Maintenance of invisible stimuli in visual short-term
memory
Maxine Gorter
University: Universiteit van Amsterdam Student number: 10733140
Mentor: Timo Stein
Subject: Bachelor Project Brain & Cognition Study: Psychology
Date: 2-5-2017 Words: 4708
Abstract
Generally, it is believed that the contents of visual short-term memory are conscious contents. A series of studies recently challenged this view by showing that details of visual images that were labeled as invisible could still be retained in short-term memory. Soto et al. (2011) claimed to found evidence for unconscious working memory (WM) which is conflicting with the classical view on WM and consciousness. In the current study, we tried to replicate Soto’s study with addition of a perception test. The results of the current did not show above chance performance for subjectively invisible stimuli, meaning that Soto et al. (2011) could not be replicated. Furthermore, we found better performance in the perception task than in a memory task. Which is suggesting that the visual information fades out over time. These results are supporting the traditional view on the relationship between WM and consciousness rather than the novel views.
Maintenance of invisible stimuli in visual short-term memory
Doing a basic task like rehearsing a phone number requires both working memory (WM) and consciousness. This assumption is the classical view on the relationship between WM and consciousness. Lately, there is a debate going on whether WM is the same as conscious experience. There has been found evidence for unconscious processing in the human WM (Soto & Silvanto, 2014). As shown in the phone number example, both WM as consciousness are important in daily life and are therefore a hot topic in research. WM is conceived as an information store and control system to meet behavior goals (Soto & Silvanto, 2014). WM addresses to the short-term maintenance of information in service of goal-directed behavior. The information is stored ‘on-line’ allowing to be easily accessed for other cognitive processes. One of the key features of WM is that accessing, manipulating and examination of its content can be conscious (Jacobs & Silvanto, 2014).
On the other hand, the term consciousness has multiple meanings and most of them are difficult to precisely define in a manner which is liable to experimentation (Dehaene & Changeux, 2011). According to Dehaene (2014) there are four signatures of consciousness. First, the ignition of parietal and prefrontal circuits which is caused by a conscious stimulus. Second, the slow wave called P3 in the EEG signal which is accompanied by conscious access. High-frequency oscillations triggered by conscious ignition are the third signature. And finally, there is a global brain web formed by regions who exchange bidirectional and synchronized long distance messages. These four signatures together are the neural correlates of consciousness (NCC). But wat is or does consciousness? Deheane (2014) proposed a global workspace theory. According to this theory is the brain-wide sharing of information what we experience as consciousness. Whenever information access the workspace, it will become consciousness. These views on WM and consciousness are assuming these two concepts are tightly coupled. This is because WM is implicated in the theory of consciousness
mechanisms. As in the global workspace theory, WM processing also involves a broadcast of information which is similar to the mentioned “conscious ignition” in the third signature of consciousness (Baars & Franklin, 2003).
The exact relation between WM and consciousness in undetermined. In the literature, there are four models describing WM and the relationship with consciousness (Jacobs & Silvanto, 2015). The first model is the standard view that WM content is always conscious. It states that all the active components of WM are accurate reportable. This accurate report is also the standard operation of consciousness. Following these arguments WM should be the same as conscious experience. This view corresponds with the global workspace theory (Jacobs & Silvanto, 2015; Dehaene, 2014). The second model states conscious experience is not equal tothe WM content. In this view, the content of WM remains unconscious until the attention is focused on the representation of conscious input. On the other hand, the third viewpoint states that WM can operate without conscious awareness. There is evidence that some items can be in WM and remain unconscious even when they are attended (Soto, Mäntylä & Silvanto, 2011). There has been a series of experiments where is shown that subliminal stimuli can be encoded and maintained and used for later retrieval (Jacobs & Silvanto, 2014). According to Soto and Silvanto (2014) WM can be called unconscious if the operational definitions that characterize WM are met in a cognitive task without awareness. Contrary to the other models, the last model states that consciousness is not directly based on the content of WM at all. The findings of Bona et al. (2013) suggest that WM maintenance and conscious awareness are separate representations. The first two models are broadly accepted and assumed as the standard models of the relationship of WM and conscious experience. The last two models are novel: they are supposing that there could be WM processes without consciousness.
