Task Termination Triggers Spontaneous Removal of Information From Visual Working Memory

Abstract

Working memory (WM) is a goal-directed memory system that actively maintains a limited amount of task-relevant information to serve the current goal. By this definition, WM maintenance should be terminated after the goal is accomplished, spontaneously removing no-longer-relevant information from WM. Past studies have failed to provide direct evidence of spontaneous removal of WM content by allowing participants to engage in a strategic reallocation of WM resources to competing information within WM. By contrast, we provide direct neural and behavioral evidence that visual WM content can be largely removed less than 1 s after it becomes obsolete, in the absence of a strategic allocation of resources (total N = 442 adults). These results demonstrate that visual WM is intrinsically a goal-directed system, and spontaneous removal provides a means for capacity-limited WM to keep up with ever-changing demands in a dynamic environment.

Keywords

working memory delay activity capacity limit forgetting open data open materials

Our environment is highly dynamic and overwhelms us with sensory information. From this information overload, we must selectively keep task-relevant information in mind so that we can complete the task at hand. To do so, our mind is equipped with an on-line memory system known as working memory (WM), which allows us to maintain a set of task-relevant information for a short period by protecting it against interference from task-irrelevant information (Cowan, 1999; Vogel et al., 2006). However, WM is severely limited in capacity (Cowan, 1999; Luck & Vogel, 1997). To be efficient in a dynamic environment, it is imperative that we remove no-longer-relevant information from WM as soon as it become unnecessary for the task at hand. If no-longer-relevant information lingers in WM, it causes interference detrimental to achieving the goal at hand (Shipstead & Engle, 2013; Souza & Oberauer, 2015; Underwood, 1957).

Previous studies have shown that no-longer-relevant information can be intentionally removed from WM in the visual, auditory, and tactile domains (Astle et al., 2012; Backer & Alain, 2012; Dames & Oberauer, 2022; Katus et al., 2012; Oberauer, 2001; Souza et al., 2014; Williams et al., 2013). However, this WM removal has only been demonstrated by requiring individuals to strategically reallocate WM resources to other information that competes for limited WM resources. Thus, the removal of information from WM may occur only indirectly through biased competition for WM resources.

This indirect-removal hypothesis is well supported in the long-term-memory literature (e.g., Bjork, 1989; Kozak et al., 2021; Tozios & Fukuda, 2020). For example, in the intentional-forgetting paradigm, reliable forgetting was observed only when cue-to-forget information was presented together with cue-to-remember information, and no reliable forgetting occurred when cue-to-forget information was presented in isolation (Bjork, 1989; Tozios & Fukuda, 2020). Thus, the directed-forgetting effect is likely caused by interference from remembering other information rather than the direct forgetting of unwanted information (Bancroft et al., 2013; Conway et al., 2000; Lee et al., 2007). Along these lines, directly suppressing an unwanted memory ironically increases the accessibility of that memory, which is known as the white-bear effect (Wegner et al., 1987). Interestingly, the white-bear effect can be avoided if one focuses on another distracting item whenever the to-be-suppressed memory comes to mind (Muris et al., 1993; Wegner et al., 1987). This distractor item interferes with the to-be-suppressed memory, preventing the unwanted memory from occupying one’s mind. Therefore, it is plausible that the removal of unwanted memories requires competition with other information.

Similarly, retro-cue studies demonstrated that no-longer-relevant WM items can be removed in the presence of other task-relevant items (Astle et al., 2012; Souza et al., 2014; Williams et al., 2013). This paradigm first requires participants to remember several visual objects; then a retro-cue indicates a to-be-tested item during a retention period, allowing participants to remove the uncued items. If an uncued item is occasionally tested, memory performance for that item is typically lower than that for the cued item, suggesting that the uncued items were removed from visual WM. This finding has been explained as the strategic biasing of memory competition. Visual WM items compete for the allocation of limited memory capacity so that directing attention toward the cued item counteracts the interference from the uncued items, resulting in the loss of the uncued items in the competition and subsequent exclusion from visual WM (Pertzov et al., 2013; Vogel et al., 2005). However, it remains unclear whether no-longer-relevant items can be removed from WM without the presence of competing task-relevant items.

In the current study, we examined whether no-longer-relevant information can be removed from WM in the absence of other task-relevant information. Here, we provide both behavioral and electrophysiological evidence that information is indeed removed from visual WM in less than 1 s from the moment that it becomes no longer relevant to the task at hand. Our finding elucidates a fundamental function of WM to allow rapid updating of task goals despite severe capacity limitations.

Statement of Relevance

At any given moment, working memory allows us to hold the limited amount of task-relevant information necessary to achieve a current goal. Considering that our goal is frequently changing in real life, we need to remove no-longer-relevant information from working memory as soon as the current goal is accomplished so that new information necessary for a new goal can be represented. By measuring participants’ electroencephalogram (EEG) and behavioral performance, we found that no-longer-relevant information is spontaneously removed from working memory less than 1 s after its use. Critically, this phenomenon was observed only when the information was deemed no longer relevant, and not when participants expected the information to be reused. These findings elucidate a critical mechanism through which we can efficiently utilize limited working memory capacity and flexibly support our goal-directed behavior with consistently changing goals.

Open Practices Statement

The data, code, and materials are publicly accessible via the Open Science Framework and can be accessed at https://osf.io/6589w. Neither of the studies reported in this article was preregistered.

Experiment 1

We examined whether no-longer-relevant items are spontaneously removed from visual WM upon completion of a task goal. Unlike in the directed-forgetting paradigm, we gave no instructions to forget no-longer-relevant items, and we simply tracked the contralateral delay activity (CDA), an electrophysiological index of the number of items currently held in visual WM (Vogel & Machizawa, 2004), after participants completed the visual WM task.

