Towards Culturally Grounded Emotional AI via a Community-Centred Approach

Abstract

This study presents a community-centred methodology for developing culturally grounded emotional AI systems for low-resource and Indigenous languages. While the work focuses on emotion recognition for te reo Māori, the Indigenous language of New Zealand, the approach is generalisable to other Indigenous languages. It examines the assumption of universal emotion categories used in current emotional AI systems by demonstrating that emotions may be shaped by cultural context. Instead of relying on predefined Western emotion categories, this work derives a culturally grounded set of emotions through community feedback. It also introduces a new emotional speech corpus and analyses how emotional expression varies acoustically across cultures, contributing to the development of more inclusive and culturally appropriate emotional AI systems. More broadly, this work highlights the need to reconsider who is represented in AI and how fundamental design assumptions can impact fairness in emotional AI.