The current study is based on the work of Soto et al. (2011). They did notable research on the question whether WM can operate without consciousness. Supporting the third viewpoint, they showed that observers could encode a subjectively invisible orientation cue and maintained it on-line even in the presence of visible distracters. Soto et al. (2011) did four experiments to get to their conclusions. Here only the first experiment will be discussed because the results of this experiments are the most remarkable. With their experiment, they showed that discrimination performance of the observers on unaware stimuli trials were above chance. Participants performed above chance in an explicit discrimination task. During this task, participants were presented with a brief masked orientation cue followed by a delay period and a test. In the test, the participants had to determine how the probe was orientated relative to the first cue. After the test, the participants gave subjective ratings for the visibility of the first cue on an 1-4 awareness scale where a rating of “1” was “did not see anything”. Soto et al. (2011) only included the trials were the stimulus was subjectively invisible, trials with awareness rating “1”, in their analysis across experiments. By implementing an objective discrimination test and the ratings of an awareness scale, Soto et al. (2011) used both objective and subjective measures. With the findings of these experiments Soto et al. (2011) argue that visual memory can encode, maintain and access unconscious content for explicit discrimination goals.
Although there seems to be some evidence for the revolutionary models of the relationships of WM and conscious content shown by Soto et al. (2011), there is also some critique on the manner how Soto et al. (2011) did their experiments. Stein, Kaiser & Hesselmann (2016) argue that these findings can be accommodated by conventional conceptions of the conscious perception and conscious WM. The evidence for non-conscious WM has been inferred from a dissociation between the subjective and the objective measure. In those trials where participants reported “1” on the awareness scale (“did not see
anything”), they showed an above-chance performance in the memory test. The rating on the awareness scale refers to a subjective measure and the performance on the memory test to an objective measure. The dissociation between the subjective and objective measures was assumed as non-conscious input in WM for later use in the memory test. According to Stein, Kaiser and Hesselman (2016) such dissociations are often used in the perception of low-visibility stimuli, but do not need to reflect non-conscious WM. The subjective ratings of stimulus visibility are susceptible to response biases, or the observers may systematically indicate partly seen stimuli as unseen. Therefore, it could reflect residual awareness of the memory cue instead of maintenance of non-conscious stimuli in WM.
Another important point of discussion in the results of Soto et al. (2011) is that there was no correlation between cue sensitivity (d’) and memory discrimination. This result suggest that the visibility of the cue does not influence the performance on the discrimination test. If the performance is not affected by the visibility of the cue, there might be a chance that the awareness rating after the test does not correspond with the first cue. Soto et al. (2011) described these analyses as being derived from signal detection theory (SDT). However, According to Stein, Kaiser & Hesselmann (2016) these analyses are not grounded in SDT and do not represent valid measures of detection sensitivity. Stein, Kaiser & Hessemann (2016) argue that the detection sensitivity is not invariant to bias but changes as a function of the decision criterion of the observer. Therefore, for conservative observers, their aberrant way of estimating detection sensitivity underestimates the true detection sensitivity.
A last thing worth noticing is that Soto et al. (2011) did not use the regular awareness scale to measure conscious experience. In their paper, they claim to use the perceptual awareness scale (PAS), but they used a different PAS as developed by Overgaard, Rote, Mouridsen and Ramsøy (2006). When using PAS, it is intended to let subjects report introspectively, instead of reporting about stimulus features they are to report what they
experience: 1= No experience, 2=brief glimpse, 3= almost clear experience, 4= clear experience. This is different from the “PAS”-scale used in Soto et al. (2011): 1= did not see anything, 2= maybe saw something, 3= saw the stimulus but not its orientation, 4= saw the stimulus and its orientation. This is an important difference because it is possible that the subjects would rate their subjective awareness different with a different scale. They could, for example, rate the same stimuli as more visible in the PAS-scale in comparison with the Soto-scale. This could influence the interpretation of the data.
