Latent memory of unattended stimuli reactivated by practice: an FMRI study on the role of consciousness and attention in learning

Abstract

Although we can only report about what is in the focus of our attention, much more than that is actually processed. And even when attended, stimuli may not always be reportable, for instance when they are masked. A stimulus can thus be unreportable for different reasons: the absence of attention or the absence of a conscious percept. But to what extent does the brain learn from exposure to these unreportable stimuli? In this fMRI experiment subjects were exposed to textured figure-ground stimuli, of which reportability was manipulated either by masking (which only interferes with consciousness) or with an inattention paradigm (which only interferes with attention). One day later learning was assessed neurally and behaviorally. Positive neural learning effects were found for stimuli presented in the inattention paradigm; for attended yet masked stimuli negative adaptation effects were found. Interestingly, these inattentional learning effects only became apparent in a second session after a behavioral detection task had been administered during which performance feedback was provided. This suggests that the memory trace that is formed during inattention is latent until reactivated by behavioral practice. However, no behavioral learning effects were found, therefore we cannot conclude that perceptual learning has taken place for these unattended stimuli.

Figure 1. Schematic general procedure.

1A: The experiment consists of two separate sessions on consecutive days; a learning phase (on day 1) and a testing phase (on day 2). During the learning phase, both groups are exposed to a letter stream, against a background of textured figures. These figures are unreportable in both groups, but for different reasons. The Inattention group performs an irrelevant letter task and hence ignores the textured background, resulting in so-called ‘inattentional blindness’ for the figures (checked with a surprise 10AFC task). The Masked group is trying to detect figures in the background, yet these figures are masked (except for luminance defined ‘catch’ figures), rendering them invisible even with attention. One day later, during the testing phase, both groups perform the same tasks. First, learning is assessed neurally by an easy letter task, while in the background both trained figures and novel figures are presented (measuring BOLD activity). Thereafter, a staircased detection task with trained and novel figures is administered without and with feedback, to measure (the effect of performance feedback on) behavioral learning effects. After a 1-hour break, post-feedback BOLD activity is measured using the same set-up as the first (pre-feedback) BOLD measurement, to assess neural effects of performance feedback. 1B: Schematic illustration of the run, block and trial structure of each fMRI task. Trial details (exact timing and target stimulus specifications) can be found in Figure 2. Note that each block consists of a continuous letter stream, there are no breaks in between trials (only in between blocks and runs).

Figure 2. Schematic representation of the task designs.

On day 1 (learning phase) both groups perform a different task. The Inattention group performs a RSVP 2-back letter task, while a textured pattern was presented that either formed a figure (containing 6 squares, in layout A or B) or a no-figure (containing no squares) is presented in the background together with every target letter. The Masked group performs a figure detection task, where they have to detect whether or not a masked figure texture is present in the background while the target letter is presented. Easily detectable unmasked luminance defined ‘catch’ figure trials (both layout A and B) have been added to keep subjects motivated, and to check whether the task instructions were being followed (in 6% of the trials, matched by 6% ‘dummy trials’ in the Inattention group). For both groups, figure and no-figure trials were presented in a block design of 12 consecutive figure or no-figure trials (every ‘trial’ being a sequence of 6–8 letters, with 1 white target letter which was paired with a figure/no-figure). On day 2 (testing phase) both groups perform the same two tasks, to measure learning effects neurally (vowel task) and behaviorally (staircase task). In the vowel task, subjects have to indicate whether the target letter is a vowel or not, while both the trained and novel figure (layout A and B) are presented in the background (again in a block design, similar to day 1). In the staircase task, subjects have to detect whether or not the masked target is a figure. Both the trained and novel figure and no-figures are now presented randomly. According to their performance the SOA between target and mask varies, in a double staircase manner for both figures independently. Two versions of the staircase task are presented, one without and one with performance feedback. Percentages refer to the percentage of all trials that a particular stimulus/trial type was presented. This figure has been adapted from our previous study, Meuwese et al. .

Figure 3. Behavioral performance on day 1 (learning phase).

For the Masked group, percentage correct does not differ significantly from chance for detection of masked figure textures (49.9% correct, t _(1,11) = −.423, p = .68), whereas unmasked ‘catch’ figures are detected above chance level (83.2% correct, t _(1,11) = 8.901, p<.0001), which indicates that subjects were performing the task correctly but just were not able to detect the masked figure textures. The Inattention group performs above chance level at the 2-back letter task (89.2% correct, t _(1,12) = 23.65, p<.0001), indicating that top-down attention was directed at the letter task instead of the background figures.