Method

Participants

Twenty-seven university students (16 females, ages 18–22 years) participated. Two participants’ data (7.4% of the sample) were excluded from analysis because of an excessive number of trials contaminated by ocular artifacts (> 25% of trials). The sample size (N = 25) was determined on the basis of similar visual WM studies measuring CDA (Drew et al., 2012; Hakim et al., 2019). Note that we replicated the basic findings in six experiments. In addition, our sample size was sufficient to detect the main effect of interest (i.e., the set-size effect in CDA amplitude) with 0.96 statistical power (Ngiam et al., 2021) when combining main event-related potential (ERP) experiments (Experiments 1 and 4; see also Experiment S1 in the Supplemental Material available online).

In all experiments of the present study, participants provided written informed consent to the protocols approved by the Research Ethics Board of the University of Toyama (and the University of Toronto in Experiment 3). They had normal or corrected-to-normal visual acuity, as well as normal color vision, and received JP¥1,500 (US$14) per hour for participation in Experiments 1, 2, and 4, and partial course credit for participation in Experiments 3 and 5.

Stimuli and behavioral procedure

We used a bilateral color-change-detection task developed by Vogel and Machizawa (2004; Fig. 1a). A bilateral array of colored squares—set-size 1 (SS1) or set-size 4 (SS4)—was presented for 100 ms. The participants remembered the items only in the left or right hemifield, which was indicated by an arrow. After a delay period (a 400-ms short delay or a 1,900-ms long delay), a single test square appeared for 100 ms, followed by a blank screen for 1,400 ms. Participants then reported whether the test color was identical to the corresponding sample square (50% of trials) as quickly as possible. Half of the participants pressed a button to report “same” and pressed no button to report “different”; the other half of the participants had inverted the response mapping so that they pressed a button to report “different” and pressed no button to report “same.” Participants were required to fixate on a central cross (0.8° × 0.8°) without blinking until it disappeared at the end of the trial. Crucially, we gave no instructions about forgetting information from visual WM. Participants completed the three conditions (long-delay SS1, long-delay SS4, and short-delay SS4) in three separate 200 trial blocks. The block order was counterbalanced across participants (see the Supplemental Material for detailed procedure and stimulus information).

Fig. 1.

Stimuli and event-related potential (ERP) results from Experiment 1. In (a) is shown an example of a trial in the long-delay and short-delay SS4 conditions. Grand average ERP difference waves (contralateral delay activity, or CDA), time-locked to the sample array presentation, are illustrated in (b). Error bands show the standard errors of the mean. The gray rectangle indicates the time period for the sample array. The blue, pink, and green rectangles indicate the time period for the test display of the corresponding condition. The mean reaction time (mean RT) to the test in the short-delay SS4 condition is also shown as the blue dotted line with the standard error of the mean. To track the difference in the time course of the CDA across conditions, CDA amplitudes were binned into 20-ms time windows and then subjected to planned t tests between each pair of conditions (brown bars; see the Method section in the main text). SS1 = set-size 1; SS4 = set-size 4.

Electrophysiological recordings and analysis

We recorded the continuous electroencephalogram (EEG) using a BrainCap recording system (Brain Products GmbH, Gilching, Germany). Recording and analysis procedures, including thresholds for rejecting trials contaminated by blinks or eye movements, were derived from our previous studies (Fukuda et al., 2016; Tsubomi et al., 2013; Vogel & Machizawa, 2004).

We recorded from 22 standard electrode sites spanning the scalp, including international 10–20 sites F3, F4, C3, C4, P3, P4, O1, O2, PO3, PO4, T5, and T6 as well as the nonstandard sites occipital left (OL) and occipital right (OR; midway between O1/2 and T5/6). The horizontal electrooculogram (EOG) was recorded from electrodes placed 1 cm to the left and right of the external canthi to measure horizontal eye movement, and the vertical EOG was recorded from an electrode beneath the right eye to detect blinks and vertical eye movements. Electrodes for use as reference sites were placed on the left and right mastoids. Electrode impedances were reduced to 5 kΩ. All signals were digitized at 500 Hz with no filtering.¹

Off-line signal processing was performed using the EEGLAB Toolbox (Delorme & Makeig, 2004) and the ERPLAB Toolbox (Lopez-Calderon & Luck, 2014). All electrodes were referenced to the average of the left and right mastoids. The signals were band-pass filtered using a noncausal Butterworth filter, half-amplitude cutoffs at 0.05 to 20 Hz, and a roll-off of 12 dB/oct. The continuous EEG was segmented into trial epochs defined as −200 to 2,000 ms after the sample array onset in Experiments 1 and 2 and −200 to 1,500 ms after the sample array onset in Experiment 4. Trials accompanied by horizontal eye movements (> 32 μV for EOG electrodes), blinks (> 70 μV for EOG electrodes), or excessive voltage for EEG electrodes (> 100 μV) were rejected before further analysis. Participants with more than 25% rejected trials were excluded from the analysis. The average proportion of rejected trials across the remaining participants was 6.7% (SD = 5.5), 5.6% (SD = 5.1), and 2.7% (SD = 3.4) for Experiments 1, 2, and 4, respectively. Contralateral waveforms were computed by averaging the activity recorded at right-hemisphere electrode sites when participants were cued to the left side of the sample array and vice versa when participants were cued to the right side of the sample array. The CDA was measured at posterior parietal, lateral occipital, and posterior temporal electrode sites (P3/4, Po3/4, T5/6, and OL/R) as the difference in mean amplitude between the ipsilateral and contralateral waveforms.

