We model the cortical dynamics underlying a free association between two memories. Computationally, this process may be realized as the spontaneous retrieval of a second memory after the recall of the first one by an external cue, what we call a latching transition. As a global cortical model, we study an associative memory Potts network with adaptive threshold, showing latching transitions. With many correlated stored patterns this unstable dynamics can proceed indefinitely, producing a sequence of spontaneously retrieved patterns. This paper describes the informational properties of latching sequences expressed by the Potts network, and compares them with those of the sentences comprising the corpus of a simple artificial language we are developing, BLISS. Potts network dynamics, unlike BLISS sentences, appear to have the memory properties of a second-order Markov chain.
Memory Attractor dynamics Information Artificial language
This is a preview of subscription content, log in to check access.