Divergent thinking modulates interactions between episodic memory and schema knowledge: Controlled and spontaneous episodic retrieval processes

Abstract

The ability to generate novel ideas, known as divergent thinking, depends on both semantic knowledge and episodic memory. Semantic knowledge and episodic memory are known to interact to support memory decisions, but how they may interact to support divergent thinking is unknown. Moreover, it is debated whether divergent thinking relies on spontaneous or controlled retrieval processes. We addressed these questions by examining whether divergent thinking ability relates to interactions between semantic knowledge and different episodic memory processes. Participants completed the alternate uses task of divergent thinking, and completed a memory task in which they searched for target objects in schema-congruent or schema-incongruent locations within scenes. In a subsequent test, participants indicated where in each scene the target object had been located previously (i.e., spatial accuracy test), and provided confidence-based recognition memory judgments that indexed distinct episodic memory processes (i.e., recollection, familiarity, and unconscious memory) for the scenes. We found that higher divergent thinking ability—specifically in terms of the number of ideas generated—was related to (1) more of a benefit from recollection (a controlled process) and unconscious memory (a spontaneous process) on spatial accuracy and (2) beneficial differences in how semantic knowledge was combined with recollection and unconscious memory to influence spatial accuracy. In contrast, there were no effects with respect to familiarity (a spontaneous process). These findings indicate that divergent thinking is related to both controlled and spontaneous memory processes, and suggest that divergent thinking is related to the ability to flexibly combine semantic knowledge with episodic memory.

Appendix

Additional information about stimuli

The scene categories and targets consisted of kitchens (target: frying pan), dining rooms (target: wine glass), bedrooms (target: alarm clock), living rooms (target: coffee mug), and bathrooms (target: toothbrush cup). Eight different object exemplars were used per category, such that the visual features of the target object varied across different scenes within a category. In each scene, only one exemplar of the target object was present, and this was kept consistent across presentations. For example, in each living room scene, there was only one coffee mug present.

The congruent location for a target object was semantically consistent across all scenes in a category, such that targets were placed relative to larger objects with which the target objects co-occur with high probability in daily life (Boettcher et al., 2018; for review of scene grammar see Võ et al., 2019). Specifically, in bathroom scenes, the toothbrush cups were located next to sinks; in dining room scenes, the wine glasses were located on tables (within arm’s reach of a chair); in kitchen scenes, the pans were on stove burners; in bedroom scenes, the alarm clocks were on nightstands; and in living room scenes, the coffee mugs were on coffee tables. The spatial locations of the targets varied across scenes, as illustrated in Fig. 4.

Fig. 4

Distributions of target location midpoints within scenes. Note. A) Distribution of the target locations in congruent scenes. B) Distribution of target locations in incongruent scenes

Search time

The target was found on 98.9% of study phase trials. On trials in which the target was found, the average search time was (1) 2045 ms in congruent scenes and 2476 ms in incongruent trials (p = .0001), and (2) 2,482 ms on first presentation and 2,040 ms on second presentation (p < .0001). Thus, both semantic knowledge and episodic memory contributed to search speed in a similar fashion as to spatial accuracy.

Model equations

The equations for the models used for the primary (i.e., non-replication) analyses are specified below (Eqs 1–4). When these equations are discussed with respect to examining fluency and originality separately (in the Sensitivity Analyses section), the “AUT score” variable below was replaced with “fluency” or “originality,” depending on the analysis in question.

Recollection effects

For the difference between recollected and strength-matched familiar scenes, the analysis included old scenes that were given a response of 6 or 5, and the model was specified as:

$$\mathrm{Target\; distance }\sim \mathrm{ response }*\mathrm{ AUT\; score }+\mathrm{ image\; intercept }+\mathrm{ subject\; intercept}$$

(1)

For the congruency effects, the analysis included recollected scenes (old scenes that were given a response of 6) and was specified as:

$$\mathrm{Target\; distance }\sim \mathrm{ congruency }*\mathrm{ AUT\; score }+\mathrm{ image\; intercept }+\mathrm{ subject\; intercept}$$

(2)

Familiarity effects

For familiarity effects, the analyses included scenes across all levels of familiarity strength (old scenes that were given a response of 1-5). For the analysis that examined familiarity irrespective of congruency, the congruency parameter was removed:

$$\mathrm{Target\; distance }\sim \mathrm{ congruency }*\mathrm{ AUT\; score }*\mathrm{ response }+\mathrm{ image\;intercept }+\mathrm{ subject\; intercept}$$

(3)

Unconscious effects

For unconscious effects, analyses were conducted in old scenes given a response of “sure new,” and new scenes. For the analysis that examined unconscious memory irrespective of congruency, the congruency parameter was removed:

$$\mathrm{Target\; distance }\sim \mathrm{ congruency }*\mathrm{ AUT\; score }*\mathrm{ old\; vs\; new }+\mathrm{ image\; intercept }+\mathrm{ subject\; intercept}$$

(4)