In this study, we introduce a novel benchmark comprising over 9,000 real-world, point-of-care, multilingual, and multimodal clinical question-answer pairs sourced from frontline health workers in Nigeria. Using the dataset, we compare local general practitioners to multiple leading open and closed LLMs. Our results reveal several critical insights into the suitability of LLMs as clinical decision support systems in low-resource contexts. The results confirm that performance varies widely by language and input modality (e.g., text vs speech): while models perform best on English text inputs, their accuracy drops significantly for local-language speech. Critically, it is possible to achieve substantial performance gains by transcribing and translating other languages into English before prompting an LLM-- an important insight for non-anglophone product developers. Finally, this benchmark highlights key limitations of SLMs in supporting frontline healthcare in low-resource settings and provides a clear opportunity to track improvements as novel solutions are developed.
Tobi Olatunji, C. Aka, C. Okocha et al.· medRxiv· 0 citations
Safety alignment in large language models (LLMs) is largely developed in English, assuming these safeguards generalize across multilingual settings. However, this assumption remains underexplored and exposes a vulnerability in low-resource languages. We investigate cross-lingual safety transfer in four African languages, Twi, Hausa, Amharic, and Swahili, using LoDNA, a new safety dataset that pairs literal translations with culturally localized prompts. To move beyond generation-based evaluation, we propose a latent geometric framework that probes hidden-state refusal representations in LLMs. Our experimental results show that cross-lingual safety transfer is severely limited; harmful prompts retain less than 10% of the English refusal signal across most language-model pairs. Literal and localized prompts are semantically aligned (cosine 0.95-0.996) but drift across layers, suggesting models encode the concepts without routing them to safety mechanisms. These findings demonstrate that current multilingual safety alignment is superficial, providing strong evidence against the assumption of a universal, language-agnostic harm manifold within the specific low-resource languages studied. Warning: This paper contains example data that may be offensive or harmful.
Abigail Oppong, P SAM SAHIL, Tadesse Destaw Belay et al.· 0 citations