Creativity ≠ Generativity: A Case Study of Attentive Machine Learning in Dance Performance 33

Dance expresses through translation. It transforms intention, memory, and affect into embodied movement through tacit and relational knowledge. When computational systems enter this space, they introduce new translation layers that render movement legible to machines and mediate how performance is perceived and remembered. This paper presents a case study of human and machine co-performance as an attentive practice of translation through a bespoke movement recognition system embedded within a twenty-minute multimedia dance theater work. Using wearable IMU sensors and a lightweight time series classifier based on MiniRocket (MINImally RandOm Convolutional Kernel Transform), the system translates performers’ movements into real time audiovisual responses with low latency suitable for live performance. The model is performer-trained and non-generative, where mappings are predefined, but their timing remains open, allowing moments of unpredictability to emerge as a generative force within performance. Through practice-based research and reflection on moments that resisted translation, we argue that human-machine systems in the arts should be evaluated not by accuracy alone, but also by how well they preserve agency, care, and artistic intention.