Time-course analysis of the CDA

To track the difference in the time course of CDA across conditions, we first computed the average CDA amplitude for every 20-ms time window and then submitted each of them to planned t tests between each pair of conditions. To avoid a Type I error arising from using a large number of t tests (Luck, 2014), we considered only the significant differences that continued longer than 100 ms. Similarly, we considered that continued significant differences diminished only when nonsignificant differences lasted longer than 100 ms. In this analysis, we show averaged statistical values—M, t, p, Cohen’s d, and 95% confidence intervals (CIs)—within a time window of interest. When a delay period is referenced, we show the averaged statistical values from 300 ms to 2,000 ms in Experiments 1 and 2 (and in Experiment S1 in the Supplemental Material) and from 300 ms to 1,500 ms in Experiment 4. We also ruled out the possibility that the CDA results reflected contamination by eye-movement-related artifacts by examining the horizontal electrooculogram (see the Supplemental Material for detailed analysis and results).²

Results

We tracked the time course of the CDA with 20-ms time windows (see the Method section). As shown in Figure 1b, a typical CDA emerged 300 ms after the onset of the memory array, which was more negative over the delay period in the long-delay SS4 condition than in the long-delay SS1 condition, M_{long-delay SS4 ₋ long-delay SS1} = −1.1 μV, t(24) = 4.14, p < .001, d = 0.83, 95% CI = [−1.6, −0.5] (see the Method section for additional statistical details). In the critical condition (short-delay SS4), the participants remembered four colored squares, but they were tested after a 400-ms delay period. The mean reaction time to the test was 591 ms, 95% CI = [559, 623]. Interestingly, the CDA reduced its amplitude following the response. The difference in the CDA amplitudes between the short-delay and long-delay SS4 conditions became statistically significant 669 ms after the response (1,760 ms in Fig. 1b), M_{short-delay SS4 ₋ long-delay SS4} = 0.8 μV, t(24) = 2.70, p = .01, d = 0.54, 95% CI = [0.2, 1.5]. In addition, the CDA amplitude of the short-delay SS4 condition became statistically indistinguishable with that of the long-delay SS1 condition 409 ms after the response, M_{short-delay SS4 ₋ long-delay SS1} = −0.3 μV, t(24) = 1.08, p = .29, d = 0.22, 95% CI = [−0.8, 0.2]. These results suggest that participants removed no-longer-relevant items from visual WM as quickly as ~700 ms after task completion without reallocating WM resources to other information. By conducting a supplemental experiment, we also ruled out the possibility that the reduction in CDA amplitude in the short-delay SS4 condition was due to test-stimulus presentation and the response execution to the test (Experiment S1; see the Supplemental Material).

Experiment 2

Experiment 1 demonstrated that WM content is removed after it fulfills its purpose in a change-detection judgment. To further illustrate the goal-dependent nature of WM item removal, we devised a task in which WM content was used in two consecutive change-detection judgments. If WM information is removed only after the completion of a task, then it should be retained in WM after its use in the first change-detection task until the second change-detection task has been completed.

Method

Participants

Twenty-six university students (14 females, ages 19–23 years) participated. The sample size was determined in the same manner as in Experiment 1. One participant’s data (3.8% of the sample) was excluded from analysis because of an excessive number of trials contaminated by ocular artifacts (> 25% of trials).

Stimuli and experimental procedure

The experiment was identical to Experiment 1 with the following exceptions. As in Figure 2a, the participants remembered one or four colored squares for 1,900 ms, and their memory was tested only once in the baseline conditions (one-test SS1 and SS4). In the critical condition (two-test SS4), the participants remembered four colored squares, and their memory was tested twice. The first test was presented 400 ms after the offset of the sample display, which was followed by a 1,400-ms blank period, and then the second test was presented for 100 ms. The second test square appeared either at the same or a different location from the first test (50% of trials for each). The colors of the two test squares could be the same or different (50% of trials for each). Participants performed each condition in separate blocks of 200 trials. The order of the blocks was counterbalanced across participants.

Fig. 2.

Stimuli and event-related potential (ERP) results from Experiment 2. In (a) we show an example of a trial in the one-test SS4 and two-test SS4 conditions. Grand average ERP difference waves (contralateral delay activity, or CDA) time-locked to the sample array presentation are shown in (b). Error bands show the standard errors of the mean. The gray rectangle indicates the time period for the sample array. The blue, pink, and green rectangles indicate the time period for the test display of the corresponding condition. The mean reaction time (mean RT) to the first test in the two-test SS4 condition is also shown as the blue dotted line with the standard error of the mean. The brown bars show statistical comparisons of the time course of CDA amplitudes among each pair of conditions. SS1 = set-size 1; SS4 = set-size 4.

Results

As shown in Figure 2b, the CDA amplitude in the one-test SS4 condition was more negative than in the one-test SS1 condition over the delay period, M_{one-test SS4 – one-test SS1} = −0.5 μV, t(24) = 2.93, p < .01, d = 0.59, 95% CI = [−0.9, −0.1]. Importantly, the CDA amplitude in the two-test SS4 condition was also more negative than in the one-test SS1 condition over the delay period, M_{two-test SS4 – one-test SS1} = −0.8 μV, t(24) = 4.42, p < .001, d = 0.88, 95% CI = [−1.2, −0.5], and it was as large as in the one-test SS4 condition, M_{two-test SS4 – one-test SS4} = −0.1 μV, t(24) = 1.25, p = .22, d = 0.26, 95% CI = [−0.5, 0.3]. These results confirm that visual WM content is removed only after it fulfills its purpose toward a task goal.

Experiment 3

The results from Experiments 1 and 2 provided electrophysiological evidence that no-longer-relevant content is spontaneously removed from visual WM after task completion. Nevertheless, it is also possible that the memory representations cease to be lateralized yet are still maintained in WM. Alternatively, our results could also be explained by assuming that WM representations are turned into a neurally inactive state (Awh & Vogel, 2020; Lewis-Peacock et al., 2018). Thus, we conducted Experiment 3 to seek behavioral evidence of WM content removal. If no-longer-relevant items are indeed removed from WM after the perceived completion of a task, participants should fail to correctly report the items 1 s after they believe the task is complete.

