Health Systems Action

Solving the Twelve Language Problem in Healthcare

Over the years, I’ve used Google Translate where I had to, including on trips to China – and found it useful enough. But Google Translate is primitive compared to the ability of today’s generative AI applications to read and write fluently in multiple languages. When it comes to languages, my mom’s not bad for a human: she speaks five, including Italian passably and chats daily on WhatsApp with her dear friend Federica in Bergamo. When I first encountered ChatGPT early last year, I asked it, in English, to write a poem for Federica (prompt: “sympathetic, but with humour”), about the miserable time she was having with a bout of shingles. Thirty seconds later, the poem was in Federica’s hands via WhatsApp and my mother. Declaring her joy and amazement at my mom’s literary skill in Italian, she immediately circulated the piece to friends and family. I had to urge Lina to reveal to Federica the poem’s true author!

As a country with 12 official languages (sign language was added in 2023) and amidst concern about the neglected status of many of them, we should be keenly interested in how AI-based translation might help tackle more significant communication tasks and language barriers than this one, such as in health care and education.

Image by Freepik

A recent podcast on AI-based language translation had some interesting insights. I learned that AI, particularly through large language models (LLMs), has dramatically changed the translation industry, enhancing the ability to communicate globally.

  • The industry is moving from initial frenzy around AI capabilities to building trust in AI-powered translations by ensuring accuracy, cultural sensitivity, and appropriateness.
  • AI is not replacing human translators, but AI-based translation is doing the heavy lifting. Human translators have an evolving role, acting as “copilots” who focus on tasks where human reasoning is crucial. They do AI model training, provide ethical oversight, and specialised content validation. Their skillful prompt engineering guides the creation of accurate and appropriate translations.
  • Better AI context understanding has improved the translation of idioms, metaphors, and context-specific language, bridging a difficult gap.
  • The importance of using AI in translation ethically and responsibly is emphasised, especially in sensitive applications (like healthcare), to avoid biases and ensure cultural and contextual accuracy.
  • AI technologies are enabling translation of lesser-spoken languages and dialects, thereby democratising access to information and services.

What do these advances mean for healthcare in SA – and other places with linguistic diversity? Here are five ideas that might enhance healthcare through AI-based language translation:

  1. Develop and integrate AI-based translation services that can handle South Africa’s linguistic diversity in healthcare settings. These systems should be trained on medical terminology across all the official languages. This will improve communication between professionals and patients, leading to better diagnosis and treatment outcomes.
  2. Use AI to translate and localise public health campaigns and educational materials into all official languages specific to regions. Accurate translation can help public health initiatives to reach further and be more effective.
  3. Integrate real-time, AI-powered language translation into telemedicine platforms to cater to patients in rural and underserved areas who may not speak English or the language of the healthcare professional. This will help democratise access to specialist consultations and medical advice.
  4. Healthcare professionals in South Africa should receive training on AI translation tools and technologies to prepare them to use these tools effectively, understand their limitations, and provide feedback for continuous improvement.
  5. We should ensure the development and use of AI translation tools in healthcare is guided by ethical standards and cultural sensitivity. This involves regular auditing of translations for accuracy, bias, and appropriateness, particularly in sensitive medical contexts, and involving local language experts in the process.

Over 50 languages are officially supported by ChatGPT including Afrikaans but not isiXhosa or Zulu, however ChatGPT will still happily translate them upon request. To move forward, we need to assess the current quality of these translations in all the indigenous SA languages, and to advocate to tech companies for their increased prioritisation. Training sets are the key to increasing the proficiency of LLMs in less used languages.

It’s interesting to note that AI-based translation into sign language is possible – one innovation being animated avatars that perform sign language gestures.

Image by Freepik

The way translators are using AI tools is a model for healthcare – not replacing humans but augmenting human capabilities.

Finally, let’s hope better translation leads to better human understanding – and not just in healthcare!

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top