NTNU-SMIL at SemEval-2026 Task 3: Logistic-Loss Regression with Same-Language Transfer for Valence–Arousal Stance Prediction in Dimensional Stance Analysis (DimStance)

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

Summary

NTNU-SMIL's system for SemEval-2026 Task 3 Track B Subtask 1 Dimensional Stance Analysis (DimStance) models target-conditioned valence–arousal regression. The approach integrates sentence-pair encoding, dual regression heads, and a logistic-loss regression formulation. For English and Chinese, the system further incorporates same-language transfer from Track A, alongside lightweight out-of-fold calibration and multi-seed ensembling to reduce cross-lingual scale mismatch. Post-hoc analysis revealed that both same-language transfer and logistic-loss regression are primary drivers of performance gains. However, the system faces a challenge with arousal variance collapse in low-resource settings, exemplified by languages such as Swahili.

Key takeaway

For NLP Engineers developing dimensional stance analysis systems, you should prioritize integrating logistic-loss regression and same-language transfer to achieve robust valence-arousal prediction. Be aware that low-resource languages like Swahili may still exhibit arousal variance collapse, necessitating specific mitigation strategies. Consider multi-seed ensembling and out-of-fold calibration to manage cross-lingual scale mismatches effectively in multilingual deployments.

Key insights

Logistic-loss regression and same-language transfer significantly improve valence-arousal stance prediction, though low-resource languages face arousal variance collapse.

Principles

• Same-language transfer boosts performance.
• Logistic-loss regression drives performance gains.
• Arousal variance collapse challenges low-resource settings.

Method

The system models target-conditioned valence–arousal regression using sentence-pair encoding, dual regression heads, and a logistic-loss regression formulation, enhanced by same-language transfer and multi-seed ensembling.

In practice

• Apply same-language transfer for performance.
• Use logistic-loss regression for stance prediction.
• Employ multi-seed ensembling for calibration.

Topics

• Dimensional Stance Analysis
• Valence-Arousal Regression
• Logistic-Loss Regression
• Same-Language Transfer
• SemEval-2026
• Low-Resource Languages

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

Related on AIssential

• From Affect Prediction to Affect Forecasting: Evidence for Distinct Information Sources in Longitudinal Text — Current affect prediction relies on text semantics, while future affective change forecasting benefits from numeric trajectory dynamics.
• Low-resource Language Discrimination Towards Chinese Dialects with Transfer learning and Data Augmentation — Transfer learning and data augmentation effectively overcome scarce annotation resources for low-resource Chinese dialect discrimination.
• A Context-Aware Dataset for Stance Detection in Bioethical Controversies on Reddit — BioStance offers a large, context-aware dataset for stance detection in complex bioethical social media discourse.
• A Scalable Tool for Measuring Manner and Result Verbs in Developmental Language Research — A RoBERTa classifier, trained on LLM-generated annotations, accurately distinguishes manner and result verbs at scale.
• Beyond Sentiment Classification: A Generative Framework for Emotion Intensity Evaluation in Text — Generative LLMs fine-tuned on continuous emotion intensity scores outperform classification and generalize to unseen affective dimensions.
• I Pitted XGBoost Against Logistic Regression on 358 Matches. The Boring Model Won. — Model complexity must match data availability and structure to avoid overfitting and miscalibration.

Open in AIssential →

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