Sonification of language networks: Acoustic discrimination of linguistic families from the Americas

Zúñiga J.V. · EPL (2026) · DOI: 10.1209/0295-5075/ae73b1

Resumen

The heat equation, long used to probe graph geometry, is used here to test whether typological information encoded in language-network spectra survives transformation into the acoustic domain. Building on the paper by Vera Zúñiga Javier, EPL, 150 (2025) 22001, we sonify Laplacian spectra of co-occurrence graphs for 44 indigenous American languages from 5 families and extract MFCC features from 10 sonification strategies varying mapping, resolution, and perceptual scale. The direct spectral baseline yields Silhouette +0.143, inter/intra ratio 2.43, and classification accuracy (Formula presented). In the acoustic domain, all strategies retain positive Silhouette scores. Heat-trace sonification gives the strongest acoustic clustering (Silhouette (Formula presented) ), whereas the square-root mapping yields the highest classification accuracy ( (Formula presented) ; macro-F1 (Formula presented) ). These results show that family-discriminative information remains recoverable from sonified spectra, supporting sonification as an alternative representational modality rather than evidence of perceptual audibility. Copyright

Instituciones: Universidad Tecnológica del Perú