Method

Participants

A set of 216 university students participated (115 females, ages 18–33 years; 155 students at University of Toyama and 61 students at University of Toronto). We used a large sample size for the following reason. We evaluated the number of participants who were able to correctly respond to the unexpected test and those who were unable to. We then examined whether they were significantly different from the expected ratio based on their mean accuracy of the memory test in the baseline block. Visual WM performance has been reliably estimated with 40 to 60 trials for a given condition in typical visual WM studies (Fukuda & Vogel, 2011; Tsubomi et al., 2013; Xu et al., 2018). Thus, we first set our target sample size to 50 participants per condition, bringing us to 100 in total. After considering that we used a new method, we doubled our target sample size (100 participants per condition). We advertised for participant recruitment on a semester basis and stopped once we reached the target.

Nineteen participants (9.6%) were excluded from the analysis for the following reasons: 8 participants anticipated that there would be an unexpected test despite the instructions; 3 participants responded to the unexpected test extremely slowly (> 3 SD from the average); 2 participants showed chance-level performance; 2 participants failed to answer the postexperiment questionnaire; 2 participants signed up and accidentally participated in the experiment twice; 1 participant left the study early; and 1 participant did not follow instructions.

Stimuli and experimental procedure

Participants performed a simple color-change-detection task (Luck & Vogel, 1997) for three blocks in a fixed order (Fig. 3a; see the Supplemental Material for detailed stimulus information). The first and second blocks served to estimate each participant’s baseline visual WM performance. In the first block (24 trials), participants remembered four sample colored squares presented for 100 ms, which was followed by a 900-ms blank delay period and the presentation of a test square. Participants reported whether the test color was identical to the corresponding sample square (50% of trials) by pressing the “F” or “J” key on a keyboard as quickly as possible. In the second block (96 trials), participants were tested twice. The second test square was presented 900 ms after the participants responded to the first test, either at the same location or at a different location from the first test (50% of trials for each). The color of the first and second test squares could be the same or different (50% of trials for each). In the third block (containing six trials), participants were instructed that their memory would be tested only once, which was true for the first five trials. On the sixth trial, however, an unexpected second test square was presented 900 ms after the participants responded to the first test, either at the same location or at a different location from the first test (50% of trials for each). The colors of the first and unexpected second test square could be the same or different (50% of trials for each). To prompt participants to respond to the unexpected test, we instructed them at the beginning of the experiment to continue the task until they see “You completed the block” on the computer screen. After the unexpected test participants reported how surprised they were by the unexpected test, using a 6-point Likert scale (1 = not at all, 2 = barely, 3 = a little, 4 = moderately, 5 = fairly, 6 = very much). The participants also indicated whether they had nonetheless anticipated the second test in the third block by checking either “I followed the instructions and expected one test” or “Despite the instructions, I prepared for two tests.”

Fig. 3.

Stimuli and behavioral results from Experiment 3. Participants performed a simple color-change-detection task for three blocks (a). Their memory was tested once in the first block and twice in the second block. In the third block, participants were instructed that their memory would be tested only once. On the sixth trial, however, an unexpected second test was presented 900 ms after the participants responded to the first test. Blue bars (b) show the accuracy in the second block. Red bars show the percentage of correct participants for the unexpected test trial in the third block (sixth trial). The error bars represent 95% confidence intervals.

Results

In the first and second blocks, we measured baseline performance. In the first block, participants were tested once. The mean accuracy was 85%, 95% CI = [84%, 86%]. In the second block, participants were tested twice. The mean accuracy was 87% for the first test, 95% CI = [86%, 88%], and 81% for the second test, 95% CI = [80%, 81%], as shown in Figure 3b.

The third block served as the critical block. Participants were instructed that they would be tested only once. However, on the sixth trial, an unexpected second test square was presented 900 ms after the participants responded to the first test (Fig. 3a). If none of the participants removed no-longer-relevant items from WM after the first test, the percentage of the participants who were able to correctly respond to the unexpected second test should be equal to the mean accuracy of the second test in the second block (81%). However, the results in Figure 3b showed that only 66% of the participants, 95% CI = [59%, 73%], were able to correctly respond to the unexpected second test (exact binomial test, p < .001).³ Computing visual WM capacity (K) by the percentage of correct participants, we found that 1.3 items were held in visual WM, 95% CI = [0.8, 1.8].⁴ In addition, we observed that the percentage of participants who correctly responded to the unexpected test was statistically indistinguishable when the second test square appeared at the same or a different location compared to the first test (see the Supplemental Material). Thus, the tested and untested items were equally removed from WM after the first test. As for the first test, 83% of the participants, 95% CI = [77%, 88%], responded correctly in the surprise trial, which was not significantly different from the expected percentage based on the mean accuracy (87%) of the first test (exact binomial test, p = .13).

We ruled out two alternative explanations that could account for the observed WM content removal in the unexpected test. First, we considered the possibility that surprise due to the unexpected test, rather than the perceived completion of the task, might have caused WM content removal. If so, those who were incorrect in the unexpected change-detection task should exhibit a higher surprise score than those who were correct. Contrary to this prediction, the mean surprise score of the correct participants was not significantly different from that of the incorrect participants with a very small effect size, M_{correct participants} = 4.3, M_{incorrect participants} = 4.2, t(195) = 0.69, p = .49, d = 0.10, 95% CI_{correct participants} = [4.1, 4.5], 95% CI_{incorrect participants} = [3.8, 4.5]. Second, those who took longer to make the first change-detection judgment might have lost the WM contents before the unexpected test because of a time-based decay. If so, those who were incorrect in the unexpected test should show a longer response time (RT) for the first change-detection judgment than those who were correct. However, the reaction time for the first change-detection judgment was statistically indistinguishable between correct and incorrect participants, M_{correct participants} = 706 ms, M_{incorrect participants} = 753 ms, t(98) = 0.91, p = .36, d = 0.13, 95% CI_{correct participants} = [650, 765], 95% CI_{incorrect participants} = [670, 840].

