AI in translation: Melanie Cole on language, innovation, and the future of healthcare communication

I grew up bilingual in Brussels, Belgium, with a Danish mother and English father. I went to an international school, so I spoke Danish, English and French from an early age, and always had a great interest and passion for languages, Spanish in particular. I studied a BA in translation and interpreting in Barcelona, where I improved my Spanish and started learning Catalan. After that, I completed an MA in Applied Translation Studies at the University of Leeds, specialising in French and Spanish. When I was offered an in-house translation position at Sandberg, it felt like a natural progression to follow this career path, using my multilingual skills and the various translation techniques I had been taught at university.

In recent years, especially since the release of ChatGPT two years ago, there has been a lot of hype surrounding Large Language Models (LLMs) and prompt-based chatbots, and their potential to impact industries, including the translation industry. Many people feared that AI might replace translators or reduce their workload. Instead of viewing this as a threat, I saw it as an opportunity to learn and understand how AI can support and complement human translation. Realising that AI could be used to enhance, not replace, the translator’s role sparked my interest in specialising in AI translation and coordination.

Since the mid-2010s, there has been a noticeable shift where many Language Service Providers (LSPs) have increasingly adopted Machine Translation Post-Editing (MTPE) as standard practice. This involves using a machine translation engine integrated into the CAT tool, followed by a translator reviewing and refining the output. Unfortunately, this trend combined with the more recent developments and use of AI in the translation industry has seen some translators steer away from their creative and linguistic side, as they focus primarily on reviewing and correcting machine-generated content rather than translating from scratch.

This has led to a fair amount of translators leaving the field, or fewer translators getting hired, as it not only reduces their workload but they are also finding their rates reduced, so their experience and qualifications become less appreciated. This trend, exacerbated by the growth of AI, is therefore a challenge for many professionals in the industry.

A common misconception is that AI will completely replace human translators. Professional translators are taught a range of techniques and theory, specializing in various subject areas. In fields like healthcare, legal, and finance, the risks are particularly high because of the legal implications involved, and AI-generated translations without human review can have serious consequences. It is important to understand that AI should be used to assist translators, not replace them. This could be by producing first drafts that are then edited and proofread. It can also support translators with admin-heavy duties, such as terminology and translation memory management and might be useful for informal communication (such as social media) or for getting the “gist” of something in another language. In other words, machine translated text without human review should never be used for public consumption.

I worry there is a growing misperception that unqualified translators or AI can easily do the job. This is particularly significant in healthcare, where there are cases of translated text being presented to patients or clinicians without any form of secondary review. Low-quality translations can have catastrophic consequences, such as inappropriate patient care, incorrect diagnoses or, in extreme cases, death. Unregulated AI translations therefore present a serious risk to patient safety, so it is equally important to have adequate policies in place when using AI.

An experienced human translator, in my opinion, is essential in the use of AI translation tools. They should be present in all stages of the process, as an inexperienced translator should not be allowed to handle such tools. This is because of the possible inaccuracies, inconsistencies that can occur in the AI translation process, which only experienced translators trained at the languages in question would be skilled at picking up. Ideally, translators will be able to fine-tune AI models using their own translations to produce a perfect target translation. Once the tool is run they can post-edit the translation and have it reviewed by a third person, ideally a translator or expert in the subject. Human translators are not going anywhere, they are simply growing alongside AI.

AI can be useful for generating initial drafts of translations, but as I keep mentioning, it is crucial that human translators review the output. The translator must also have a perfect understanding of the source text. This review process ensures linguistic accuracy, consistency and cultural sensitivity, and compliance with industry-specific regulations. In healthcare, we need to ensure that translations are reviewed by either medical professionals or specialised translators, or both, to avoid critical mistakes which can have serious consequences. It is also a good idea to have a layered review process, where one translator reviews the machine generated output and a second translator reviews the translation before it is finalised.

There are many risk factors involved around coordinating translations that rely on AI. Firstly, the type of damage end users can incur and who ultimately is liable for any factual or errors with serious consequences. Secondly, risks in terms of data protection where sensitive information is used in free online tools, and who would be liable for this. This, as well as ensuring that translations are adequate, is why it is important to ensure that adequate policies are in place and that adequate tools are appropriately used. This is of a major concern, because not many policies have yet been written for AI and translation as the concept is still relatively new. Having appropriate guidelines and disclaimers in place where AI is used would solve this issue in the first instance. When I work across multiple languages as a translations coordinator, I make sure I use accredited translation companies or translators from the Association of the Translation Companies, ITI or CIOL, as they will have appropriate procedures in place for handling AI.  

One of the most rewarding projects I have worked on at EIDO Healthcare was the translation of our patient information leaflet library into Hindi and Marathi to be piloted for hospitals in India. For this project, the initial translation draft is AI translated into Hindi and Marathi while using our translation memory as the base text. This is then proofread and edited by human translators and sent for a final review by an accredited medical organisation before publishing. This layered review process ensures that any inaccuracies and mistranslations are picked up along the way, and by using a translation memory, translations and terminology are kept consistent and up to date with previously translated content. This ensures patients in India can be given information leaflets that are both understandable and precise as well as being culturally sensitive to their needs.

The best way to address cultural nuances and idiomatic expressions when working with AI is, as I keep saying, by ensuring human review. While AI can generate translations quickly, it cannot understand the cultural or idiomatic context in the same way a human translator can. A professional translator with cultural expertise can refine the output to ensure it is accurate and culturally appropriate

In healthcare, cultural relevance is critical. We ensure that translations are reviewed by native-speaking experts who understand both source and target languages as well as the cultural context. This helps avoid misinterpretations or misunderstandings that could impact patient care. In addition, professional translators with specialised medical training are essential in ensuring that technical and medical terms are used correctly.

In the next five years, I see AI becoming an increasingly integral tool in the translation process. AI will likely improve in accuracy and efficiency and can increase productivity for translators. However, the role of translators will remain essential, as human oversight will still be needed to ensure quality, consistency, and cultural relevance. Translation coordinators will likely be more involved in managing AI tools and ensuring that the right models are applied, as well as overseeing the review process. I believe the role will evolve into one that focuses more on curating and refining AI outputs, providing a higher level of expertise in combining human insight with AI technology.

Tools such as BERTScore or BLEU have been developed to assess and score the accuracy of translations, which are fairly good and is useful for myself as a translation coordinator. This still has room for improvement, but the more these models are trained, the better they will become, and both AI and human translations will find themselves reviewed by these models.

The most surprising lesson I’ve learned is that the "hype" around AI potentially replacing translators is not as detrimental as it seems. While AI is advancing, it has not yet reached a level where it can replace the nuanced skills of a professional translator. In fact, I’ve learned that AI can be a powerful tool to assist translators, but it cannot replace the human touch that ensures quality, accuracy, and cultural relevance that is so vital for producing translations. From a translator’s perspective, keeping up to date with all the new developments in AI and ensuring that it is more of an AI-human collaboration rather than a replacement is key.

If I could change one thing, it would be to place more emphasis on the importance of professional translators in industries like healthcare, law, and finance, where the stakes are high. AI tools should be viewed as aids, not replacements, and I believe more industry-wide recognition is needed regarding the value that qualified, specialised human translators bring to these sectors. Additionally, there should be greater awareness around the importance of secondary review in AI-assisted translations, enabling them to evolve together.