The need for quality assurance in Neural Machine Translation

Neural Machine Translation (NMT) marks a major milestone in the evolution of machine translation, creating perfect results at low cost. Perfect? Well, not always. And this is where translation quality assurance comes in.

Text by Michael Schneider


Image: © metaworks/

Whenever the conversation with clients turns to the use of automatic machine translation (MT), language service providers (LSP) face the same misconceptions and myths: MT is almost free and produces perfect results.

With a bit of explanation, you might get clients to understand that the use of first-class technology generally does cost money and that not every language pair is equally suited for MT. But the need for (expensive) quality assurance (QA) and its impact on the overall achievable reduction of translation cost is a lot harder to illustrate. Let’s give it a try...

A high-level view of the process

In an industrialized LSP context, machine translation is not used as a “stand-alone”, but rather as one of several tools in a complete translation process. This process includes quality assurance, standard translation memory systems, and process and workflow management components.