Episodic memory processes modulate how schema knowledge is used in spatial memory decisions

Abstract

Schema knowledge can dramatically affect how we encode and retrieve memories. Current models propose that schema information is combined with episodic memory at retrieval to influence memory decisions, but it is not known how the strength or type of episodic memory (i.e., unconscious memory versus familiarity versus recollection) influences the extent to which schema information is incorporated into memory decisions. To address this question, we had participants search for target objects in semantically expected (i.e., congruent) locations or in unusual (i.e., incongruent) locations within scenes. In a subsequent test, participants indicated where in each scene the target had been located previously, then provided confidence-based recognition memory judgments that indexed recollection, familiarity strength, and unconscious memory for the scenes. In both an initial online study (n = 133) and replication (n = 59), target location recall was more accurate for targets that had been located in schema-congruent rather than incongruent locations; importantly, this effect was strongest for new scenes, decreased with unconscious memory, decreased further with familiarity strength, and was eliminated entirely for recollected scenes. Moreover, when participants recollected an incongruent scene but did not correctly remember the target location, they were still biased away from congruent regions-suggesting that detrimental schema bias was suppressed in the presence of recollection even when precise target location information was not remembered. The results indicate that episodic memory modulates how schemas are used: Schema knowledge contributes to spatial memory judgments primarily when episodic memory fails to provide precise information, and recollection can override schema bias completely.

Keywords: Memory; Recognition; Recollection; Schemas; Semantic knowledge; Visual search.

Fig. 1.

Sample stimuli and procedure. A) The congruent version of a sample scene, with the target object (toothbrush cup) next to the sink. The green ring appeared around the target after participants clicked on the scene in the study phase. B) The incongruent version of the scene. C) Closeup of the target object in the congruent scene (for visualization only; this was not part of the experiment). D) Closeup of the target object in the incongruent scene. E) The trial sequence in the study phase, which consisted of 60 scenes presented two times each (120 trials). In each trial, a target probe appeared (e.g., “Find the toothbrush cup”), followed by the scene with target object. Participants were required to click on the target object within 10s. After clicking or after 10s, whichever occurred first, a green ring appeared around the target for 3s. F) The trial sequence in the test phase, which consisted of 80 scenes (80 trials). A target probe appeared, followed by the scene without the target object, and participants were given 10s to click on the scene location that they thought had contained the target when the scene was presented in the study phase. After 10s or clicking, whichever occurred first, participants gave a confidence-based recognition memory response for the scene.

Fig. 2.

Object location memory for schema-congruent and incongruent target objects in new scenes (A) and old scenes (B). Each heat map illustrates the distribution of recalled locations for the scenes, normalized such that the center of the heatmap represents the location of the target object. Thus, heatmaps tightly focused on the center-point—as in the recollected scenes—indicate high spatial accuracy, whereas more distributed heatmaps indicate poorer spatial accuracy. C and D) Spatial memory accuracy measured as the distance between the recalled location and the studied object location (i.e., target distance). Higher values indicate lower accuracy. To control for subject and image effects, the least-squares means derived from a linear mixed effects model with random effects of subject and image are plotted, and the error bars represent the standard error of these estimated means from the model.

Fig. 3.

The effect of memory on spatial recall errors for incongruent scenes. A) Example of the effect of schema congruency on spatial accuracy. The heatmaps are smoothed aggregate density maps of the click locations made on the congruent and incongruent version of the same scene. For example, the congruent heatmap includes the test phase click location from each participant who saw the congruent version of the scene. The incongruent heatmap suggests that many of the errors in these trials were due to participants erroneously choosing the congruent region. B) The congruent and incongruent versions of the scene. The target is circled in green for each case. C) Distance between the recalled object location and the schema-congruent location for objects studied in an incongruent location, plotted for each type of recognition response. Trials that were correctly recalled (i.e., <25 pixels from the studied location) have been excluded. The dashed line represents chance performance, which is the average distance between the recalled locations and a randomly selected target location. Values below the chance line indicate that participants’ selected locations were more likely than chance to be near the schema-congruent region, and indicate that errors on those incongruent scenes may be driven by schema bias. Values above the chance line, as in the recollect responses, indicate that the selected locations were less likely than chance to be near a schema congruent region; this suggests that errors in these trials were not driven by schema bias. To control for subject and image effects, the least-squares means derived from a linear mixed effects model with random effects of subject and image are plotted, and the error bars represent the standard error of these estimated means from the model.

Fig. 4.

Replication experiment data, showing object location memory for schema-congruent and incongruent target objects in new scenes (A) and old scenes (B). Each heat map illustrates the distribution of recalled locations for the scenes, normalized such that the center of the heatmap represents the location of the target object. Thus, heatmaps tightly focused on the center-point —as in the recollect responses—indicate high spatial accuracy, whereas more distributed heatmaps indicate poorer spatial accuracy. C and D) Spatial accuracy measured as the distance between the recalled location and the studied object location. To control for subject and image effects, the least-squares means derived from a linear mixed effects model with random effects of subject and image are plotted, and the error bars represent the standard error of these estimated means from the model.