Our overview paper for the Shared Task on Patient-Centric Question Answering at the second NLP-AI4Health workshop (IJCNLP, Mumbai, December 2025) is now in the ACL Anthology.
The task targets inclusive NLP systems for two clinical domains — Head and Neck Cancer (HNC) and Cystic Fibrosis (CF) — and introduces the NLP4Health-2025 Dataset: more than 45,000 validated multi-turn patient–provider dialogues across 10 languages (Assamese, Bangla, Dogri, English, Gujarati, Hindi, Kannada, Marathi, Tamil, and Telugu).
Participants built lightweight models (< 3B parameters) for:
- Clinical summarization — abstractive summaries and structured clinical extraction (SCE)
- Patient-centric QA — empathetic, factually accurate answers in the dialogue’s native language
The paper covers the hybrid human–agent dataset pipeline, task definitions, evaluation metrics, and results from 9 submissions by 6 teams, showing that small language models can work in low-resource medical settings when tuned with approaches such as LoRA and RAG.
Authors: Arun Zechariah, Balu Krishna, Hannah Mary Thomas, Joy Mammen, Dipti Misra Sharma, Parameswari Krishnamurthy, Vandan Mujadia, Priyanka Dasari, and Vishnuraj Arjunaswamy.