Summarizing al above, Soto et al. (2011) claim to have found evidence for nonconscious working memory. If these results could be replicated it would mean that the standard view on the relation between WM and consciousness is violated. But there are some points of critique which make this claim questionable. Firstly, Stein, Kaiser & Hesselman (2016) mention the probability of response bias of subjects. Then second, there was an incorrect calculation of the cue sensitivity. Finally, Soto et al. (2011) used a different PAS scale which could have influenced the results. To reexamine the relation between WM and consciousness properly, we will replicate the experiment of Soto et al. (2011) and in addition we will do a perception test. The perception test will be right after the first cue, so that we can scale the visibility and orientation. By adding a perception test right after the first cue we can compare this visibility ratings with the visibility ratings after the probe and examine whether those later ratings are based on unconscious memory or if they are just guesses. Besides, adding a perception test enables us to compare the discrimination performance right after the cue and after a delay and examine whether this delay effects the accuracy of discrimination. We expect to find above-chance memory performance even in trials where participants reported no subjective awareness of the memory cue, as found in Soto et al (2011). Furthermore, we expect higher performance in the perception-only condition, as no
maintenance of unconscious or weakly conscious information is required. Finally, we expect to find similar visibility ratings in both conditions.
Methods
Participants
A total of 16 healthy volunteers (5 males, mean age: 21,2 and 11 females, mean age: 21,5) were recruited from the University of Amsterdam. The study conformed to the ethical standards of the University of Amsterdam. Participants provided informed consent and were rewarded afterwards with 1 research credit per hour. All participants followed the same standardized experiments. People with eye disorders (glasses or contact lenses are allowed), color blindness and problems with depth view were excluded.
Materials
We used a computer with Matlab R2017a (9.2.0.538062) 64-bit (maci64) with Psychophysics Toolbox Version 3 (PTB-3) to control the stimulus displays and responses. Stimuli were presented on a 24-inch monitor with a refresh rate of 60-Hz.
Procedure
To replicate and extend the findings of Soto et al. (2011) the participants were instructed to do two experiments. In both the experiments, each trial began with a black fixation cross in the middle of the screen for 100 ms, followed by a blank screen for 500 ms and the memory cue. The memory cue was shown for 16.7 ms. The memory cue was a Gabor grating with 10% contrast (spatial frequency: 1 cycle/deg and diameter of 4 degree of visual
angle from a view in distance of 57 cm). The grating could be tilted 10, 40, 70, 100, 130 and 160 degrees to the right from the vertical. On each trial, the cue was randomly selected. The memory cue was masked by a 15% circular grating mask. Both Experiments included catch trials (50% of all trials) on which no memory cue was presented.
In Experiment 1 participants were instructed to remember the memory cue’s orientation over a delay of 2 seconds, and compare the orientation to the probe. The probe was a similar Gabor grating and was tilted 30 degrees left or right relative to the memory cue. Participants were asked to indicate the direction of the tilt of the probe relative to the memory cue via button press, in a non-speeded response, as accurately as possible. Finally, they rated their subjective awareness of the memory cue with the following rating: 1= did not see anything, 2= maybe saw something, 3= saw the stimulus but not its orientation, 4= saw the stimulus and its orientation (see Figure 1).
Experiment 2 was a perception-only experiment. Here participants were presented only with the memory cue, followed by a circular grating mask. The participants were asked to discriminate the cue’s orientation relative to a vertical line, without any temporal delay via button press, in a non-speeded response, as accurately as possible. After the discrimination participants were asked to rate their subjective awareness of the cue with the same rating scale as in Experiment 1 (see Figure 2).
The participants did both experiments twice in counterbalanced order. The experiments contained 144 trails leading to a total of 576 trials (288 perception experiment trials and 288 in the memory experiment). After every 72 trials, there was a pause of at least 10 seconds. Participation took about 1.5 hour.
