Spatial Distortions in Visual Short-Term Memory

Abstract

Two-dimensional topographic maps represent a major coding principle for spatial information in the primate brain. In principle, neuronal maps can be subject to systematic distortion caused by interactions between neighboring cells as well as attentional influences. In this thesis, it is shown that when observers try to reproduce the exact position of a target dot on the monitor screen after a brief retention interval, their memory is systematically distorted by the presence of visual landmark stimuli serving as spatial reference. Landmarks provide regions of high positional certainty, markedly reducing response variance in their vicinity. At the same time, configurations of one, two, or three landmarks induce predictable distortional fields that are governed mainly by a stimulus-based (intrinsic) frame of reference but also interact with extrinsic reference systems, e.g., the allocentric vertical. Distortional fields can be invariant with image transformations, closely following changes in orientation, translation, or elongation of the landmark configuration, and are established as soon as 100 ms after the target stimulus has disappeared. Furthermore, it is shown that the distortional field of two landmarks can be predicted on the basis of knowledge of the single landmarks presented individually because the fields are locally invariant when a second landmark is presented at some distance, which leads to a partitioning of the visual field into regions of influence dominated by single landmarks. These results are inconsistent with previous theories of spatial memory distortions. Instead, an attentional model is advanced where spatial distortions arise from a preactivation of spatial reference systems in topographical cortical memory maps.