Two Signatures of Context Failure: A Computational Framework for Disambiguating WCST Perseveration

Abstract

Perseveration in patients with schizophrenia, defined as difficulty updating card-sorting behavior upon a rule change in the Wisconsin Card Sorting Task (WCST), has been classically ascribed to deficits in rule inference. However, patients might instead detect the rule change and fail to apply a new rule. Current measures of WCST performance cannot differentiate between these potential sources of perseveration. In this thesis, I present computational evidence showing that the failure to utilize a known rule is sufficient to cause perseveration in the task. I propose a neural network called the GEE which combines features of both Webb’s ESBN and Giallanza’s EGO architectures into a hybrid network. The GEE model employs a dual-pathway architecture whereby the task identity information is represented by a static pathway (h0), and the current rule is encoded in the dynamic pathway (RCM). Presentation of the rule via an oracle input removes the need for inference. Specific disruptions to either pathway result in distinct error patterns within the accuracy, consistency, and perseveration dimensions. Static disruption induces disorganized errors; dynamic disruption generates perseverative errors in the presence of the rule. This interaction pattern holds even under biologically motivated gain modulation. The resulting theoretical framework leads to a testable clinical hypothesis for separating perseverators by accuracy–consistency–perseveration criteria.