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

2026-05-08

AI Summary

This study addresses the challenge of identifying psychological defense mechanism categories in supportive dialogues, which is hindered by semantic ambiguity and low inter-annotator agreement due to superficial linguistic similarity despite divergent pragmatic functions. To tackle this issue, the authors propose a multi-axis voting ensemble method that constructs nine heterogeneous models along three orthogonal dimensions: classification granularity, training paradigm (generative vs. discriminative), and base architecture. Rather than optimizing individual model performance, the approach prioritizes error independence among ensemble members to effectively manage ambiguous category boundaries. Evaluated on the hidden test set of the PsyDefDetect shared task, the method achieves an F1 score of 0.420, ranking first among 21 participating teams and demonstrating substantially improved discriminative capability for ambiguous psychological defense categories.

Abstract

Detecting levels of psychological defence mechanisms in supportive conversations is inherently ambiguous. In the PsyDefDetect shared task at BioNLP 2026 the eight positive defence categories share surface language and differ only in pragmatic function and trained raters reach only moderate inter-annotator agreement. On such a task the decisive lever is not a stronger single model but error independence, since any single representation will waver on the overlapping defence boundaries. We translate this insight into a 9-voter ensemble spanning three orthogonal axes: class granularity (all nine classes for the gatekeeper, only the eight defence classes for the specialists), training method (generative and discriminative) and base model. The system reaches $F1_{test}{=}.420$ on the hidden test set, placing first among 21 registered teams.

Problem

• psychological defence mechanisms
• supportive conversations
• inter-annotator agreement
• defence classification
• pragmatic function

Innovation

• multi-axis ensemble
• psychological defence mechanism classification
• error independence
• orthogonal model diversity
• generative-discriminative hybrid