Modelling the psychological representation underlying perceptual and cognitive tasks
Talk presentation by William Ngiam at the Australasian Society for Philosophy and Psychology 2025 conference.
Abstract
Many cognitive models assume a set of processes operating upon a psychological representation. This representation is often derived using multidimensional scaling (MDS) of similarity judgments. While MDS-based models have been useful in exploring cognitive phenomena, the representation itself should not be left unexplained. Further, some take issue with a similarity-based projection as the basis for cognition. This issue has recently emerged in the field of visual working memory (VWM). Most VWM models are constructed on the physical stimulus space (e.g. Neural Resource model; Bays et al., 2014; Interference Model, Oberauer and Lin, 2017), but Schurgin et al. (2020) recently argued for a signal-detection model in terms of psychological similarity. It seems likely that there are both perceptual and psychological contributions to VWM. Thus, there remains a need to be able to model the underlying VWM representation. We took a Bayesian generative modelling approach and modelled open data of three tasks: a perceptual reproduction task, a quad psychophysical scaling task, and a memory reproduction task (Tomic and Bays, 2024). We show that the underlying representation can be sensibly recovered with this approach, finding non-uniformity in the representation across all tasks that is theoretically tractable. We find the recovered representation differs slightly across all tasks, likely due to the task context, which raises some questions for measurement models and theories of VWM.