Prior to each of the experiments, participants fulfilled two blocks of practice, each containing 4 trials. In the first block the duration of the memory cue was slower compared to the main experiment, to get used to the lay-out of the experiment. In the second block, it was
similar to the duration of experiment 1 and 2. The practice trials on each of the experiment used the same cue-target delay as in the actual experiment trials. The instructions given to the participants regarding the cues were as follows: “The memory cue can be difficult to see. When you think you didn’t see the memory cue, try to follow your intuition and do your best in the memory task. If you are sure that you didn’t see anything just before the circular mask when the memory cue should appear, press “1” in response to the awareness question, but still try your best in the memory task.”
Figure 2. Timeline of an experimental trial in Experiment 2.
Results
A total of 16 subjects participated to the experiments. There was no dropout of
subjects. The data of all subjects is used for the statistical analyses. By analyzing the data, we handled a significance criterion of α = 0.05. The following section contains two parts. In the first part, the visibility ratings will be examined. The proportion of visibility ratings will be plotted in a graph. And there will be examined whether the average visibility rating was influenced by condition and presence of the stimuli. Finally in section 1.2., the d-prime, a sensitivity measure of visibility, of both conditions will be compared.
In second part, the main analyses of WM and perception performance will be executed. First in section 2.1., will be examined whether the accuracy was influenced by condition and visibility. Then in section 2.2., will be examined whether the participants performed above chance on the discrimination task.
1. Visibility ratings
Fist, the proportion of responses will be displayed. The proportion of responses of experiment 1 (memory) are plotted in Figure 3 and the proportion of responses in experiment 2 (perception) are plotted in Figure 4. In both figures can be seen that the proportion pressed 1 “Did not see anything” is higher for absent cues than for present cues. Furthermore, the proportion pressed 4 “Saw the stimulus and its orientation” is higher for present cues than for absent cues.
Table 1
Mean proportion of responses across all experiments as a function of awareness response. The values in brackets are the standard deviations.
Figure 3. Bar graph of the mean proportion of awareness responses in Experiment 1
(Memory).
Cue Present Cue Absent
Response 1 2 3 4 1 2 3 4 Memory 0,40 (0,23) 0,34 (0,18 ) 0,07 (0,07) 0,19 (0,18) 0,72 (0,18 ) 0,21 (0,13) 0,03 (0,06) 0,04 (0,07) Perception 0,32 (023) 0,30 (0,21 ) 0,06 (0,08) 0,32 (0,25) 0,72 (0,17 ) 0,18 (0,12) 0,03 (0,04) 0,07 (0,11)
1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Visibility rating P ro p or ti on
Figure 4. Bar graph of the mean proportion of awareness responses in Experiment 2
(Perception).
To examine whether there was effect for the memory or perception condition and the presence of the cue on the average rating of the cue, we run a repeated measures ANOVA on the average awareness rating. There were two within-subject variables (cue absent or present and memory versus perception). The dependent variable was the average awareness rating. The assumptions for this test were normality and sphericity. Because the repeated-measures variables have only two levels the assumption for sphericity is met (Field, 2013). Normality was checked by the Shapiro-Wilk test. This assumption was only violated in the memory condition when the cue was absent W (16) = .859, p = .018. Whereas the data is robust for normality, the violation of this assumption is not disadvantageous for the primary analysis (Field, 2013). The Shapiro-Wilk test was not significant for the other conditions. In Table 2 the conditions, the number of participants, the mean score and the standard deviation (SD) are displayed.