Experiment 4

The experiments so far demonstrated that memory content was spontaneously removed from WM when it became no longer relevant because of the perceived completion of the task. However, if WM is truly goal oriented, task completion should not be required to trigger this removal, but instead, removal should be triggered as soon as WM content is deemed unnecessary (because of task termination) even before the completion of the task. We tested this hypothesis by signaling that WM content is no longer necessary in the middle of the retention interval with an auditory cue indicating task termination.

Method

Participants

Twenty-nine university students (16 females, ages 19–24 years) participated. The sample size was determined in the same manner as in Experiment 1. One participant’s data (3.4%) was excluded from analysis because of an excessive number of trials contaminated by ocular artifacts (> 25% of trials).

Stimuli and experimental procedure

The experiment was identical to Experiment 1 with the following exceptions (Fig. 4a). The experiment consisted of three conditions (test SS4, no-test SS4, and test SS1). In the test SS4 and no-test SS4 conditions, a 50-ms auditory cue was presented 400 ms after the offset of the sample colored squares. For half of the participants, a low-tone cue (650 Hz) indicated that the test would not be presented, whereas a high-tone cue (2500 Hz) indicated that the test would be presented. For the other half of the participants, the tone-test relationship was inverted. Importantly, we only informed participants that the auditory cue would indicate whether the test would be presented or not; we gave no instructions about forgetting. In the test SS1 condition the auditory cue was not presented, and the test was presented in all trials. The delay period was fixed at 1,400 ms for all three conditions. Participants were required to fixate on a central cross without blinking until they responded to the test in the test SS1 and test SS4 conditions. In the no-test SS4 condition, participants were required to do the same until the central cross shrank in size (0.8° × 0.8° to 0.4° × 0.4°) after a 1,400-ms delay period. Participants performed three conditions with 200 trials per condition. The test SS4 and no-test SS4 conditions were randomly intermixed in one block. The test SS1 condition was run in a separate block. The order of the two blocks was counterbalanced across participants.

Fig. 4.

Stimuli and event-related potential (ERP) results from Experiment 4. In (a) we show an example of a trial in the test SS4 and no-test SS4 conditions. An auditory cue indicated whether a test square would be presented or not. Grand average ERP difference waves (contralateral delay activity, or CDA) time-locked to the sample array presentation are shown in (b). Error bands show the standard errors of the mean. The gray rectangle indicates the time period for the sample array. The yellow bar shows the time period for the auditory cue. The pink and green rectangles indicate the time period for the test display of the corresponding condition. The brown bars show statistical comparisons of the time course of CDA amplitudes among each pair of conditions. SS1 = set-size 1; SS4 = set-size 4.

Results

As shown in Figure 4b, the difference in CDA amplitudes between the no-test SS4 and the test SS4 conditions became statistically significant 300 ms after the auditory cue, M_{no-test SS4 – test SS4} = 0.5 μV, t(27) = 2.92, p < .01, d = 0.55, 95% CI = [0.2, 0.9]. In addition, the CDA amplitude of the no-test SS4 condition became statistically indistinguishable with that of the test SS1 condition 280 ms after the auditory cue, M_{no-test SS4 – test SS1} = −0.1 μV, t(27) = 0.30, p = .77, d = 0.13, 95% CI = [−0.5, 0.4]. The CDA amplitude of the test SS4 condition was more negative than that of the test SS1 condition over the delay period, M_{test SS4 – test SS1} = −0.6 μV, t(24) = 3.43, p < .005, d = 0.74, 95% CI = [−0.9, −0.3]. These results suggest that WM item removal is triggered as soon as WM contents are deemed unnecessary even in the middle of a task. In addition, the CDA decreased without perceiving any new visual information. In Experiment 1, the CDA began to decrease less than 1 s after the participants responded to the test, but the amplitude did not go lower than that in SS1. This left the possibility that the reduced CDA might have been the CDA elicited by the single test item. However, the results in Experiment 4 showed that the CDA decreased as much as in Experiment 1 without participants perceiving any new visual information. This ruled out the possibility that no-longer-relevant items were pushed out of visual WM by the visual test item.

Experiment 5

In Experiment 5, we sought further behavioral evidence to support the CDA results in Experiment 4. We hypothesized that, if the CDA reflects the number of items currently held in visual WM, participants should fail change detection when tested 1 s after the auditory cue indicated the WM contents became obsolete because of task termination.

Method

Participants

One hundred and twenty-one university students (76 females, ages 18–22 years) participated. The sample size was determined in the same way as in Experiment 3. Seventeen participants (14%) were excluded from the analysis for the following reasons: 9 participants anticipated that there would be an unexpected test despite instructions; 4 participants responded to the unexpected test extremely slowly (> 3 SD from the average); and 4 participants considered the unexpected test a bug in the computer program and called the experimenter before they responded to the test.

Stimuli and experimental procedure

The experiment was identical to Experiment 3 with the following exceptions. Participants performed a simple color-change-detection task for a single block of 61 trials. Four colored squares were presented for 100 ms. After a 1,400-ms blank delay period, a single test square appeared. Participants reported whether the test color was identical to the corresponding sample square (50% of trials) by pressing the “F” or “J” key on a keyboard as quickly as possible. In a random half of the trials, an auditory cue (650 Hz) was presented for 100 ms after the offset of the sample colored squares, with an interstimulus interval (ISI) of 400 ms. The auditory cue indicated that the test would not be presented. A 900-ms blank period followed the auditory cue, and the trial ended. We simply informed participants that the auditory cue would indicate that the test would not be presented, giving no instructions about forgetting. The auditory cue was 100% valid until the 60th trial, during which we estimated participants’ baseline visual WM performance. The 61st trial served as a critical trial in which an unexpected memory test followed 900 ms after the offset of the auditory cue. To prompt participants to respond to the unexpected test, we instructed them at the beginning of the experiment to continue the task until they saw “You completed the block” on the computer screen. After the participants responded to the unexpected test, they rated a degree of surprise to the unexpected test with a questionnaire sheet, as in Experiment 3. The participants also reported whether they had followed the instructions by checking either “I expected no test when the auditory cue was delivered” or “I prepared for the test regardless of the auditory cue.”

