To Newsroom

Tempted by free machine translations? A fluent translation may be full of content errors

While machine translation can be a useful tool, translating medical texts with machine translation alone may put patient safety at risk, writes Elsa Lehtonen from Oriola’s translation team in her blog post.


Over the past few years, machine translation technology has taken great leaps forward. Text produced by modern machine translation tools often appears quite fluent: conjugations are mostly correct, terms – even unusual ones – are translated right, and the vocabulary seems to fit the context of the text. Free online machine translation can be very helpful when trying to get the gist of a foreign-language text. However, it is not a good idea to publish unedited machine translations on your website or print them on a medicine carton – particularly if you don’t understand the target language yourself.

The apparent fluency of the translated text conceals one of the pitfalls of machine translation: the machine doesn’t actually understand the text it is translating. A seemingly fluent machine translation may in fact be full of content errors. In a medical translation, this could, at worst, put patient safety at risk. A machine also doesn’t understand different text types or their purposes: it doesn't know when to use the European Medicines Agency's templates or the Standard Terms database in a medical translation.

From a professional translator’s perspective, a machine translation engine can be a valuable part of the translator's toolkit. A professional translator’s principal tool is typically a translation memory system, i.e. a software programme that stores translated text segments in a database. Like most machine translation tools that professionals use, the machine translation engine used by Oriola’s translation team is integrated into our translation memory system. To improve the quality and usability of the machine translations, the translation models used in our engine have been fine-tuned using Oriola’s own high-quality translation memories, which contain translations produced by our experienced professional translators. In other words, the machine translation models we use have been tailored specifically for medical texts.

Oriola’s translation team works with dozens of different texts every week. In addition to the Nordic languages, we also help pharmaceutical companies with translations into many other languages. Having a good machine translation tool can significantly speed up our work as the machine provides an initial rough translation for our professionals to edit. The machine translation tool suggests a translation if there is no suitable match in our translation memory. Our translators use their linguistic expertise to assess the quality of the machine-translated suggestion. They can then either edit the translation as necessary or reject it and translate the segment from scratch. This ensures that any errors produced by the machine do not end up in the final translation. By combining the speed of the machine translation tool and the expertise of our human translators, we are able to produce high-quality translations more quickly and cost-effectively.

Compared with many free online translation programmes, machine translation tools designed for professional use also tend to have significantly better data security. At Oriola, the machine translation engine is installed locally on each translator’s computer. The engine never sends or uploads the translated text to any external location, such as a machine translation server. The machine translation is produced securely on the translator’s own computer and no one else can access it.

Sometimes customers ask us to review and correct translations produced by their own machine translation tool. This approach is best suited to simple texts with a predefined structure, as these are the kinds of texts that machine translation engines tend to do well with. For more challenging texts, it’s faster to translate the document from scratch using the translation memory system, which allows the translator to seamlessly utilise translation memories, termbases and machine translation as needed.

Machine translation has its time and place. In the hands of a professional, a machine translation engine is an excellent tool that speeds up work and saves both the translator’s and the customer’s time. Nevertheless, even the best machine translation still requires human review before publication to eliminate any machine-generated mistakes and clumsy turns of phrase. Patient safety should not be left in the hands of a machine.

Elsa Lehtonen works as a medical translator at Oriola.