[Re: https://www.hifa.org/dgroups-rss/please-send-details-international-patie... ]
Dear Neil,
Thank you for giving me this opportunity to talk about my work on the use of language in health. I have been interested in the use of language and the interplay between the use of language, culture, knowledge formation and dissemination for a long time. My mother tongue is Romanian, and I grew up in a multi-lingual environment (Romanian and Hungarian). Since my PhD studies, I have been completely immersing myself in English, a language I use now the most (I think I think in English :). In the last few years, I have learnt some Chichewa because I have been living and working in Malawi.
As a computer scientist, I have channeled some of this interest in computational linguistics projects: e.g., developing interfaces in natural language, building linguistic tools and datasets for machine translation. My interest in language for health emerged indirectly, through my work in analysing data from case surveillance forms for capturing COVID-19 cases in Malawi. The forms themselves were in English, but the actual data collection took place in local languages, primarily in Chichewa. This meant that data collectors had to interpret and translate questions and patient responses, often on the spot, before transcribing them into English on the forms. Many of the data collectors had limited proficiency in medical terminology in English and local languages, leading to significant inconsistencies, omissions, and ambiguities in the records. In many cases, what appeared to be data gaps could have been in fact, language gaps. Additionally, I looked at medical records and the use of language and patient understanding.
Thank you for your suggestion to join the multilingualism group. I had been in touch with some people at Clear Global by email. I look forward to learning more about what they are doing. In my work I collected a vocabulary with medical condition, symptoms and anatomy in Chichewa. I want to now put that together in a concise dictionary as an electronic and possibly printed resource. I see this work also as contributing to language preservation! At the moment I do not have monetary resources specific for this work, so any incremental progress is made as time and work permits. There is a need to consult more widely with medical professionals and linguists. Patients are often treated primarily as recipients of healthcare, rather than active participants in it. In my work, I emphasised engagement with people who interact with health systems.
I assume that when answering your question: "I would be interested to learn from your experience about the future role of AI in health communications”, you expect me to think beyond Malawi. I will try.
1) Health communication during consultations: AI can help support automatic interpretation in local languages. Speech-to-text to transcribe what the patient / doctor says + automatic translation to and from English / local languages. I do not think that we are anywhere close to developing a completely autonomous / competent AI conversational agent in Chichewa or many other languages spoken in Africa for at least two reasons: gaps in datasets and a lack of morphological tools.
2) Telemedicine and health-related chatbots / assistants: It could be argued that telemedicine is promoted more aggressively as a transformative solution in African countries than in European countries for example, being seen as a way to leapfrog infrastructure barriers. For these to be adopted effectively, one needs to have access to well-defined health vocabularies, that support the use of medical and health terms in context. How can AI help? At the moment, it can help at the foundational level (data and morphological tools) to achieve a similar linguistic progress as for English. For example, could Large Language Models (LLMs) support the development of foundational tools such as morphological analysers and parsers? Automatic AI tools for information extraction, entity recognition and annotation, can significantly speed up knowledge gathering, and structuring.
3) Medical records management and utilisation. One can think for example of (a) speech to text tools that document the medical conversation as it is taking place, (b) or tools that summarises the consultation, or (c) tools that highlight critical parts of the consultation transcript in relation to the medical history. Such uses of AI can only be possible if the current data and processes can support them. For English, and other languages, speech-to-text can be quite effective, but for languages such as Chichewa, more work needs to be done. Automatic Speech Recognition systems, a technology that converts spoken language into written text, is being experimented with, by using synthetic data for models training.
In Malawi, I also see a more direct application of AI as part of Optical Character Recognition (OCR) systems tailored to the unique challenges of medical record preservation in low-resource environments.
AI tools cannot be developed without the availability of good quality data and insights, and by rigurous experimentation. And all these come from us, the humans. One of the most challenging aspects of AI system is how to represent “insights” (or heuristics) and how to incorporate them in the reasoning used by the AI in a coherent way.
To make rigorous experimentation possible, better trust and collaborations between researchers is needed.
Kind regards
Amelia
HIFA profile: Amelia Taylor is a Lecturer in AI at Malawi University of Business and Applied Sciences. Interests: AI, NLP, Health Informatics, Data Visualisation. ataylor AT mubas.ac.mw