Are Expressions for Music Emotions the Same Across Cultures?

TL;DR

This study investigates whether music emotion terms generalize across languages by addressing bias in stimulus selection and Western-centric taxonomies. It introduces a data-driven, bottom-up pipeline (STEP) that generates culture-specific emotion taxonomies from open-ended tagging across Brazil, Korea, and the US, followed by dense ratings of a shared song corpus. The results reveal two robust clusters corresponding to high arousal/high valence emotions across cultures, while cross-cultural term mappings show meaningful alignments but also notable translation-driven divergences, underscoring the limits of dictionary translations for music emotion semantics. Additionally, an in-group effect indicates higher agreement when raters assess music from their own culture, highlighting the role of familiarity in cross-cultural emotion perception. Overall, the approach reduces cultural biases and offers a scalable framework for cross-cultural emotion research with applications in music recommendation and multi-domain emotion studies.

Abstract

Music evokes profound emotions, yet the universality of emotional descriptors across languages remains debated. A key challenge in cross-cultural research on music emotion is biased stimulus selection and manual curation of taxonomies, predominantly relying on Western music and languages. To address this, we propose a balanced experimental design with nine online experiments in Brazil, the US, and South Korea, involving N=672 participants. First, we sample a balanced set of popular music from these countries. Using an open-ended tagging pipeline, we then gather emotion terms to create culture-specific taxonomies. Finally, using these bottom-up taxonomies, participants rate emotions of each song. This allows us to map emotional similarities within and across cultures. Results show consistency in high arousal, high valence emotions but greater variability in others. Notably, machine translations were often inadequate to capture music-specific meanings. These findings together highlight the need for a domain-sensitive, open-ended, bottom-up emotion elicitation approach to reduce cultural biases in emotion research.

Figures (4)

• Figure 1: Experimental Setup. A: Music collection in three cultures. B: Human-in-the-loop pipeline conducted in the following order: (C) STEP paradigm to obtain a list of words per culture, (D) discrete emotionality rating to select emotional words, and (E) dense rating of each stimulus along the select terms.
• Figure 2: Pearson correlation across 50 emotional terms in Portuguese (Brazil, A), Korean (Korea, B), and English (US C). Correlation matrices are sorted by the order in the dendrogram obtained via agglomerative clustering.
• Figure 3: Pearson correlation across all three pairs of cultures: (A) Brazil and Korea, (B) Korea and US, and (C) US and Brazil. Correlation matrices are sorted by the order in the dendrogram obtained via agglomerative clustering.
• Figure 4: Network analysis on the full correlation matrix. Negative correlations are removed. Korean terms are in bold, Portuguese in italics, and English in plain font. The coloring is based on the modularity.