Animal Vocal Communication as a Structured System: A Bioacoustic, Veterinary and Computational Perspective

Abstract. Artificial intelligence (AI) is increasingly used to analyse animal vocalisations, and public discussion often frames the goal as "decoding" or "translating" animal language. This framing risks assuming that animal vocal communication is equivalent to human language. This Perspective argues that vocalisations are more productively studied as structured communication systems whose acoustic form, context, sequence and receiver responses can be measured and tested without presupposing language. We synthesise evidence on physiological constraints, emotion-related variation, context-specific and combinatorial signals, individual address and receiver dependence, and we distinguish what machine learning can establish (reproducible acoustic patterns, predictive associations) from claims that require behavioural or clinical validation. We propose "bioacoustic logic" as an operational framework with six graded criteria, a reporting checklist for AI-based veterinary vocal analysis that complements STARD and TRIPOD+AI, and a controlled design for testing informal clinical observations of receiver-dependent behaviour. A simulation illustrates how data leakage, confounding and receiver shift can inflate or erase apparent performance.