Learning Perspectivist Social Meaning via Demographic-Conditioned Fusion Embeddings

2026-06-05

AI Summary

This work addresses the perspective-dependent nature of social meaning in natural language, which conventional NLP systems often oversimplify by assigning a single label while disregarding interpretive differences across demographic groups. To systematically model this diversity of perspectives, the authors propose a demographic-conditioned fusion embedding approach that jointly incorporates textual and demographic features. The method is evaluated across zero-shot, few-shot, and fine-tuning learning paradigms on a dataset of 28k human annotations. Experimental results demonstrate that the proposed fusion model consistently and significantly outperforms text-only baselines under all learning strategies, achieving a relative improvement of 5.9%–6.5% in macro-averaged PR-AUC. Ablation studies with randomized demographic labels confirm the robustness of these gains, verifying that demographic attributes contribute genuine predictive signals rather than spurious correlations.

Abstract

Social meaning in language is inherently perspectival, varying across annotator backgrounds, demographics, and ideological positions. However, most NLP systems collapse this variation into a single ground-truth label, ignoring the diversity of interpretations. In this work, we model social dimensions along a perspectivist spectrum, capturing how interpretations vary across demographic groups on a dataset consisting of 28k human annotations. We benchmark multiple modeling paradigms, including zero-shot, few-shot, and fine-tuned approaches, and propose fusion embeddings that integrate textual and demographic representations. Our fusion models yield consistent and statistically significant improvements over text-only baselines across all fusion strategies (+5.9-6.5% relative macro PR-AUC), with shuffle ablations confirming that demographic profiles carry genuine predictive signal rather than spurious correlations.

Problem

• social meaning
• perspectivism
• demographic variation
• annotation diversity
• NLP

Innovation

• perspectivist modeling
• demographic-conditioned fusion
• social meaning
• annotation diversity
• fusion embeddings