Neuromechanics of learned sensorimotor vocal integration
How sequential, rapid vocal production is represented in the cortex, and how representations change across cortical regions remains poorly understood. This gap in knowledge complicates development of an adequate cortical model for speech and language production which could help to explain speech and language pathologies. Here, a biomechanical perspective is taken to develop a mathematical dynamical systems model of the zebra finch syrinx and upper vocal tract. The model successfully introduces simplifications in describing the complex singing behavior, identifying a sequence of elemental gestures that comprise the bird's song. Gestures are small vocal movements, coordinated changes in pressure (varying the amount of air going through the syrinx) and changes in tension (of the syringeal membrane) as a function of time as a bird sings. Examining neural correlates of the time-varying features of pressure and tension in a cortical song system area HVC has identified that the activity of HVC neurons encode significant moments during movements (extreme or maximal points in the gesture trajectories). This indicates that HVC has an internal model of vocal behavior, represented by sequences of gestures. The timing of neuronal activity is precisely associated with near zero time lag to the time a bird is singing the gesture encoded by the neurons. Since there should be a substantial conduction delay between neuronal activity and singing, this motivates the hypothesis that the activity in HVC is a prediction of motor output. This suggests a new model of organization of vocal motor control: the output of the population of HVC projection neurons represents a predication (forward model) of the dynamics of gestures, used to process feedback, and the conversion to pre-motor activity occurs in primary motor cortex RA. Two specific aims will test these hypotheses. The first aim will use recordings of single identified HVC neurons in sleeping birds and in singing birds to test predictions of the gesture hypothesis. Feedback will be perturbed to assess if it affects non-linear temporal summation hypothesized to be the mechanistic basis for the prediction. The data will be assessed in terms of a statistical model sensitive to parameters of gesture trajectories. The second aim will test the hypothesis that the change of information from HVC to RA involves transformations from representations of gesture extrema to continuous paths through song. The topographic organization of RA will be assessed by systematic recordings in RA. In a related aim, the model will be improved through introducing bilateral interactions and time-depending upper vocal tract filtering, and creating automated procedures to facilitate rapid analysis of song stimuli for the other aims. These studies can generate new insight into vocal motor coding, and inform studies of human speech production.