Visual Working Memory is more tolerant than visual long-term memory
As a result, the only difference in behavioral performance could be a result of whether an object representation was probed over the short-term (in VWM) or the long-term (in VLTM).
Figure 2. Results of Experiment 1a & 1b. The 0% noise conditions in the figure were obtained in Experiment 1a and all other noise condition results came from Experiment 1b (i.e. a different group of participants). Across both experiments, memory performance was unaffected by noise level at test in VWM, with comparable performance across 0-75% noise. In contrast, VLTM was greatly affected by noise at test, with a linear decrease in performance observed across 0-75% noise. * designates p < 0.01. Error bars represent SEM.
Figure 3. Results of Experiment 2. VWM performance was equivalent across recognizing a previously seen object at the same or a different orientation. In contrast, VLTM performance was significantly worse when recognizing a previously seen object at a different orientation than the one it originally appeared in. * designates p < 0.01. Error bars represent SEM.
We addressed this concern in Experiment 3, a one trial version of Experiment 1a. Participants did only one trial of the visual working memory task, in which they were shown two objects and then tested immediately for memory of one of them. The other item was saved while the participant completed an unrelated study (one of a number of the other experiments in the lab that did not include images of real-world objects). Finally, the participant was asked to identify the second previously shown object from a pair of images that included a foil, i.e. the participant was given a surprise VLTM test probing the saved image from the previous VWM trial. In this way, the VLTM test took place without the potential for interference among a large set of exposure images.
We found similar performance relative to Experiment 1a, suggesting average performance differences were not the result of interference between memoranda. This is the only single trial experiment on VLTM that we are aware of. It provides evidence that VLTM intolerance is not caused exclusively by interference, and also reinforces the motivation to distinguish between VLTM and VWM.
We conducted three additional experiments in order to address potential alternative accounts or confounding factors in the primary experiments.
Encoding Time (Experiment 5): It may be that the relative intolerance demonstrated by VLTM reflects inadequate encoding time. To investigate the possibility, we replicated Experiment 2 with increased encoding time from 0.5 to 1 seconds. Again, in VWM we did not observe significant difference for recognizing a previously seen object at the same or different orientation, but VLTM performance was significantly worse at recognizing a previously seen object at a different orientation.
The Role of Color (Experiment 6): We have emphasized the encoding of objects into working and long-term visual memory. It is possible though that discrimination responses in the tasks rely on lower-level display features. Color, in particular, could be available as a means to discriminate choices in the immediate VWM test, but not the VLTM test. To evaluate this possibility, we replicated Experiment 1b but with completely grayscale images. Despite all stimuli being presented without color information, performance was remarkably similar to Experiment 1b.
Test-induced Forgetting (Experiment 7). The general instructions for the experiments so far presented the VLTM test as a surprise, with the participants at first only told that they would be tested following each trial. Perhaps testing one item in a display immediately led to the strategic forgetting of the other. To evaluate this possibility, we modified the basic paradigm so observers were told they would be tested for long-term memory after the VWM portion of the experiment. Additionally, in half of trials (unpredictably sorted) there was no immediate VWM test after the encoding display. Thus items were available for VLTM testing that were not paired with an already-tested image. We did not find a significant difference in VLTM performance for items that appeared in immediately tested compared to untested displays