Neural machine translation and postediting


Neural machine translation, also called neural MT or NMT, refers to the translation of a text by a computer, without human involvement. Based on the current state of technology, a machine-translated text is suitable for information purposes. However, such texts are not ready to publish.

What are the advantages of a neural machine translation?

  • Large amounts of data can be translated very quickly; a raw version is essentially available at the push of a button
  • Low costs
  • Can serve as the basis for a professional translation

What is postediting (PEMT)?

Postediting is when a machine-translated text is then edited by a translator. We distinguish between simple (“light”) postediting and extensive (“full”) postediting. Light postediting involves correcting only the obvious errors. In full postediting, the machine-translated text is edited to the extent that it reads like a human translation.

What is the difference between a generic MT engine and a proprietary MT engine?

Generic translations are available from Google, Microsoft and DeepL, for example. The disadvantage of these MT engines is that they cannot be combined with a company’s own technology, which results in a higher level of effort in the postediting process. Generic MT engines use many variants of expression in order to appear as close as possible to natural language. This practice is not desirable in technical documentation, however, making it necessary for human translators to make corrections here, too. Data security is also an important factor there. Free, generic MT engines do not offer such security, since all of the data entered is stored on third-party servers.

A proprietary MT engine takes specific terminology into account and offers the better solution. However, very large pools of data are required in order to train the engine.

What role does terminology play in machine translation?

The terminology is of prime importance, both with generic and with proprietary engines.
The better the engine is able to isolate terms and eliminate synonyms, the less postediting is necessary.

Do machine-translated texts have to be postedited?

Yes, proofreading by a human translator is absolutely necessary. A machine translation can accelerate and simplify the work of the human translator in certain cases, but it cannot replace it. A machine translation can be sufficient if a translation is needed for information purposes only.

Can all texts be machine translated?

No, texts with a very sophisticated style or with cultural codes - such as advertising texts - are only suitable for human translation. We are happy to advise you on which texts are best suited for machine translation.

Would you like to learn more about neural machine translation?

We are happy to discuss with you how you can use the advantages of such a system in your business, thereby saving costs on translations.