Fig. 1.

A predictive coding formulation of the Hermeneutic Circle: e.g., Speech Chain in the setting of language (Denes & Pinson, 1993). This schematic provides a simple example of neural hermeneutics in the form of a control system diagram. It is a simple example because the internal (generative) model predicting the behavioural consequences of action (of self and other) is the same. In other words, it is neither a model of my behaviour or your behaviour – but a model of our behaviour. When both agents adopt the same model, generalised synchronisation is guaranteed and prediction errors are minimised. The implicit architecture highlights the fact that the top-down predictions from a dynamical generative model (labelled Narrative) come in two flavours: exteroceptive predictions predicting the external consequences of action (c.f., corollary discharge) and proprioceptive predictions that predict the internal consequences of action (c.f., motor commands). These predictions are compared with sensory input to provide prediction errors. In control diagrams of this sort ⊗ denotes a comparator. Exteroceptive (e.g., auditory) prediction errors are used to update the generative model at various timescales to produce inference and learning. In contrast, the proprioceptive prediction errors drive classical reflexes to produce the predicted action. When the (dynamics of the) generative models in the two brains are identical, both exteroceptive and proprioceptive prediction errors are minimised and the dynamics will exhibit (generalised) synchrony. The red arrows denote learning or control by prediction errors that compare (descending) predictions with (ascending) sensations.