Machine translation

Machine translation is a translation generated by a computer program – i.e. fully automatically. We work with baseline engines, and we also train neural models. This means the output is representative and therefore more suited to your specific domain.

The technology has been around for more than 60 years and has improved tremendously over the last decade, particularly with the introduction of ‘neural models‘. Although the quality depends on the language pair, the source text and the customisation of the MT system, the output is usually at least good enough to understand the message and increase the productivity of the translator who edits the ‘pre-translation’.

Untranslate helps its clients deploy existing translation technology or build their own engines that understand the specific terminology. This is done by feeding the engines with parallel data in the source and target language. One condition is that the data is representative and qualitative, and meets a number of formal requirements. It may therefore be necessary to first clean up the data using routines and special tools, and identify problematic passages using automatic checks.

Are you convinced that, in certain cases, post-editing machine translation leads to a faster and therefore cheaper production process than ‘manual’ translation, but are your translators sceptical? Untranslate has been organising experiments involving objective productivity measurement at companies and organisations for more than 10 years. Through examples and practical exercises, linguists discover the real potential of MT applied to their own personal situation. A clear explanation of how the technology works contributes to a better understanding of possible error patterns so they can be recognised and corrected quickly. Examples of successful use cases serve as a source of inspiration for building innovative and integrated solutions within the organisation.