Vinland_Vector at #SMM4H-HeaRD 2026: Multilingual ADE Detection and Query-Augmented Clinical NER for English

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

Summary

Vinland_Vector's submission to SMM4H-HeaRD 2026 addresses two critical natural language processing tasks: multilingual adverse drug event (ADE) detection and clinical named entity recognition (NER) for English. For ADE detection (Task 1), the team formulated it as a binary classification problem across social media posts in German, French, Russian, English, Mandarin, and Japanese, with zero-shot capability for Farsi. Their XLM-RoBERTa-Large model, enhanced with a dual-pooling head, stratified sampling, language-conditioned inputs, translation-based augmentation, and calibrated ensembling, achieved a macro F1 score of 0.6088, surpassing the competition's mean (0.5465) and median (0.5798). For MultiClinNER (Task 8), targeting English clinical text, they utilized GLiNER-large with sliding-window inference, query augmentation, and calibrated thresholds. This approach yielded strict F1 scores of 0.7591 for Disease, 0.7263 for Procedure, and 0.6733 for Symptom, outperforming a PubMedBERT baseline across all entity types.

Key takeaway

For NLP Engineers developing multilingual adverse drug event (ADE) detection or clinical named entity recognition (NER) systems, this work demonstrates effective strategies. You should consider integrating language-conditioned inputs and translation-based augmentation with XLM-RoBERTa-Large for robust multilingual classification. For English clinical NER, adopting GLiNER-large with query augmentation and calibrated thresholds can significantly improve F1 scores for entities like Disease, Procedure, and Symptom, surpassing traditional baselines.

Key insights

Combining XLM-RoBERTa-Large with ensembling and GLiNER-large with query augmentation significantly improves clinical NLP tasks.

Principles

• Multilingual NLP benefits from language-conditioned inputs.
• Translation-based augmentation enhances cross-lingual performance.
• Query augmentation improves clinical NER F1 scores.

Method

Method for ADE detection involves XLM-RoBERTa-Large with dual-pooling, stratified sampling, language-conditioned inputs, translation augmentation, and calibrated ensembling.

In practice

• Apply XLM-RoBERTa-Large for multilingual text classification.
• Use GLiNER-large with query augmentation for clinical NER.
• Implement calibrated ensembling for robust model predictions.

Topics

• Adverse Drug Event Detection
• Clinical Named Entity Recognition
• Multilingual NLP
• XLM-RoBERTa-Large
• GLiNER-large
• Query Augmentation
• Ensembling

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

Related on AIssential

• 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.
• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• 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.
• KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts — Domain-specific pre-training significantly enhances BERT models for Norwegian clinical NLP tasks.
• Predicting relations between SOAP note sections: The value of incorporating a clinical information model — Fine-tuning a RoBERTa-base model on manually annotated clinical data improves NER for SOAP note sections.
• Intro to NLP with spaCy (4): Detecting programming languages — Training a custom spaCy Named Entity Recognition (NER) model automates programming language detection by learning from labeled data.

Open in AIssential →

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