A continuous dual-process accumulation model of recognition judgments

Theories of recognition have shifted from a single process approach to a dual-process view, which distinguishes between knowing that one has experienced an object before (familiarity) and knowing what it was (recollection). The remember/know procedure, in which remember judgments are assumed to reflect recollection and know judgments are assumed to be based on familiarity, is widely used to investigate these two processes. While most recent dual process models can account for relationships among accuracy, remember/know judgments, and study factors that influence recognition (under a range of different assumptions), none of these models address the time course of the recognition process. As a results, paradoxical findings that familiarity is available faster than recollection but remember responses are on average faster than know responses, cannot be convincingly explained by any existing dual process model. In this dissertation, we resolve this paradox by proposing an elaborated dual process model of recognition called the Continuous Dual Process Accumulation (CDPA) model. The CDPA model uses the dual-system hypothesis of mammalian memory (Packard and McGaugh, 1996) as its neurological basis, describing the interplay between the hippocampus and the caudate in making recognition judgments, which allows it to make detailed predictions regarding the time course of recollection and familiarity, and explain how the information available through these two processes is applied to make the recognition decision . In the first half of the dissertation, a neuro-imaging study is presented, which tests a key assumption of the CDPA model that quick familiarity signals are based on perceptual judgments produced by the instrumental system (which includes the hippocampus), while the slower recollection signals require the habit system (which includes the caudate nucleus of the striatum) to generate the memory trace. The second half presents the CDPA model, which is implemented computationally as a collapsing bound diffusion model. A conventional recognition task for previously studied words is used to test the predictions of the model qualitatively and quantitatively. The model therefore extends signal detection theory, and allows, for the first time, predictions of hits and false alarms for remember and know judgments based on confidence, accuracy and RT.Ph.D.Includes bibliographical referencesby Neha Sinh