Music Therapy is not Relaxation

Semantic Isomorphism and Musical Agency in Music Therapy with Elderly People

Abstract

Research Question/Problem

There is a common perception that music therapy acts primarily as a tool for passive relaxation or as an adjunct to the treatment of mental disorders (Bruscia, 2016). However, in geriatric clinical practice, music therapy often demands intense cognitive and motor engagement (Thaut & Hoemberg, 2014). This study investigates whether the structure of musical cognition in the elderly, measured by the Cognitive Assessment of the Elderly in Music Therapy (ACPIM-5; Pedrosa & Arruda, 2026), is encoded in language as a psychopathological form or as an autonomous domain of musical action. The central problem lies in verifying whether Artificial Intelligence (AI) models can predict how humans respond to music therapy scales and whether this prediction depends on a semantics of symptoms or on intentional musicking.

Objectives

The objective of this study was to perform the pre-empirical calibration of the ACPIM-5 instrument via Generative Psychometrics, testing the isomorphism between the semantic space of Large Language Models (LLMs) and the covariation of human responses, identifying the logical pillars that sustain musical cognition in senescence.

Method

A comparative Generative Psychometrics protocol was used, extracting embeddings from the Cohere v3.0 architecture for the items of ACPIM-5, Musical Coping scales (ECOM; Pimentel et al., 2012), and Psychopathology scales: Beck Depression Inventory (BDI; Beck et al., 1996), Beck Anxiety Inventory (BAI; Beck et al., 1988), and the Depression, Anxiety, Stress Scale (DASS-21; Lovibond & Lovibond, 1995). The training of the Elastic Net model was carried out in two stages: (1) training on a robust database regarding psychopathological variables (N=1,838 Brazilian participants) and (2) training on a musical domain database (N=457). The validity of the transfer to the music therapy cognition scale for elderly people was assessed via Permutation Tests (5,000 iterations) to ensure that the results were not random.

Results and/or Main contributions

The results revealed that the model trained on 1,838 cases of depression and anxiety failed to predict the music therapy structure within the elderly population (), providing evidence that “musical doing” in music therapy does not share the same grammar as psychopathology. In contrast, the model trained on Musical Coping demonstrated a robust isomorphism with the ACPIM-5 (; ). Explainable AI (XAI) analysis indicated that dimensions of Passive Relaxation and Stress Reduction presented massive negative weights in the prediction, while the Brazilian regional term “Desopilar” (Cathartic Activation) and dimensions of Motivation and Action were the stable positive anchors ().

Conclusions and/or Implications

The research concludes that music therapy, from the perspective of embodied cognition, is defined by Agency rather than relaxation. AI revealed that the structure of cognition in elderly people within music therapy is an activation structure. These findings imply that interventions and assessment tests in music therapy should be calibrated by active engagement domains, as the semantics of music in the human mind pertains to intentional musical action.