The localization industry has been watching machine translation (MT) for years. Despite some concerns about the practical aspects of using the most radical change the language industry has ever seen, most professionals agree that new technologies are here to stay, certainly when neural machine translation (NMT) is part of the picture. NMT is the biggest megatrend to hit the language industry, the blockbuster of localization. We call it a megatrend because it is particularly fundamental, persistent, and not limited to singular aspects of our work. It will affect everything: business, process, technology, people, quality management, decisions on rates, and the kinds of professional roles needed to do this work well. But NMT can’t stand alone. It is the only language processed by a machine, after all. Humans train and operate the NMT systems and make choices about style, tone, and feeling. The human role will be more important than ever, with new skills required for working with NMT. In order to facilitate the interaction of technology and humans, workflows and functionalities are used that make it easy to interact– (not to type new translations from scratch, but to edit, in most cases) and to rely on previous good translations that are still valid. memoQ, a state-of-the-art translation environment, integrates NMT with translation memory, myriad editing functionalities, QA tools, and the most comfortable working environment available. This all goes toward supporting the humans behind the machines, to get the very best out of the human-technology interaction. We see smart, hybrid workflows coming in which NMT and translation memories go hand-in-hand to enable humans to translate ever increasing volumes. These workflows can be extremely flexible, facilitating translations for many purposes and in many language combinations. We believe MT should not stand alone. MT raw translation for many languages (as long as large corpora are available and collected in advance), but it is the many other functionalities of a translation environment that allow editing and QA of the MT output so that it can be leveraged as validated material for future use. Not only can NMT systems learn and improve their own output, they can also profit from the extended options of a smart, time-tested translation technology when used in a hybrid process, improving results from both systems. How can you exploit NMT and translation memories to translate faster and more efficiently than ever before? Let’s look at two challenges related to NMT: terminology and context. One solution that addresses both is Microsoft’s machine translation plugin, Translator Text API v3.0. This version is based entirely on NMT and supports over three dozen languages. One of the features, Custom Translator, allows you to build and train custom NMT models by uploading term bases, translation memories, and other documents. Custom Translator is a paid service that requires some of your time and expertise. With it you can create a private NMT engine. You train the system with your resources for your own use, and it’s never accessed by third parties. Challenge of Terminology Whether a human or machine translates, terminology needs to be appropriate for the subject matter at hand. NMT can learn from subject-matter specific glossaries to improve accuracy. For example, with Amazon’s NMT service, which is accessible in memoQ through the Intento MT gateway plugin, you can upload your glossaries in various topics (such as law, economy, audiovisual, etc.) for later use in memoQ projects. You select the appropriate glossaries prior to beginning work on a translation. If the system detects a match for a word in a sentence to translate and term in the glossary, it will use the preferred term during translation. Challenge of Context To assure that the context for your translations is correct when using NMT, you can supplement NMT with matches from your translation memories. Let’s say you have a general NMT model (not customized for a specific subject matter) to which you add matching source and target segments from a subject-specific translation memory. In such a case, the MT service translates the given sentence and also modifies the translation according to the uploaded translation memory result, thereby providing the user with a contextually accurate translation. The Human Element in a Hybrid Solution The best solution is undoubtedly one that involves humans at critical points. Even the best possible NMT engine using a custom model and incorporating appropriate terminology and relevant segments from translation memory needs a skilled language professional to edit the NMT result. Armed with memoQ’s suggestions from a wide array of resources like translation memories, term bases, and LiveDocs corpora, the human element in hybrid workflows can increase your confidence in future reuse of translated content. Simply put, when NMT engines are joined with productivity and collaboration tools like memoQ in one workflow, you can handle more localization volume. We are proud to help our industry advance with tools to bring more content to ever-growing international audiences.