Even 'the patron saint of translation' got it wrong sometimes. Jerome of Stridon - most famous for supposedly removing a thorn from a lion's paw - was a fourth century linguist who set about translating the Bible into Latin from the original Hebrew. As a second-language speaker, he failed to grasp that the Hebrew word for a horn was used as a euphemism for a ray of light, instead describing Moses' descent from Mt. Sinai in wholly literal terms. For centuries, artists depicted Moses with a pair of horns.

Jerome could have benefited from the other kind of SME: a Subject-Matter Expert who possessed not only a solid grasp of Hebrew vocabulary, spelling and grammar but also an intuitive understanding of how these would have been applied in a highly-specific context. In other words, he needed a technical translator. The distinction between general and technical translations - something we sought to reflect in our recent website redesign - can be demonstrated by a straightforward example. For a general translation (such as an internal memo about not discussing intellectual property with third parties), it may not matter if you choose 'confidential' or 'restricted' for 'vertrouwelijk' as long as the meaning is clear. But if you translate the state secret classification 'Departementaal VERTROUWELIJK' as anything other than 'NLD RESTRICTED', you might be opening up sensitive information to thousands of people with a lower level of clearance.

In the broadest sense, technical translations are a matter of finding le mot juste for the situation at hand. More so than with general translations, this implies a high degree of specialism in a specific area: the right technical translator for court documents might not be a good fit for manufacturing handbooks, and neither is likely to have the required expertise for translating software code. Within their respective domains, however, they should be able to provide more reliable and consistent terminology than even the most advanced machine learning tools, which have a broader technical vocabulary than any human but a shaky grip on context.

Somewhat counterintuitively, perhaps, technical translators should also be able to recognise when their own knowledge doesn't cut it. In a recent assignment for TNO, for instance, I was tasked with translating instructions for their internal reservation and incident reporting system in a Word document. As a third party, however, I had no access to the system itself and was unable to verify whether my choice of vocabulary in the instructions ultimately matched this. It can be easy to simply take an informed guess: the online dictionary Linguee gives 'aanvrager' as 'applicant' in 24 of its 30 examples, so it wouldn't be unreasonable to go for this. Fortunately, TNO was kind enough to provide screenshots of the system when I raised the issue. 'Aanvrager' was exclusively translated as 'requester'.

In short, technical translations are a collaboration between insider knowledge and outside expertise. In the science and technology domain, translators need to keep up to date with changes in a language but must also be aware that their work reflects a specific organisation. Machine learning, as mentioned, is often very useful. But a terminology guide - created in cooperation with the customer - is usually the best way to ensure that a certain style carries through into even the strictest of documents, wherever possible. As the field changes, technical translators ensure that the guidelines evolve. After all, there's a reason we haven't seen a horned Moses these last few centuries.

- Josh