Results

As shown in Figure 5, the mean accuracy was 88% in the baseline trials (from the first to the 60th trial), 95% CI = [86%, 89%]. At the 61st trial, the auditory cue indicated that the test display would not be presented, but an unexpected test was presented. We assumed that if none of the participants removed no-longer-necessary items from WM after the auditory cue, the percentage of participants who correctly responded to the unexpected test should be equal to the mean accuracy of the baseline trials. The results in Figure 5 showed that only 68% of the participants correctly responded to the unexpected test, 95% CI = [58%, 77%], which was significantly lower than the mean accuracy of the baseline trials (exact binomial test, p < .001).⁵ Computing visual WM capacity (K) with the percentage of correct participants, we found that 1.5 items were held in visual WM for the unexpected test, 95% CI = [0.7, 2.2].⁴ This low capacity was not due to surprise or response time to the unexpected test, as neither was significantly different between participants with correct and incorrect responses—surprise score: M_{correct participants} = 4.0, M_{incorrect participants} = 4.4, t(98) = 1.06, p = .29, d = 0.23, 95% CI_{correct participants} [4.1, 4.5], 95% CI_{incorrect participants} = [3.8, 4.5]; response time: M_{correct participants} = 3,550 ms, M_{incorrect participants} = 4,145 ms, t(102) = 1.16, p = .25, d = 0.25, 95% CI_{correct participants} = [2,954, 4,146], 95% CI_{incorrect participants} = [3,393, 4,895].

Fig. 5.

Behavioral results from Experiment 5. The blue bar shows accuracy in baseline trials (from the first to the 60th trial). The red bar shows the percentage of participants with a correct response for the unexpected change-detection task (in the 61st trial). The error bars represent 95% confidence intervals.

Discussion

Our study provides novel evidence that visual WM content is spontaneously removed after it becomes obsolete. More precisely, we demonstrated that the CDA, a neural correlate of WM maintenance, began to decrease less than 1 s after its contents were used for the current task’s goal. Consistent with this neural measure, once the WM content became no longer necessary, participants could not recall it even though they could have just 1 s earlier. Critically, this information removal occurred only when participants deemed WM content no longer necessary. We also replicated these findings while controlling for visual stimuli input and interference, assuring that the WM content removal was not due to remembering new items or reallocating WM resources to other visual information.

This goal-driven WM content removal extends previous demonstrations of competition-based removal from WM. Previous studies have shown that no-longer-relevant items can be removed from visual WM if attention is strategically biased away from these items and toward another to-be-remembered item (Astle et al., 2012; Dames & Oberauer, 2022; Williams et al., 2013). However, it remained unclear whether no-longer-relevant information can be removed from WM without the presence of competing task-relevant information. Our results directly demonstrated that such strategic attentional allocation is not necessary to remove no-longer-relevant representations from WM. Interestingly, the spontaneous WM removal occurred ~700 ms after task completion. This is far faster than passive forgetting by decay and interference (~10 s), which occurs when the participant actively engages in WM maintenance (Ricker et al., 2014; Zhang & Luck, 2009). Thus, it seems likely that the spontaneous WM content removal we demonstrated is separate from the passive forgetting that occurs as a failure of WM maintenance.

Compatible findings with goal-directed WM item removal have been reported in paradigms that utilized attentional capture (Oberauer, 2001; Sasin et al., 2017) and retro cues (Ecker et al., 2014; Souza et al., 2014). For instance, in an attentional-capture paradigm, no-longer-relevant WM items do not capture attention in a visual-search task after a directed-forgetting instruction. In a retro-cue paradigm, when the retro-cue indicates one of the items in WM as relevant and others as no longer irrelevant, it becomes easier and faster to add another item to WM. These observations might suggest that no-longer-relevant items are removed from WM, but it is also possible that these items remained in WM in an activity-silent state (Awh & Vogel, 2020; Lewis-Peacock et al., 2018) so that they could have been called upon if a subsequent task required it (i.e., a surprise second test). Our study directly tested this possibility by asking participants to retrieve the information in WM immediately after it was used. Participants’ failure in doing so ensures that no-longer-relevant information was, indeed, removed from visual WM.

Additionally, spontaneous WM removal is in sharp contrast to forgetting in long-term memory. Previous studies have demonstrated that it is extremely difficult to forget items from long-term memory without biasing attention toward another to-be-remembered item (Bjork, 1989; Conway et al., 2000; Lee et al., 2007; MacLeod & Macrae, 2001; Tozios & Fukuda, 2020), and thus, that the interference between to-be-remembered items and to-be-forgotten items is necessary to forget unwanted items in long-term memory. In contrast, our study showed that WM removal does not necessitate such an indirect process. Together with previous studies, our results elucidate the flexible, goal-directed nature of WM.

Last, the present study has two limitations to be examined in future studies. First, we demonstrated that removal of no-longer-relevant information from WM is possible without the need to remember an external item or reallocate WM resources to other visual information. However, a participant’s own internal thoughts or tasks, including mind wandering or pressure for the workspace of WM to be used for something else, may act as a competitive source of WM contents. Second, although the removal of no-longer-relevant items was initiated as soon as they became obsolete, it was not complete in less than 1 s. This observation suggests the following two possibilities. First, complete removal is possible, but it simply takes longer than 1 s. Alternatively, the removal might be completed at different times across trials, and it might even fail probabilistically in some of them. Future study is needed to tease apart these hypotheses.

In sum, we have demonstrated that WM is a flexible, goal-directed memory mechanism that maintains information as long as it is necessary for the task at hand. Once it becomes unnecessary, WM content is largely removed without relying on strategic interference with additional task-relevant information. This removal mechanism is critical to explain how capacity-limited WM can keep up with a dynamic environment in which task-relevant information is consistently changing.

