Prior Knowledge and Recognition Memory - a Computational Modeling Approach

For more than a century, an immense interest has been devoted to the study of recognition memory, where a multitude of memory phenomena has been explained. Recognition memory is usually described with parsimonious measurement and statistical models, stemming from dual process theory and signal detection theory. In the present thesis, the most often used models of recognition memory are reviewed and compared to a novel implementation of the variance theory, abbreviated the VT (Sikström, 2001) in the account of frequency and familiarity effects, and a new model of item variability (the multidimensional signal detection theory, abbreviated the MSDT). The focus of the thesis lies on the effects of prior knowledge on recognition memory, investigated with behavioral, electrophysiological and modeling approaches. In Study 1, a novel paradigm for measuring frequency and familiarity effects in recognition memory was introduced (the name paradigm), where recognition memory was tested on rare and common names that were famous and non-famous. The name paradigm was experimentally implemented in different conditions that provided a detailed description of fame and familiarity effects in recognition memory in four experiments. The study showed that pre-experimental knowledge both facilitates and impairs memory. Fame and frequency were selectively related to specific and non-specific semantic knowledge, where the former enabled retrieval of more and detailed information whereas frequency lacked such specificity at retrieval. The second study elaborated on prior knowledge on recognition memory with the name paradigm by recording Event-Related Potentials, a method with which electrophysiological signatures of cognitive processes can be linked to experimental manipulations. More specific, it was investigated whether old/new effects previously related to familiarity and recollection, the FN400 old/new effect (Mecklinger, 2006) and the late positive component (the LPC, see Rugg & Yonelinas, 2003), respectively, would be selectively induced by frequency and fame, thereby linking the experimental variables to the two memory processes. Further, in a second experiment, the proposed link between familiarity and conceptual priming (Paller, Voss & Boehm, 2007) was investigated. The behavioral findings replicated those in Study 1, and the ERP analysis revealed that low frequent names elicited the FN400 effect, whereas fame to a higher extent than frequency gave rise to the LPC. Experiment 2 demonstrated that familiarity (i.e., the FN400) is insensitive to conceptual priming. Study 3 provided a comprehensive account of fame and frequency effects by a novel implementation of the VT (Sikström, 2001). In two experiments the name paradigm was implemented in conditions where item, source and associative memory was assessed, which replicated the memory findings in Study 1 and 2. In the VT, fame was defined as a pre-experimental encoding of the stimulus. When a famous name was encoded the reinstatement of the item, based on previous experiences (prior to the experimental test) was associated with an increase in the specificity of the representation. This lead to an increase in net input to the underlying at retrieval, due to the high degree of similarity between the encoded and the retrieved item, and low degree of variability. Frequency, on the other hand affected the variability but not the magnitude of the net input, which resulted in lower memory performance. In Study 4, a new model of item variability was introduced, the MSDT. The MSDT describes memory with three parameters, similar to the account provided by signal detection theory (SDT), but introduces non-linearity’s to SDT, relies on binomial rather than normal latent distributions, and provides a multidimensional account of memory phenomena. The MSDT suggests novel predictions on changes in item variability as a function of attentional skill (i.e., ADHD versus healthy controls and varying degrees of attentional disabilities) as well as for the mediators in the differences in response variability in attentive and inattentive people. These predictions were tested on attentive and inattentive people, and provided augmented support for the model. The MSDT was conceptually and mathematically compared to the unequal-variance signal detection theory and the dual-process signal detection model (Yonelinas, 1994), and provided a more comprehensive account of the studied memory phenomena. Because attentive people yield a higher number of active nodes than attentive, and a lower variability in the activation threshold, the former group performs better and yields a higher ratio of new to old item variability than the latter. The MSDT also account for higher level of false alarms in inattentive than attentive, and suggests that the difference in new to old item variability is a result of increased new item variability relative that of old items. Further, the model provides a unified account of item- and response variability