ICB-UMA at #SMM4H–HeaRD 2026: Hybrid Clinical Entity Projection for MultiClinAI: Adaptive Candidate Windows, XGBoost, and LLM Refinement

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, medium

Summary

The ICB-UMA team submitted a hybrid system to the MultiClinAI Shared Task (Gallego-Donoso et al., 2026) for cross-lingual clinical entity annotation projection, specifically transferring expert annotations for Diseases, Symptoms, and Procedures from Spanish to English. Their approach combines three core components: adaptive candidate window generation, an XGBoost classifier that utilizes both surface and semantic features, and a Large Language Model (LLM)-based post-processing stage designed to resolve complex misalignments. This system achieved a 3rd place ranking on the official leaderboard, with strict F1 scores of 0.737 for Diseases, 0.549 for Symptoms, and 0.538 for Procedures. These results highlight the effectiveness of integrating supervised candidate scoring with targeted LLM refinement for robust clinical entity projection.

Key takeaway

For NLP Engineers developing cross-lingual clinical entity extraction systems, consider integrating a hybrid approach that combines traditional supervised learning with Large Language Model (LLM) refinement. You should implement adaptive candidate window generation and an XGBoost classifier for initial scoring, then utilize LLMs for targeted post-processing to resolve complex misalignments. This strategy, demonstrated by the 3rd place MultiClinAI system, can significantly improve the accuracy of transferring expert annotations for entities like Diseases and Symptoms across languages.

Key insights

Hybridizing supervised candidate scoring with LLM refinement robustly projects cross-lingual clinical entities.

Principles

• Hybrid approaches improve cross-lingual entity projection.
• LLM refinement resolves complex misalignments.
• Supervised scoring enhances candidate selection.

Method

The method involves adaptive candidate window generation, an XGBoost classifier utilizing surface and semantic features, and an LLM-based post-processing stage for misalignment resolution.

In practice

• Apply XGBoost for feature-rich candidate scoring.
• Use LLMs for post-processing complex entity misalignments.
• Generate adaptive windows for entity candidates.

Topics

• Cross-lingual NLP
• Clinical Entity Recognition
• Large Language Models
• XGBoost
• MultiClinAI Shared Task
• Annotation Projection

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Cross-Lingual Empirical Evaluation of Large Language Models for Arabic Medical Tasks — English-centric LLMs exhibit significant performance gaps and reliability issues in Arabic medical tasks.
• LLMs for automatic annotation of Mandarin narrative transcripts — LLMs can automate complex discourse-level annotation in non-English speech, significantly reducing time.
• Beyond Bilingual Transfer: Multilingual Code-Switching in Instruction Tuning — Multilingual code-switching instruction tuning improves LLM performance beyond bilingual settings.
• EDEN: A Large-Scale Corpus of Clinical Notes for Italian — EDEN provides the largest Italian clinical note corpus and a CRF-filling benchmark for LLM-based information extraction.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• Authorship Attribution in Multilingual Machine-Generated Texts — Multilingual authorship attribution for LLM-generated text is complex, with current monolingual methods showing limited cross-lingual transferability across diverse language families.

Open in AIssential →

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