Analyzing Prompt Design Choices in Biomedical Information Extraction for Low-Resource Languages

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, medium

Summary

A study presented at BioNLP 2026 by Khatun et al. investigates optimizing biomedical named entity recognition (NER) using large language models (LLMs), particularly for low-resource languages such as Bangla and Basque. The research, published on pages 31–44, reveals that prompt design significantly impacts model performance. Question-style prompting emerged as the most effective method across all tested languages, leading to substantial accuracy improvements, especially in Bangla and Basque compared to high-resource languages like English and Spanish. Furthermore, the study highlights the critical role of output format, demonstrating that span-based extraction, which directly extracts text phrases, outperforms traditional BIO tagging by yielding higher F1 scores. This is attributed to LLMs' natural text generation capabilities. Error analysis identified common issues like hallucination and missing entities, noting that translation-based prompts can mitigate hallucination, while question-style prompts reduce empty outputs in biomedical NER tasks.

Key takeaway

For NLP Engineers developing biomedical information extraction systems, especially for low-resource languages, you should prioritize question-style prompting. This approach significantly boosts accuracy and reduces empty outputs. Furthermore, adopt span-based extraction for your output format, as it aligns better with LLM generation capabilities and yields higher F1 scores than traditional BIO tagging. Consider translation-based prompts to mitigate hallucination risks in your models.

Key insights

Question-style prompts and span-based extraction significantly enhance biomedical NER performance in LLMs, especially for low-resource languages.

Principles

• Prompt design critically impacts LLM NER accuracy.
• Question-style prompts clarify LLM task understanding.
• Span-based extraction aligns with LLM text generation.

Method

The study systematically compared various prompt styles and output formats, including question-style prompting and span-based extraction versus BIO tagging, across multiple languages for biomedical NER.

In practice

• Implement question-style prompts for NER tasks.
• Prioritize span-based output over BIO tagging.
• Use translation prompts to mitigate hallucination.

Topics

• Biomedical NER
• Large Language Models
• Prompt Engineering
• Low-Resource NLP
• Span-based Extraction
• Information Extraction

Best for: AI Engineer, Machine Learning Engineer, AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning? — LLMs offer a viable few-shot alternative for biomedical relation extraction, outperforming prior few-shot methods, especially for rare relations.
• Automatic Generation of Highlights for Academic Paper Via Prompt-based Learning — Prompt-based learning with LLMs can generate academic paper highlights effectively without extensive labeled data.
• Prompt Engineering for Named Entity Extraction from Portuguese Legal Documents — Prompt engineering with LLMs offers a viable alternative for legal NER in low-resource settings.
• Analysing Lightweight Large Language Models for Biomedical Named Entity Recognition on Diverse Ouput Formats — Lightweight LLMs offer competitive biomedical NER performance, with specific output formats proving more effective than multi-format tuning.
• Measuring the sensitivity of LLM-based structured extraction to prompt, model, and schema choices in clinical discharge summaries — LLM-based clinical data extraction sensitivity varies significantly with model and schema, especially regarding "absence-versus-silence" distinctions.
• Can AI Save India’s Dying Languages? | Building a Tulu LLM Without Training Data Ft. Prathamesh — Structured prompting and custom romanization enable LLMs to generate low-resource languages without extensive training data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.