Whilst artificial intelligence (AI) is rapidly becoming integrated into our daily lives, pitfalls in accuracy and nuance remain – precluding its use independently in technical translation work. But by combining the powers of highly efficient new “large language models” with the skilled human touch Physiopedia Plus (Plus) has a tried and tested system for translating courses faster and better – improving access to high-quality rehabilitation knowledge all over the globe.

It arrived faster than anyone thought. AI has taken the world by storm since the release of ChatGPT (OpenAI), and the speed at which it is now being applied in a variety of fields is increasing at a dizzying rate. In the translation of specialised knowledge, certain elements of AI have already been increasingly present in recent years. However, the combination of pure translation with the actual “learning” of an interpretation and the offering of solutions to problems is an innovation that breaks previous boundaries.

At Plus, specialist knowledge for rehabilitation has been translated into five languages – Ukrainian, Spanish, Italian, German and French – since 2022. Our translators are all physiotherapists who have experience in fluent, multilingual communication in the rehabilitation sector. Now, it would be easy to simply leave the translation of the courses and pages solely to one of the highly efficient new “large language models” that form the basis for AI systems such as ChatGPT, Jasper, Perplexity or Claude. But several factors show that this is not always an ideal solution.

Some of the main pitfalls of AI-based translation are still terminological errors. A harmless example is the case of a “tender” muscle, an attribute for sensitivity on palpation that is often mistaken in German AI-translation as “romantically affectionate”. But what may seem amusing in such smaller issues can quickly become a problem with medical content. If, for example, substances are misinterpreted, interventions mistaken or laboratory values mixed up, patients may come to harm. This underlines the importance of context sensitivity, an ability that AI systems are still beginning to learn in order to understand, for example, if the abbreviation “KET” in a text means “ketamine” or “ketoconazole”.

However, our translators also face other challenges while working with emerging technologies. Depending on the system and the language, for example, AI-based machine translation may “forget” parts or entire sentences in long texts. An even more serious problem happens when cause and effect are swapped in a section of text. This changes the entire meaning of what is being said, something unacceptable in technical literature.

The more complex AI systems become, the more convoluted some types of misunderstandings may emerge. Sophisticated systems can, in certain circumstances, look for alternative ways to solve a task more efficiently, instead of using the exact given instructions (“prompts”). This might occur because the system is designed to provide assistance, even when it doesn’t have the optimal solution to our query. In this case, the results may not be as good as expected, having to be overworked. In the very early and experimental phase of artificial intelligence in which we are right now, such errors may appear and should not be ignored.

However, these challenges should not deny the fact that AI systems are a considerable help and improvement in everyday translation work. Thanks to machine translation software, we can translate more, faster and better. At a time when up-to-date evidence is more important than ever and the need for high-quality rehabilitation knowledge is clear, a translation result must not only be very good, but also available quickly and in large volumes.

For this reason, the Technical and Translation teams at Plus work together on a daily basis. New programs are tested and results are evaluated, accepted or rejected in order to improve our output with every text. Every day, our translators work hard to turn the raw, machine-generated substance into an excellent knowledge base – a manufacture of words, made by people for people and for the benefit of our patients.