Some texts are perfect for neural machine translation. Which ones?

Translation customers today are becoming increasingly aware of the vastly improved quality of neural machine translation from DeepL to Amazon and others. However, neural machine translation remains far from perfect, especially for particular kinds of texts. The current revolution in machine translation thus involves a paradox: Despite (or indeed because of) the improved quality of neural machine translation, customers are more uncertain than ever about how about their texts should be processed.

Here’s a hint: Texts from the legal domain, but also from areas of medicine, business, and the natural sciences, are often ideal for machine translation because their word choice and grammar are often highly standardized.

Unsurprisingly, the quality of neural translation engines is high for these kinds of texts because they were trained trained on them. DeepL and other services love standard legal contracts.

Some examples of standardized texts that are usually great for neural machine translation include:

Legal:

  • General Terms and Conditions

  • Data Privacy

  • Basic contracts

Business:

  • Financial reports

  • Informational emails

  • Corporate communication

  • Websites

Natural sciences/medical

  • Patents

  • Brochures