DimStance: Multilingual Datasets for Dimensional Stance Analysis

2026-01-29

AI Summary

This work addresses the limitation of traditional stance detection, which relies on discrete labels (e.g., support/oppose/neutral) and fails to capture the fine-grained affective nuances in stance expression. To overcome this, we introduce the valence–arousal (VA) continuous dimensional model from affective science into stance analysis, presenting DimStance—the first multilingual dimensional stance dataset spanning five languages and two domains—and formalizing a dimensional stance regression task. We conduct cross-lingual VA prediction experiments using both pre-trained language models and large language models under regression and prompting paradigms. Results demonstrate that fine-tuned large language models achieve strong performance on dimensional regression, though low-resource languages remain challenging. This study establishes a foundation for emotion-aware, fine-grained, and cross-lingual stance analysis.

Technology Category

• Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument Mining
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Stance detection is an established task that classifies an author's attitude toward a specific target into categories such as Favor, Neutral, and Against. Beyond categorical stance labels, we leverage a long-established affective science framework to model stance along real-valued dimensions of valence (negative-positive) and arousal (calm-active). This dimensional approach captures nuanced affective states underlying stance expressions, enabling fine-grained stance analysis. To this end, we introduce DimStance, the first dimensional stance resource with valence-arousal (VA) annotations. This resource comprises 11,746 target aspects in 7,365 texts across five languages (English, German, Chinese, Nigerian Pidgin, and Swahili) and two domains (politics and environmental protection). To facilitate the evaluation of stance VA prediction, we formulate the dimensional stance regression task, analyze cross-lingual VA patterns, and benchmark pretrained and large language models under regression and prompting settings. Results show competitive performance of fine-tuned LLM regressors, persistent challenges in low-resource languages, and limitations of token-based generation. DimStance provides a foundation for multilingual, emotion-aware, stance analysis and benchmarking.

Problem

• stance detection
• dimensional stance
• valence-arousal
• multilingual
• fine-grained analysis

Innovation

• dimensional stance analysis
• valence-arousal modeling
• multilingual stance dataset
• fine-grained affective modeling
• cross-lingual stance regression