Boundary extension is a well established psychological phenomenon in which people consistently recall a scene with visual information beyond its boundaries (Intraub & Richardson, 1989). For example, when asked to draw the below image of a house from memory, participants drew empty space around the house, even though the house in the actual image is cut off by the image borders.
As such, boundary extension has been taken to mean that mental representations of scenes combine sensory information and schematic information that “extrapolates the views of a presented scene” in a process called scene construction (Bainbridge & Baker 2020a).
However, Dr. Bainbridge and her colleague Dr. Chris Baker noticed that in boundary extension studies, almost all of the stimuli that showed the phenomenon were images with one or a few central, close-up objects and a generic background. While it’s true that those types of scenes exist, the scenes we see in real life are typically made up of many objects, both near and far. Given this, to say that boundary extension is a universal phenomenon of scene memory might be too broad a generalization. To investigate, they conducted a study with a large and varied set of 1,000 images on 2,000 people (Bainbridge & Baker 2020a). Participants viewed an image, then after a brief delay where they saw a nonsense scrambled image, viewed the original image again. However, participants did not know that the two images were the same and answered whether the second image was closer or farther than the first.
Importantly, half of the 1,000 images were close-up images (object-oriented) like the ones in previous boundary extension experiments, while the other half were broader scenes (scene-oriented).
They found that while object-oriented images almost exclusively produced boundary extension (participants said the second image was closer), scene-oriented images were split evenly between displaying boundary extension and boundary contraction. Furthermore, the responses for the images were consistent across people, meaning that whether an image extends or contracts is a property of the image itself and not a result of random responding by participants. The researchers also looked at what these images contained and they found that images with “several, smaller, dispersed, and distant objects” tend to cause boundary contraction and images with “fewer, larger, more central, and close objects” tend to cause boundary extension.
These results show that for scene images that actually look like the scenes we see in real life, boundary contraction is equally as common as boundary extension. This goes against the idea that boundary extension only is a universal phenomenon of scene memory. They also found a strong relationship between how an image’s boundaries transform in memory and the visual make-up of the image, which implies that this effect is not driven largely by schematic knowledge of the scene as the original boundary extension work suggests, but by the images themselves.
In response to this finding, Dr. Helene Intraub, the original author of the foundational boundary extension work, argues that the experimental methods used in the study are not actually examining the scene construction mechanism that causes boundary extension but rather normalization, a different memory mechanism. Scene construction is a process that automatically constructs visual information from a scene beyond its boundaries, which is separate from normalization, the process in which we ‘normalize’ our scene memory to an existing average or schema that we already have. Dr. Intraub argues that because scene construction results in only boundary extension, the presence of both boundary extension and boundary contraction in the study means that the experiment design is activating additional cognitive mechanisms, including normalization (Intraub, 2020).
However, if boundary extension and boundary contraction are equally likely to occur in a large set of real-world representative scenes, is there truly a special link between boundary extension and scene construction to begin with? This is what Dr. Bainbridge and Dr. Baker argue in their reply to Dr. Intraub’s response (Bainbridge & Baker, 2020b). The reason that prior work only found boundary extension might be due to the fact that the scenes they used were limited to object-focused images and weren’t representative of the scenes we see naturally. In addition, previous studies show that this experimental set up not only elicits boundary extension but also eliminates normalization effects, refuting Dr. Intraub’s claim that the study’s method doesn’t actually test scene construction. In conclusion, they reiterate that their fundamental claim is not that boundary extension doesn’t exist, but that boundary contraction and extension are equally likely and are not necessarily separate mechanisms in the brain.
This discussion on boundary extension highlights the importance of open science and of data sharing in particular. Many of the ongoing questions could be tested and answered if the stimulus sets from prior boundary extension work were made available, Dr. Bainbridge and Dr. Baker point out. To learn more about this ongoing discussion, the articles mentioned in this post (also listed below) are a great place to start!
Articles:
- Intraub, H., and Richardson, M. (1989). Wide-angle memories of close-up scenes. J. Exp. Psychol. Learn. Mem. Cogn., 15, 179–187.
- Bainbridge, W. A., & Baker, C. I. (2020a). Boundaries Extend and Contract in Scene Memory Depending on Image Properties. Current Biology, 30(3), 537-543.e3. https://doi.org/10.1016/j.cub.2019.12.004
- Intraub, H. (2020). Searching for boundary extension. Current Biology, 30(24), R1463–R1464. https://doi.org/10.1016/j.cub.2020.10.031
-
Bainbridge, W. A., & Baker, C. I. (2020b). Reply to Intraub. Current Biology, 30(24), R1465–R1466. https://doi.org/10.1016/j.cub.2020.10.031