Enigma at #SMM4H–HeaRD 2026: Leveraging Multilingual Pre-trained Models for Clinical Named Entity Recognition

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

Enigma participated in the MultiClinAI challenge's MultiClinNER subtask, focusing on clinical Named Entity Recognition across seven languages: Czech, Dutch, English, Italian, Romanian, Spanish, and Swedish. The primary objective was to identify and extract clinical terms related to diseases, procedures, and symptoms from discharge summaries. The team explored various methods, including monolingual and multilingual pre-trained, zero-shot, domain-adapted, and fine-tuned transformer models, alongside ensemble modeling. Data augmentation using external resources significantly improved the models' ability to recognize clinical entities. Both monolingual and multilingual approaches demonstrated complementary strengths depending on the language and entity type, achieving an average F1 score of 0.6502 across the best models.

Key takeaway

For NLP Engineers developing clinical NER systems across multiple languages, you should consider integrating ensemble modeling with diverse transformer architectures. Crucially, prioritize data augmentation using external resources, as this significantly enhances entity recognition performance. Evaluate both monolingual and multilingual approaches, as their complementary strengths can optimize results for specific languages and entity types in your deployment.

Key insights

Leveraging multilingual pre-trained models and data augmentation enhances clinical Named Entity Recognition across diverse languages.

Principles

• Data augmentation significantly improves entity recognition.
• Monolingual and multilingual models offer complementary strengths.

Method

Explored pre-trained, zero-shot, domain-adapted, and fine-tuned transformer models, combined with ensemble modeling and data augmentation from external resources for clinical NER.

In practice

• Identify diseases, procedures, symptoms.
• Apply data augmentation for clinical NER.

Topics

• Clinical NER
• Multilingual Models
• Transformer Models
• Data Augmentation
• Ensemble Modeling
• Discharge Summaries
• MultiClinAI Challenge

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• 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.
• Data Augmentation for Named Entity Recognition in Domain-Specific Scenarios in Portuguese — Data augmentation significantly improves NER performance in low-resource, domain-specific Portuguese datasets.
• Linguistic Symptoms: Augmented Generation by Retrieval and Reasoning in LLMs under Portuguese-English Variation in Medical Contexts — Linguistic variation significantly impacts smaller LLMs in medical contexts, with larger models showing greater cross-language robustness.
• From BERT to T5: A Study of Named Entity Recognition — BERT and T5 models were fine-tuned and compared for Named Entity Recognition using different architectural approaches.
• Label-Aware Pseudo-Training Sample Generation for Text Classification — A new data augmentation method uses LLMs to subtly alter sentences, improving NLP performance with limited data.
• KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts — Domain-specific pre-training significantly enhances BERT models for Norwegian clinical NLP tasks.

Open in AIssential →

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