Work

A mini-review of medical question answering for Bangla: the methodological landscape and cross-lingual perspectives

Year: 2026

Type: review

Abstract: Large language models (LLMs) have advanced medical question answering (QA) significantly, yet low-resource languages such as Bangla remain critically underserved. This review examines the current landscape of LLM-based medical QA with a primary focus on Bangla, analyzing 15 representative studies (2022-2026), including English and multilingual works reviewed to contextualize findings for Bangla. We examine methodological approaches, datasets, and evaluation frameworks across diverse paradigms, including transformer architectures, retrieval-augmented generation (RAG), prompt engineering, and parameter-efficient fine-tuning, to identify pathways toward reliable and linguistically inclusive medical AI. The findings reveal a persistent performance gap driven by large-scale domain-specific pretraining and standardized benchmarks in English. Persistent challenges include dataset scarcity, heavy reliance on translated corpora, inconsistent evaluation practices, and a lack of clinician-validated benchmarks. This review identifies these gaps and outlines a structured research roadmap emphasizing native-language corpus development, multimodal data integration, standardized benchmarking, and bias auditing as essential steps toward building reliable and linguistically inclusive medical QA systems for Bangla healthcare.

Source: Frontiers in Digital Health

DOI: 10.3389/fdgth.2026.1856734