Neural machine translation (NMT)

Neural machine translation uses deep neural networks to translate text, weighing the context of an entire sentence rather than swapping words one at a time. This is why modern translation reads fluently and handles idiom and word order far better than older rule-based or phrase-based systems.

In a real-time pipeline, NMT is the middle stage between speech recognition and speech synthesis, and is tuned to begin translating before a sentence is complete to keep latency low.