Supplemental Material

sj-pdf-1-pss-10.1177_09567976241246709 – Supplemental material for Task Termination Triggers Spontaneous Removal of Information From Visual Working Memory

Supplemental material, sj-pdf-1-pss-10.1177_09567976241246709 for Task Termination Triggers Spontaneous Removal of Information From Visual Working Memory by Hiroyuki Tsubomi, Keisuke Fukuda, Atsushi Kikumoto, Ulrich Mayr and Edward K. Vogel in Psychological Science

Footnotes

Acknowledgements

We thank Shiho Kawata, Mika Sasa, and Yoshiko Demura for assistance in data collection, and Richard Matullo for English language editing.

Transparency

Action Editor: Krishnankutty Sathian

Editor: Patricia J. Bauer

Author Contributions

Hiroyuki Tsubomi: Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Software; Validation; Visualization; Writing – original draft; Writing – review & editing.

Keisuke Fukuda: Conceptualization; Funding acquisition; Investigation; Methodology; Writing – original draft; Writing – review & editing.

Atsushi Kikumoto: Conceptualization; Investigation; Methodology; Writing – review & editing.

Ulrich Mayr: Conceptualization; Writing – review & editing.

Edward K. Vogel: Conceptualization; Funding acquisition; Writing – review & editing.

ORCID iDs

Hiroyuki Tsubomi

Ulrich Mayr

Supplemental Material

Additional supporting information can be found at

Notes

References

Astle

D. E.

Summerfield

Griffin

Nobre

A. C.

(2012). Orienting attention to locations in mental representations. Attention, Perception, & Psychophysics, 74, 146–162. https://doi.org/10.3758/s13414-011-0218-3

Awh

Vogel

E. K.

(2020). Online and off-line memory states in the human brain. In Poeppel

Mangun

G. R.

Gazzaniga

M. S.

(Eds.), The cognitive neurosciences (pp. 347–356). MIT Press.

Backer

K. C.

Alain

(2012). Orienting attention to sound object representations attenuates change deafness. Journal of Experimental Psychology: Human Perception and Performance, 38, 1554–1566. https://doi.org/10.1037/a0027858

Bancroft

T. D.

Hockley

W. E.

Farquhar

(2013). The longer we have to forget the more we remember: The ironic effect of postcue duration in item-based directed forgetting. Journal of Experimental Psychology: Learning, Memory, and Cognition, 39, 691–699. https://doi.org/10.1037/a0029523

Bjork

R. A.

(1989). Retrieval inhibition as an adaptive mechanism in human memory. In Roediger

H. L.

Craik

F. I. M.

(Eds.), Varieties of memory and consciousness: Essays in honour of Endel Tulving (pp. 309–330). Erlbaum.

Conway

M. A.

Harries

Noyes

Racsma’ny

Frankish

C. R.

(2000). The disruption and dissolution of directed forgetting: Inhibitory control of memory. Journal of Memory and Language, 43, 409–430. https://doi.org/10.1006/jmla.2000.2706

Cowan

(1999). An embedded-processes model of working memory. In Miyake

Shah

(Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 62–101). Cambridge University Press. https://doi.org/10.1017/CBO9781139174909.006

Cowan

(2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–185. https://doi.org/10.1017/S0140525X01003922

Dames

Oberauer

(2022). Directed forgetting in working memory. Journal of Experimental Psychology: General, 151(12), 2990–3008. https://doi.org/10.1037/xge0001256

10.

Delorme

Makeig

(2004). EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of Neuroscience Methods, 134, 9–21. https://doi.org/10.1016/j.jneumeth.2003.10.009

11.

Drew

Horowitz

T. S.

Wolfe

J. M.

Vogel

E. K.

(2012). Neural measures of dynamic changes in attentive tracking load. Journal of Cognitive Neuroscience, 24, 440–450. https://doi.org/10.1162/jocn_a_00107

12.

Ecker

U. K. H.

Oberauer

Lewandowsky

(2014). Working memory updating involves item-specific removal. Journal of Memory and Language, 74, 1–15. https://doi.org/10.1016/j.jml.2014.03.006

13.

Fukuda

Kang

M.-S.

Woodman

G. F.

(2016). Distinct neural mechanisms for spatially lateralized and spatially global visual working memory representations. Journal of Neurophysiology, 116, 1715–1727. https://doi.org/10.1152/jn.00991.2015

14.

Fukuda

Vogel

E. K.

(2011). Individual differences in recovery time from attentional capture. Psychological Science, 22, 361–368. https://doi.org/10.1177/0956797611398493

15.

Hakim

Adam

K. C. S.

Gunseli

Awh

Vogel

E. K.

(2019). Dissecting the neural focus of attention reveals distinct processes for spatial attention and object-based storage in visual working memory. Psychological Science, 30, 526–540. https://doi.org/10.1177/0956797619830384

16.

Katus

Andersen

S. K.

Müller

M. M.

(2012). Nonspatial cueing of tactile STM causes shift of spatial attention. Journal of Cognitive Neuroscience, 24, 1596–1609. https://doi.org/10.1162/jocn_a_00234

17.

Kozak

Herz

Bar-Haim

Censor

(2021). Indirect modulation of human visual memory. Scientific Reports, 11, Article 7274. https://doi.org/10.1038/s41598-021-86550-2

18.

Lee

Y.-S.

Lee

H.-M.

Tsai

S.-H.

(2007). Effects of post-cue interval on intentional forgetting. British Journal of Psychology, 98, 257–272. https://doi.org/10.1348/000712606X120410

19.

Lewis-Peacock

J. A.

Kessler

Oberauer

(2018). The removal of information from working memory. Annals of the New York Academy of Sciences, 1424, 33–44. https://doi.org/10.1111/nyas.13714

20.

Lopez-Calderon

Luck

S. J.

(2014). ERPLAB: An open-source toolbox for the analysis of event-related potentials. Frontiers in Human Neuroscience, 8, Article 213. https://doi.org/10.3389/fnhum.2014.00213

21.

Luck

S. J.

(2014). An introduction to the event-related potential technique (2nd ed.). MIT Press.

