Maternal Gait as a Potential Scaffold for Rhythm Processing: A Predictive Processing Network Model.

Humans uniquely perceive periodic structure in complex rhythms and spontaneously move to music-abilities rare among animals. Though training and experience contribute to our sense of rhythm, basic beat perception and the urge to move to rhythm are present even in young infants. But might prenatal experience play an essential role in shaping these faculties? We propose that maternal gait during pregnancy may provide scaffolding for rhythm development through correlated auditory-vestibular inputs that train predictive neural circuits. We implemented a recurrent predictive coding network with parallel vestibular and auditory sensory pathways, trained via Hebbian learning to minimize prediction error. Training paired discrete auditory pulses with continuous triangular vestibular waveforms mimicking maternal locomotion. These networks learned to anticipate a periodic auditory pulse in the absence of vestibular input. Continuous vestibular input greatly facilitated learning an appropriate resonance in the network, while temporally coordinated auditory input was necessary to give the auditory channel access to that resonance for auditory prediction. Auditory anticipation was accompanied by vestibular predictions, which, under the principles of active inference, are expected to produce spontaneous periodic movement. This work illustrates how simple sensory correlations during development can give rise to complex musical behaviors.