Contextual modulation and dynamic grouping in perception

Abstract

First paragraph: Evidence for specialization of function within and between modules dominates research in cognition and neuroscience. Interactions that coordinate those specialized activities are also necessary, however, and, although often taken for granted, they are much less studied and much less well understood. Two major forms of coordination can be distinguished: dynamic grouping and contextual modulation1. I use the term ‘dynamic grouping’ to be more precise about issues that are commonly discussed under the heading of ‘binding’. Grouping can be divided into two fundamentally distinct classes: pre-specified grouping and dynamic grouping. Pre-specified grouping is due to the convergence of signals in a hierachy of feature detectors. It is ubiquitous in neural computation, and is the primary mechanism by which feature or object detectors compute whatever it is that they detect. It is pre-specified in the sense that although it adapts gradually to the statistical structure of input, it is specified as a possible grouping prior to the occurrence of the particular inputs processed at each moment. Dynamic grouping, by contrast, forms groupings that cannot be specified prior to the particular input being processed, and are computed by processes that configure themselves to that input. Dynamic grouping was emphasized by the Gestalt psychologists, and can occur pre-attentively. Processes that group features dynamically can therefore be clearly distinguished from the processes that compute the features that they group.