What is Neural MT?

Translations are part of everyday business. They are necessary to address customers across the globe and serve international markets. Nevertheless, many companies have a hard time managing their translation efforts. Even if a translation agency has been found, translations still require a lot of manual work, which not only takes up a lot of time, but also financial resources. With companies depending on them, this in the worst case might result in campaign delays and sluggish communication. Neural Machine Translation seeks to address this exact problem. The technology enables automated and fast translations coupled with transparent processes.

Machine translations are based on a highly-innovative training process for Artificial Intelligence building on the language data enterprise customers already have available. These training data is available in the Translation Memory File (tmx) which is also used by traditional translation agencies. Roughly speaking, a tmx is a database of all previous translations. Together with the data from the specialist field, the tmx forms the foundation to train individual Machine Translation engines. This is what sets the process apart from generic translations. The data used by generic engines is vast and covers a very wide scope, instead of limiting it to custom, company-specific data. They are not assigned to a specific specialist field, which is reflected in low-quality translations. This makes generic machine translations unsuitable for professional applications. Neural MT, on the other hand, exclusively uses the existing language data of the customer. It is tailored precisely to customer requirements and “understands” the way in which the company communicates. This allows the engines to produce pre-translations that approach the quality of human translations. The engines are trained using neural network technology, which collects relationships between the source and target language.

One major advantage of Neural MT revolves around the constant learning process of the engines. Professional linguists proofread and finalize the pre-translations, which are in turn used to train the engine at the end of the process. This constant training process helps the system deepen its understanding of the way in which the company communicates over time. Translations are not only carried out faster; their quality is also higher. This learning process saves costs, and, with it, resources, which companies otherwise have to dedicate to professional translations. The technology is of interest to large enterprises. They already have a large pool of translation data that can be used to train the engines. This results in up to 50% cheaper translations.

Benefits of Neural Machine Translation

Artificial Intelligence in form of machine translations comes with many benefits. By automating the translation process, which is slow in its traditional form, Neural MT is not only quicker, but also more affordable and of a higher quality.

How is this unrivaled quality realized? The Translation Memory File (tmx) grows with every new translation, and the engine understands the specific language of the company better over time. This results in greater linguistic consistency than traditional translation agencies can ensure.

Because the engines are trained continuously, translations not only improve but also become more affordable. The TMX grows, with correspondingly lower translation effort and costs. Pre-translating is the most time-consuming process in translation, and automating this step saves a lot of time. This is without considering the time it takes to search for the right translator.

The translation process can additionally be integrated into the company’s own systems and processes. This makes it possible to design data transfers, which can be difficult, in the easiest-possible way. Every department that needs translations within the company can independently place orders. The process is made as transparent as possible, to make sure that the responsible localization and quality managers always keep an overview of the progress of translation processes.


A certain data quantity is required to train neural networks. It allows the engines to learn the company language on the basis of existing language data and produce high-quality translations. The larger the tmx, the better the engines can be trained. This technology is particularly interesting for larger enterprises, who are sitting on this valuable data resource — often without even knowing it.

This means that a more essential requirement is finding an external partner. Very few companies have the capacities to develop their own machine learning process. Experienced service providers fill this gap, transforming previously unused translation data into custom engines. This ensures optimal use of resources, and translations quickly deliver good returns. Expressed in figures, established providers can deliver translations up to three times faster and at prices up to 50% lower.

AI in the form of Neural Machine Translation has the potential to revolutionize the translation process of companies. The technology not only allows for immense cost savings, but also boosts the quality of translations while making processes more transparent. This is an important step in the push to drive digitalization in companies. With every new translation, costs sink while collecting valuable data. Companies benefit in a range of ways from the use of Neural MT. This makes it the perfect technology to invest in when looking to use resources in a more meaningful way.