Nürnberg NLP at PsyDefDetect: Multi-Axis Voter Ensembles for Psychological Defence Mechanism Classification

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Social Sciences & Behavioral Studies · Depth: Expert, quick

Summary

The Nürnberg NLP system secured first place among 21 teams at the BioNLP 2026 PsyDefDetect shared task, achieving an F1 score of .420 on the hidden test set. This system addresses the challenging and ambiguous classification of psychological defence mechanisms in supportive conversations, where eight positive defence categories share surface language and exhibit only moderate inter-annotator agreement. Recognizing that error independence is more crucial than a stronger single model for such tasks, the team developed a 9-voter ensemble. This ensemble spans three orthogonal axes: class granularity, differentiating between all nine classes for a gatekeeper and eight defence classes for specialists; training method, incorporating both generative and discriminative approaches; and diverse base models.

Key takeaway

For NLP engineers tackling highly ambiguous classification tasks, particularly those with overlapping categories like psychological defence mechanisms, your focus should shift from optimizing a single model to building robust ensembles. You should explore multi-axis voter ensembles that leverage error independence across diverse dimensions. Consider varying class granularity, integrating both generative and discriminative training methods, and employing different base models to achieve superior performance where individual models struggle.

Key insights

Error independence in multi-voter ensembles is key for ambiguous classification tasks with overlapping categories, outperforming single stronger models.

Principles

• Error independence is crucial for ambiguous classification.
• Ensembles spanning orthogonal axes enhance robustness.

Method

A 9-voter ensemble combines models across three orthogonal axes: class granularity (9 vs. 8 classes), training method (generative/discriminative), and base model diversity.

In practice

• Apply multi-axis ensembles to ambiguous NLP tasks.
• Vary class granularity and training methods in voters.

Topics

• Psychological Defence Mechanisms
• Ensemble Learning
• Multi-axis Ensembles
• NLP Classification
• BioNLP Shared Task
• Error Independence

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• From TF-IDF to Transformers: A Comparative and Ensemble Approach to Sentiment Classification — A soft voting ensemble of diverse machine learning and transformer models significantly enhances sentiment classification accuracy, particularly for nuanced text.
• LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification — Iterative imbalance-aware fine-tuning of LLMs, combined with augmentation and post-processing, significantly improves rare-class NLP performance.
• Implicit and Indirect: Detecting Face-threatening and Paired Actions in Asynchronous Online Conversations — Detecting face-threatening actions in asynchronous conversations requires handling ambiguity and multiple interpretations.
• Decorrelation, Diversity, and Emergent Intelligence: The Isomorphism Between Social Insect Colonies and Ensemble Machine Learning — Ant colonies and random forests are mathematically isomorphic, both achieving collective intelligence via decorrelated stochastic units.
• Bootstrapping-based Regularisation for Reducing Individual Prediction Instability in Clinical Risk Prediction Models — A novel regularisation method improves clinical prediction model stability without sacrificing interpretability, crucial for healthcare adoption.
• Learning Perspectivist Social Meaning via Demographic-Conditioned Fusion Embeddings — Social meaning is perspectival; models can predict how interpretations vary across demographic groups using fusion embeddings.

Open in AIssential →

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