Exploring automatic terminology extraction from historical medical data

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Health & Medical Research · Depth: Advanced, quick

Summary

This study evaluates various automatic terminology extraction methods applied to historical medical texts with non-standard orthography. Researchers tested two linguistic pattern-based methods, four prompt-based Generative AI models, and one BERT-like model, some of which were fine-tuned for terminology extraction, including one specialized in Portuguese medical terms. Four distinct prompting strategies were employed for the GenAI models. The test data comprised chapter fifteen of "Aviso 'a Gente do Mar sobre a sua Saude," a 1794 Portuguese translation of an 18th-century French medical text, manually annotated for terminology. Evaluation focused on f-measure and pure precision to assess how automatic methods could augment manual token-based annotation. Results indicate that combining multiple automatic extraction methods can significantly enhance coverage, achieving over 90% recall on the test data, despite individual models showing limited extraction quality.

Key takeaway

For NLP Engineers working with historical or orthographically inconsistent datasets, consider a multi-model approach to terminology extraction. Your current single-model solutions may underperform, but combining linguistic pattern methods, fine-tuned BERT-like models, and prompt-engineered Generative AI can achieve over 90% recall, significantly improving data coverage for downstream tasks.

Key insights

Combining multiple automatic methods significantly improves terminology extraction from historical texts with non-standard orthography.

Principles

• Historical texts challenge modern NLP.
• Model combination boosts recall.
• Prompting strategies impact GenAI.

Method

The study tested linguistic patterns, fine-tuned BERT-like models, and prompt-based GenAI with four strategies on 18th-century Portuguese medical texts, evaluating f-measure and precision against manual annotations.

In practice

• Combine extractors for higher recall.
• Fine-tune models for specific domains.
• Experiment with GenAI prompting.

Topics

• Automatic Terminology Extraction
• Historical Medical Data
• Generative AI Models
• BERT-like Models
• Linguistic Pattern Methods

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Data Augmentation for Historical NER: A Systematic Comparison of Lexical and LLM-based Approaches — This paper systematically compares lexical and LLM-based data augmentation for Historical Named Entity Recognition.
• Severity-Aware Curriculum Learning with Multi-Model Response Selection for Medical Text Generation — Severity-aware curriculum learning with multi-model response selection significantly improves medical text generation quality and relevance.
• KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts — Domain-specific pre-training significantly enhances BERT models for Norwegian clinical NLP tasks.
• Closing the Last Mile in Document AI: Improving Extraction Quality in Azure Content Understanding — Labeled examples significantly boost document field extraction accuracy in generative AI systems like Azure Content Understanding.
• NLP Landscape — NLP's evolution from rule-based to deep learning reflects increasing automation in feature extraction and pattern recognition.
• 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.

Open in AIssential →

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