Computational Modeling of Updating Operations in the Performance of Basic Verbal Working-Memory Tasks.

Abstract

Verbal working memory is a mental system for the temporary maintenance and manipulation of information supporting ongoing performance. The ability to update knowledge representations in working memory is critical to this system's support for flexible and goal-driven behavior. While maintenance has been extensively studied, theorized and modeled, the understanding of working-memory updating has not progressed apace. At an abstract level, updating has been conceived as a control process handled by a central executive acting on information maintained in an articulatory loop (Baddeley, 1986). I revise and elaborate this theory by developing a set of explicit computational models that account for performance in four verbal working-memory updating tasks. Performing these tasks requires the operations of adding and deleting words from the beginning and end of serially ordered verbal sequences. Across four experiments, I compare updating operations and develop a series of computational models that account for important aspects of performance. This work reveals that updating is not implemented by unique processes of executive control, but rather by the same perceptual, motoric, and control processes that mediate maintenance and rehearsal. I also found that deletion of words is performed by passively omitting them from rehearsal, not by an active process of inhibition; and that when multiple updating operations are required they are performed each in turn through cycles of articulation. Through careful experimentation and detailed computational modeling, I have decomposed the abstract executive process of updating into an explicit theory of performance.