COMPUTATIONAL MODELING ANALYSIS OF MEMORY BINDING

This research is designed to isolate a particular operation that is of fundamental importance in human episodic memory--the operation underlying our ability to remember the parts of an event as having occurred together. is operationally defined by situations in which to-be-remembered items are tested by probes that consist either entirely of elements that co-occurred in the same original to-be- remembered event, or of elements, all of which are old, but which occurred in different events. If people call the former old' with greater frequency than the latter we may say that binding has occurred. Mechanisms of episodic memory binding are investigated using computational models of memory. Detailed modeling studies, yielding the predictions of different mechanisms of episodic memory binding will be conducted for CHARM, TODAM1, MINERVA, MINERVA* (without binding), competitive learning, and a back propagation model. The Configural Features hypothesis will also be investigated in each model. Eight experimental paradigms address different aspects of binding: (1) The Basic Paradigm--Composites and Conjunctions versus old items, (2) Binding in Implicit versus Explicit Memory, (3) Generalization versus Binding, (4) Binding in a Continuous Blends Paradigm, (5) Effects of Criterion on Conjunction and Composite Probes, (6) Binding in Between versus Within-Pair Mixtures, (7) Binding as a Function of Name Linking, and (8) Caricature. Parametrically varied modeling studies and experimental tests with humans will be conducted on these eight paradigms. This combined modeling and experimental research promises to deepen our understanding of human memory, and may also contribute to our ability to assist people with memory disorders.