Dynamics of symbol systems

To implement diverse and partial flow of information in cognitive processes, we need a design method without explicit stipulation of domain/task-dependent information flow, together with a control scheme for guiding information processing to concern only important information depending on contexts. A computational architecture is proposed which is based on a first-order logic program with a dynamics. The declarative semantics of the logic program is defined by measuring the degree of violation in terms ofpotential energy, and a control scheme for both analog and symbol computation is derived from the resultant dynamics, which replaces domain/task-dependent procedures, hence avoiding intractable complexity in the system design. This inherent integration of the control scheme with the declarative semantics guarantees that inferences are naturally centered around relevant information in a context sensitive manner. The essence of inference mechanisms proposed so far, such as weighted abduction and marker passing, are also subsumed by this dynamical control. All this justifies further exploration for improving this sort of formalism to deal with large, real-world problems.