Medical accuracy in the age of AI: Why context still reigns supreme 

In the world of AI-powered medical translation, the rise of artificial intelligence has been nothing short of revolutionary. AI-powered translation tools now offer unprecedented speed, cost-efficiency, and scalability. For global pharmaceutical and MedTech companies, this means faster time-to-market, smoother localisation workflows, and the ability to manage dozens of languages simultaneously. 

But as powerful as AI is, it has one significant blind spot: context. 

The hidden danger of isolated strings 

Despite its capabilities, machine translation for healthcare has a critical limitation: it often lacks contextual understanding.  

Medical communication is not just about words. It’s about meaning, nuance, and cultural clarity. Even a single misplaced term can have significant consequences in a healthcare setting. AI, for all its linguistic training, often lacks the ability to distinguish between homonyms, understand tone, or interpret intent without clear context. 

Over the past year, we’ve had multiple clients approach us with content generated by AI translation engines for their medical apps. In each case, the source content was composed of isolated strings — interface terms, labels, buttons, and error messages — often stored in spreadsheets or code repositories, completely detached from their real-world usage. 

At first glance, translating these strings seems like a quick task. But we quickly identified major issues. 

Take the word “lead”. In one app, it was used in the cardiology sense (ECG leads). But AI, unaware of this, translated it as the verb “to lead” or the noun referring to leadership. The result? Grammatically correct, but completely wrong — and potentially confusing in a medical context. 

Another recurring issue: the term “control”. In clinical research, this might refer to a control group. But when presented alone, the AI rendered it as a verb (“to control”) or as a UI element (like a button), depending on its training. Without contextual cues, accuracy fell apart. 

Why context is essential in AI-assisted medical translation 

In user interfaces, patient apps, and digital health tools, isolated strings are the norm — but that doesn’t mean they can be translated blindly. Especially in healthcare, even a short phrase can carry critical meaning. Without additional context, even the best-trained AI (and even a human linguist!) can make the wrong call. 

Clients often assume that by providing source files alone, the job is 90% done. But in reality, the quality of the output depends largely on the input: metadata, usage notes, screenshots, reference files, and clarification about tone or audience all make a massive difference. 

How to prevent these errors? 

At Novalins, we’ve built processes to handle this challenge: 

• We request contextual notes for isolated strings. 

• We encourage the use of tools that allow string tagging, screenshots, or developer comments. 

• Our medical linguists are trained to flag ambiguities early in the process, prompting dialogue instead of assumptions. 

• We recommend creating and approving glossaries with the client before the project begins. 

• We ask for reference material, such as previous translations or design mockups, to better understand the content’s function. 

• We suggest involving someone from the client’s internal team who can provide live feedback during the project, directly within our TMS, to resolve issues in real time — not just after delivery. 

• And most importantly, we emphasize constant communication between the client and our project team to ensure full alignment from start to finish. 

Collaboration makes quality possible 

Whether you’re using AI to pre-translate content or relying on a human team from the start, the takeaway is clear: context must be shared. A skilled LSP can help guide the process, optimize string documentation, and ensure that your app content is safe, accurate, and compliant. 

When working with AI translation, speed and cost savings are real. But so are the risks. The best results come from collaboration — combining automation with expert review, and rich context with linguistic judgment. 

In the fast-paced world of digital health, where every word can impact a user decision or a regulatory outcome, context isn’t just helpful — it’s critical. 

So if you’re preparing app content, don’t just send the strings. Send the story behind them. 

References 

1.  Genovese A, Borna S, Gomez-Cabello CA, Haider SA, Prabha S, Forte AJ, Veenstra BR. Artificial intelligence in clinical settings: a systematic review of its role in language translation and interpretation. Ann Transl Med. 2024 Dec 24;12(6):117. doi: 10.21037/atm-24-162. Epub 2024 Dec 17. PMID: 39817236; PMCID: PMC11729812. 

2. Delfani, J., Orasan, C., Saadany, H., Temizoz, O., Taylor-Stilgoe, E., Kanojia, D., Braun, S., & Schouten, B. (2024). Google Translate error analysis for mental healthcare information: Evaluating accuracy, comprehensibility, and implications for multilingual healthcare communication (arXiv:2402.04023). arXiv. https://doi.org/10.48550/arXiv.2402.04023