AI in medical translation: balance speed and safety

How to best use AI technology for your medical translations


Improved access to information, the ability to weigh up the benefits and risks of suggested treatments and make an informed decision, accurately following medical guidelines, informed consent, and much more… medical translations can save lives and provide your patients and users with a better, more equitable experience.

Several examples of erroneous machine translation have been noted recently by The University of Aberdeen. A series of package leaflets were machine- translated by New York pharmacies across six different languages.


“Once a day” in English was translated as “eleven a day” (because “once” has the meaning of “eleven” in Spanish). Similarly, boca (which means “mouth” in Spanish) was spelt as “poca” instead (meaning “little” in Spanish), which caused customers to incorrectly assume that the medicine should be taken in small quantities rather than simply orally.

Given the severity of the subject matter (and clear potential consequences!), enlisting the help of Artificial Intelligence in the medical translation sector was traditionally considered a risky endeavour. But AI-powered Machine translation tools are becoming increasingly efficient and more widely accepted in the medical sector. Neural Machine translation (NMT), and more recently, Machine Translation development, has changed the game. Instead of decoding word-by-word or phrase-by-phrase, the technology looks at the entire content of a sentence, understands it, and produces a much more fluid and usable result.

But to get the most out of technology, we need to use it to the best of its ability, and understand its limitations. Here are some best practices and guidelines on how to leverage AI technology for your medical translations.

Find out if AI is suitable for your needs

Machine Translation output can’t be reduced down to being either ‘good’ or ‘bad’. The quality varies depending on the languages, subject areas, and the engine being used. It generally performs best with technical content, such as clinical and regulatory documentation, research papers and studies, reports, medication inserts and product descriptions. It’s weaker for creative and high-value content, so we would advise exercising caution for translation of ads, landing pages, marketing copy and psychometric content.

Choose the right engine for your content

You’ve probably already heard of the major players, like DeepL and Google Translate, but there are plenty of other available machine translation engine options to choose from. Some specialise in specific language pairs, others in certain subject areas – Rozetta T-400 and CrossLang are among some of the MT engines that specialise in medical translations.

Give your tool a helping hand

With a large budget, you can develop and train tailor-made models with your own data. This kind of investment requires a lot of time and input to get to a usable state. For smaller budgets, utilising Translation Memories and adding glossaries to your MT engines is also effective.

Enlist human help

Time and accuracy are of the essence when you’re dealing with people’s lives.

Machine Translation speeds up the translation process considerably, and can be a cost-efficient solution. But that’s not the only thing: AI can support human translators in many ways. Through research – by helping them quickly find resources and information that is relevant to their translations – and by helping them maintain accuracy (through the use of glossaries and Translation Memories) and keep content consistent throughout.

On the other hand, professional human translators are excellent at spotting and mitigating a machine’s potential problem points. They have a refined ability to recognise the context and reproduce the desired tone of voice and can localise the text if needed.

If you’re publishing materials or working with critical documents, then it’s a good idea to have a human edit your machine-translated materials ahead of distribution, to ensure they are accurate and ready for publishing (this is called Machine Translation Post-Editing, or MTPE).

Feed your processes with feedback

Establishing a feedback process afterwards will be helpful for future projects and can provide you with the opportunity to refine and review your Translation Memories and Glossaries.

Don’t hesitate to ask us for support

Monitoring devices that keep patients safe and trigger for early detection, robotic-assisted surgeries with lower risks and faster recovery times, the ability to analyse data sets with lightning-fast accuracy, and much, much more… AI is already being leveraged in the healthcare industry for the better. If used properly, it can significantly improve patient experience and safety across the board. We believe that AI-powered translation services can also contribute to positive, life-changing developments in healthcare and medicine – if administered properly.

As always, if you have any further questions or would like some specialist advice, contact us and we’ll set up a meeting to discuss your requirements.

Share this post