Table 2
The Number of Participants, the Mean and the Standard Deviation (SD) of the Ratings per Condition
Condition Presence N Mean rating SD
Memory PresentAbsent 1616 2,051,38 ,54,32
Perception PresentAbsent 1616 2,381,44 ,69,39
The results are showing a main effect for condition F (1,15) = 12.79, p = .003. Reflecting higher mean ratings in the perception condition than in in the memory condition (mean memory = 1.72, SD = .33 and mean perception = 1.91, SD = .45). Second, there was a main effect for presence of the cue F (1,15) = 31.6, p < .001. Reflecting higher mean rating for present cues than for absent cues. Finally, there is an interaction effect of condition and presence of the cue F (1,15) = 15,50, p = .001. This interaction effect means that for present cues the increase of mean awareness rating (from memory to perception) is higher than for absent cues (see Figure 5). A paired simple t-test illustrates that for absent stimuli: there was no significant difference between memory and perception t (1,15) = 1.21, p = .245. For present cues, mean ratings were significant higher in the perception than memory condition t (1,15) = 4.33, p = .001. These results are partly in line with our expectations.
Figure 5. Mean awareness rating for presence of cue.
Figure 5 illustrates a significant difference in mean awareness rating for cue presence
in the memory condition.
1.2. D-prime
D-prime is measured whereas a hit is a cue-present trial with an awareness rating >1 and a False Alarm is a cue-absent trial with an awareness rating >1. To get an objective measure of cue visibility we run an independent t-test and examined whether d-prime significantly differs from 0. The normality assumption was met. The results show that the d-prime scores of both memory and perception significantly differ from 0. For d-prime memory t (1,15) = 2.73, p = . 015 and for d-prime perception t = (1,15) = 3.17, p = .006.
To examine whether d-prime memory significantly differs from d-prime perception we run a paired simple t-test. The assumptions for this test are the test are the same as the previous test and were met. The results show that the d-prime memory significantly differs from d-prime perception t (1,15) = -2.14, p = .049. Meaning that d-prime in the perception condition is significantly higher than in the memory condition (d-prime memory= .99, SD= . 72 and d-prime perception = 1.22, SD= .90).
2. Working Memory
2.1. Does the accuracy differ for condition (memory vs perception) and cue visibility (invisible vs visible)?
To investigate whether there was an effect for memory or perception condition and visibility of the cue on accuracy of responding, we run a repeated measures ANOVA on the accuracy on the discrimination task. There were two within-subject variables: condition (memory versus perception) and visibility (visible versus invisible). The dependent variable was the accuracy measurement. The assumptions for this test are the same as in section 1.1. Again, the assumptions were met. In Table 3 the conditions, mean accuracy, the number of participants and the standard deviation are displayed.
Table 3
The Number of Participants, the Mean Accuracy and the Standard Deviation (SD) per condition.
Condition Visibility N Mean Accuracy SD
Memory InvisibleVisible 1616 ,505,774 ,058,14
Perception Invisible 16 ,623 ,15
The results are showing a main effect for condition F (1,14) = 18.5, p = .001. Reflecting higher accuracy in the perception condition than in the memory condition. Second, there was a main effect for visibility of the cue F (1,14) = 103.5, p < .001. Reflecting a higher accuracy for visible than for invisible stimuli. Finally, there is no interaction effect found for condition and presence of the cue F (1,14) = .035, p = .855 (See Figure 6). This means that the effect of visibility did not significantly differ between the conditions memory and perception.
Figure 6. Mean accuracy for visibility of the cue.
Figure 6 illustrates a lower mean accuracy for invisible stimuli in both memory and
perception condition. For both the visible and the invisible the accuracy is higher in the perception condition. As noted before, there is no interaction effect for condition and visibility. These results are in line with our expectations.
2.2. Did they perform above chance?
To examine whether the participants performed above chance we run an independent t-test. Assumptions for this test are were met. In the perception condition the accuracy for both the invisible and visible stimuli was significantly different from chance, respectively t (1,14) = 3.15, p = .007 and t (1,14) = 24.9, p < .001. In the memory condition, the accuracy for visible stimuli was also significantly different from chance t (1,14) = 7.58, p < .001. Contradictory to the other results, the accuracy for invisible stimuli was not significant different from chance t (1,14) = .329, p = .747. This is not in line with our expectations and the results of the study of Soto et al. (2011) about unconscious WM.