Figure 4. Trained figure-evoked BOLD signal on day 1 (learning phase).

5A: In the ROI of the trained figure, in the Inattention group, a higher BOLD signal can be observed for the trained figure (minus no-figure) evoked signal than in the Masked group (F_(1,23) = 7.945, p = .0097). Within the Inattention group, the trained figure evoked signal differs from the no-figure signal at trend level (F_(1,12) = 3.361, p = .092). Thus, figure-ground segregation processes proceeded during inattention (see also Meuwese et al., 2013; Scholte et al., 2006), although marginally. For the Masked group, trained figure-related activity is blocked by the mask, such that the trained figure evoked even a lower BOLD signal than no-figure (F_(1,11) = 7.111, p = .02, note that this is an effect in the opposite direction; less figure activity than no-figure activity). The mask thus successfully blocked figure processing, which is also reflected in the performance on the detection task (which is at chance level, see Figure 3). 5B: In the ROI of the novel figure (which is not yet presented on Day 1) as expected, the trained figure evoked BOLD activity does not differ from that of the no-figure, neither within (Inattention group: F_(1,12) = 1.175, p = .30; Masked group: F_(1,11) = 3.736, p = .08 (although note that this is an effect at trend level in the opposite direction: less trained figure than no-figure evoked activity)) nor between groups (F_(1,23) = .100, p = .75).

Figure 5. Behavioral performance on the Staircase task on day 2 (testing phase).

No learning effects were found in the Staircase task, for either the version without or with performance feedback (Inattention group, no feedback: t _(1,12) = .462, p = .65, feedback: t _(1,12) = −.595, p = .56; Masked group, no feedback: t _(1,11) = 1.103, p = .29, feedback: t _(1,11) = .549, p = .59). There was no effect of feedback on the learning effect (trained minus novel figure)) (Inattention group: t _(1,12) = .709, p = .49, Masked group: t _(1,11) = .319, p = .76).

Figure 6. Trained figure and novel figure-evoked BOLD signal on day 2 (testing phase).

6A: We did not find a significant neural learning effect in the Inattention group for the pre-feedback BOLD measurement. For both groups, figure activity did not differ significantly from no-figure activity, for both the trained figure (Inattention group: F_(1,12) = 1.595, p = .23; Masked group: F_(1,11) = .154, p = .70) and the novel figure (Inattention group: F_(1,12) = .569, p = .47; Masked group: F_(1,11) = 1.670, p = .22). Nor did we find any learning effects (trained figure versus novel figure) in the BOLD signal for both groups. In fact, there was more novel figure than trained figure activity in both groups, even significantly so in the Inattention group (F_(1,12) = 5.902, p = .03; Masked group: F_(1,11) = 2.590, p = .14). Between-groups, we did not find any significant differences. 6B: The post-feedback BOLD signal for the trained figure was significantly higher than for the no-figure in the Inattention group (F_(1,12) = 8.512, p = .013), not in the Masked group (F_(1,11) = .188, p = .67) (between group difference: F_(1,23) = 4.929, p = .037). Trained figure activity was higher than novel figure activity in the Inattention group at trend level (F_(1,12) = 3.561, p = .08), in the Masked group the trained figure evoked a significantly lower signal than the novel figure (F_(1,11) = 8.136, p = .02) (between group difference: F_(1,23) = 8.110, p = .009). In both groups, novel figure activity does not deviate from no-figure activity. 6C: When post- and pre-feedback BOLD signal are compared, we found a large post-feedback increase of BOLD activity for the trained figure and for the learning effect (trained figure minus novel figure), only in the Inattention group (trained figure: F_(1,12) = 9.807, p = .009, learning effect: F_(1,12) = 13.099, p = .004) (Masked group, trained figure: F_(1,11) = .001, p = .98, learning effect: F_(1,11) = .048, p = .83). Between groups, the BOLD signal increase of the trained figure differs at trend level (F_(1,23) = 3.620, p = .07), the increase of the learning effect (trained versus novel figure) differs significantly (F_(1,23) = 6.023, p = .02). For the novel figure there were no significant BOLD changes in the post-feedback signal compared to the pre-feedback signal in both groups.