gains exist compared to older systems, up to 25% reduction in post-editing effort according to Läubli, CTO of TextShuttle.

 

 

And what about quality?

Since Google Translate’s first beginnings in 2006, automated translation has had rather bad press, and with good reason. Old rules-based systems or statistical models offered disappointing results. However, with the acceleration of progress in machine learning and artificial intelligence, the new place of MT is that of a translator’s tool, just like the legendary CAT tool SDL Trados or the useful Antidote from the Quebec firm Druide informatique. And anyway, and for several more years, the translator will have the last word on the final translation during the post-editing process. The quality control functions integrated into CAT tools and the use of spell-checkers and grammar checkers together with careful revision guarantee the superior quality of translations produced with the help of MT.

I invite you to watch these two recordings (French) of the presentation on MT that I gave at the first Meetup of the Machine Translation Montreal group, on May 16th. Next meeting is July 11th; feel free to join us to learn more or to discuss with group members.

Laurent Chevrette holds a Bachelor of Science degree from the Université de Montréal and has over 10 years of experience in translation. Passionate about the progress of artificial intelligence, he is an expert in the deployment of automated translation solutions in small and large companies in Europe and America. When he is not implementing a neural translation system, he can be found on ski slopes or diving.

Also available in: Français English