22.

Luck

S. J.

Vogel

E. K.

(1997). The capacity of visual working memory for features and conjunctions. Nature, 390, 279–281. https://doi.org/10.1038/36846

23.

MacLeod

M. D.

Macrae

C. N.

(2001). Gone but not forgotten: The transient nature of retrieval-induced forgetting. Psychological Science, 12, 148–152. https://doi.org/10.1111/1467-9280.00325

24.

Muris

Merckelbach

de Jong

P. F.

(1993). Verbalization and environmental cuing in thought suppression. Behaviour Research and Therapy, 31, 609–612. https://doi.org/10.1016/0005-7967(93)90112-8

25.

Ngiam

W. X. Q.

Adam

K. C. S.

Quirk

Vogel

E. K.

Awh

(2021). Estimating the statistical power to detect set-size effects in contralateral delay activity. Psychophysiology, 58, Article e13791. https://doi.org/10.1111/psyp.13791

26.

Oberauer

(2001). Removing irrelevant information from working memory: A cognitive aging study with the modified Sternberg task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 948–957. https://doi.org/10.1037/0278-7393.27.4.948

27.

Pertzov

Bays

P. M.

Joseph

Husain

(2013). Rapid forgetting prevented by retrospective attention cues. Journal of Experimental Psychology: Human Perception and Performance, 39, 1224–1231. https://doi.org/10.1037/a0030947

28.

Ricker

T. J.

Spiegel

L. R.

Cowan

(2014). Time-based loss in visual short-term memory is from trace decay, not temporal distinctiveness. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40, 1510–1523. https://doi.org/10.1037/xlm0000018

29.

Sasin

Morey

C. C.

Nieuwenstein

(2017). Forget me if you can: Attentional capture by to-be-remembered and to-be-forgotten visual stimuli. Psychonomic Bulletin & Review, 24(5), 1643–1650. https://doi.org/10.3758/s13423-016-1225-0

30.

Shipstead

Engle

R. W.

(2013). Interference within the focus of attention: Working memory tasks reflect more than temporary maintenance. Journal of Experimental Psychology: Learning, Memory, and Cognition, 39, 277–289. https://doi.org/10.1037/a0028467

31.

Souza

A. S.

Oberauer

(2015). Time-based forgetting in visual working memory reflects temporal distinctiveness, not decay. Psychonomic Bulletin & Review, 22, 156–162. https://doi.org/10.3758/s13423-014-0652-z

32.

Souza

A. S.

Rerko

Oberauer

(2014). Unloading and reloading working memory: Attending to one item frees capacity. Journal of Experimental Psychology: Human Perception and Performance, 40, 1237–1256. https://doi.org/10.1037/a0036331

33.

Tozios

C. J. I.

Fukuda

(2020). Indirect, but not direct, down-regulation of visual long-term memory encoding through strategic biasing of attentional allocation. Journal of Experimental Psychology: General, 149, 1294–1310. https://doi.org/10.1037/xge0000712

34.

Tsubomi

Fukuda

Watanabe

Vogel

E. K.

(2013). Neural limits to representing objects still within view. The Journal of Neuroscience, 33, 8257–8263. https://doi.org/10.1523/JNEUROSCI.5348-12.2013

35.

Umemoto

Inzlicht

Holroyd

C. B.

(2019). Electrophysiological indices of anterior cingulate cortex function reveal changing levels of cognitive effort and reward valuation that sustain task performance. Neuropsychologia, 123, 67–76. https://doi.org/10.1016/j.neuropsychologia.2018.06.010

36.

Underwood

B. J.

(1957). Interference and forgetting. Psychological Review, 64, 49–60. https://doi.org/10.1037/h0044616

37.

Vogel

E. K.

Machizawa

M. G.

(2004). Neural activity predicts individual differences in visual working memory capacity. Nature, 428, 748–751. https://doi.org/10.1038/nature02447

38.

Vogel

E. K.

McCullough

A. W.

Machizawa

M. G.

(2005). Neural measures reveal individual differences in controlling access to working memory. Nature, 438, 500–503. https://doi.org/10.1038/nature04171

39.

Vogel

E. K.

Woodman

G. F.

Luck

S. J.

(2006). The time course of consolidation in visual working memory. Journal of Experimental Psychology: Human Perception and Performance, 32, 1436–1451. https://doi.org/10.1037/0096-1523.32.6.1436

40.

Wegner

D. M.

Schneider

D. J.

Carter

S. R.

White

T. L.

(1987). Paradoxical effects of thought suppression. Journal of Personality and Social Psychology, 53, 5–13. https://doi.org/10.1037/0022-3514.53.1.5

41.

Williams

J. R.

Robinson

M. M.

Schurgin

M. W.

Wixted

J. T.

Brady

T. F.

(2022). You cannot “count” how many items people remember in visual working memory: The importance of signal detection–based measures for understanding change detection performance. Journal of Experimental Psychology: Human Perception and Performance, 48, 1390–1409. https://doi:10.1037/xhp0001055

42.

Williams

Hong

S. W.

Kang

M.-S.

Carlisle

N. B.

Woodman

G. F.

(2013). The benefit of forgetting. Psychonomic Bulletin & Review, 20, 348–355.

43.

Adam

K. C. S.

Fang

Vogel

E. K.

(2018). The reliability and stability of visual working memory capacity. Behavior Research Methods, 50, 576–588. https://doi.org/10.3758/s13428-017-0886-6

44.

Zhang

Luck

S. J.

(2009). Sudden death and gradual decay in visual working memory. Psychological Science, 20, 423–428. https://doi.org/10.1111/j.1467-9280.2009.02322.x

45.

Zhao

Vogel

Awh

(2023). Change localization: A highly reliable and sensitive measure of capacity in visual working memory. Attention, Perception, & Psychophysics, 85, 1681–1694. https://doi.org/10.3758/s13414-022-02586-0

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.35 MB