In the age of digitalisation, the world of translation is also facing a revolution: post-editing. Specifically, content is translated by a machine translation engine (which may or may not employ “artificial intelligence”), then reviewed and corrected by a translator capable of grasping the nuances of human discourse written in another language.
The latest advance in this field is called neural machine translation or NMT. Underpinned by deep learning, it mimics human neural networks with a view to learning and improving performance over the long term. This improvement can be optimised for a specific area of application by customising the engine with terminology supplied by the user.
In order to respond to the increasing volume of content to be translated and to reduce the costs associated with this activity, certain large groups – with the financial clout needed to support such an initiative – have invested in developing their own neural translation engines.
Since 2020, fidestra has been working on a regular basis with a major client that uses this solution, carrying out so-called “advanced” post-editing projects.
This qualifying adjective sets the process apart from traditional post-editing, which is widespread in the industry and which consists of modifying the translation provided by a generic engine (DeepL, Google Translate, etc.). However, if the aim is to ensure optimal quality and confidentiality as well as respecting legal obligations around data protection, this solution accessible in just a few clicks very quickly reaches its limits.
By contrast, in advanced post-editing, the company’s internal NMT engine is trained by a team of linguists and benefits from guaranteed maintenance and development carried out by knowledgeable IT teams. This solution therefore provides not only results that are better and more usable, but also a reinforced level of security, since the information never leaves the servers where it is stored.
Over the past three years, fidestra has developed real expertise in this advanced form of post-editing and a keen awareness of the pitfalls that the machine – as sophisticated as it may be – can unwittingly throw up.