Discussion
In the current study, the relation of WM and consciousness was reexamined by replicating the study of Soto (2011). First, to give a more in depth interpretation of our results, we looked into the awareness ratings and examined whether the average visibility rating was influenced by condition and presence of the stimulus. Furthermore, our main analysis was whether subjects performed above chance and whether the accuracy was influenced by condition and visibility.
First, the analysis of visibility is showing higher visibility reports for present than for absent cues. These results show that subjects could indicate whether the cue was present or not. Also, analysis of d-prime shows that subjects were able to discriminate between the presence and absence of a cue. In both conditions the d-prime is above 0. Then, there was an effect for presence of the cue: subjects gave higher visibility ratings when the cue was present than when the cue was absent. This difference depends on the type of test (memory or condition). For absent stimuli, there was no increase of visibility rating (from memory to
perception), but for present stimuli the visibility ratings were higher in the perception condition than the memory condition. This result is also confirmed by the d-prime analysis: the d-prime is higher for the perception condition than for the memory condition. This is indicating that the cue sensitivity is higher when there is no delay and no additional cognitive operational comparison.
The main analyses of visual WM performance are showing an effect for visibility of the cue: the performance was higher for visible stimuli than for invisible stimuli. This effect does not depend on the condition. In both the memory and perception condition the performance was lower for invisible stimuli and higher for visible stimuli. Which is logical because when people report they did not see anything, they had to guess and therefore the performance is lower. These results are in line with our expectations. In addition, there is an effect for condition: the performance is higher in perception condition than for the memory condition. So, the delay in the memory condition caused a lower performance than no delay in the perception task. This result is also in line with our expectations. Furthermore, in the perception condition the performance for both visible and invisible stimuli was above chance. In the memory condition, the performance for visible stimuli was also above chance. Contradictory to our expectations, the performance for invisible stimuli in the memory condition did not differ from chance.
An explanation for this unexpected result could be that we did not replicate Soto’s (2011) study properly because in the current study, subjects report less “1” (did not see anything) ratings (40% in the current study versus 55% in Soto et al. (2011)). This could mean there were less trials where subjects pressed “1”. But this is not the case because we used more trials than Soto et al. (2011) did. Another argument could be that the subjects of the current study were more liberal in giving higher perception ratings. This argument is implausible because the current study used the same awareness scale as in Soto et al. (2011) which means
the subjects in the current study should rate the same stimuli in the exact same way as the subjects in Soto et al. (2011). A third argument for not replicating Soto et al. (2011) is that the present study would not have enough power. But also this argument is invalid. Based on the effect for above-chance discrimination in the unaware trials of Soto et al. (2011) (Experiment 1, effect size d=1.52, with N=8, which is a very large effect according to Cohen), the current experiment had with N=16 more than 99% power to detect the effect.
The fact that the visibility is better in the perception condition than in the memory condition suggests that the visual information fades out over time. In previous research, they never came up with the possibility of forgetting the visibility. The results from the current study could mean that in the memory condition the accuracy is lower not because they could not see the memory cue, but because the information about the cue is quickly forgotten. This would support the idea that unconscious WM performance is more like a blind sight phenomenon, where non-conscious cues are only short-lived (Dutta, 2014).
A methodological limitation of the current study is the distinction between the memory and perception condition. The differences between these conditions is not only temporal but it is also a comparison between a simple versus a complex task. Whereas the complex task is the comparison between the memory cue and probe plus the awareness rating, the simple task only contains a perception test and awareness rating. To examine whether this difference between conditions influenced the result, there should be an extra condition taken into account. This condition should contain the same comparison as in the memory condition but only with a shorter delay in between. For further research on the relation of WM and consciousness a design with this additional condition could be useful.
Concluding, the current study could not replicate the findings of Soto (2011). We found no above chance performance for invisible stimuli in the memory condition. Furthermore, we found better performance in the perception task than in a memory task. When looking further
into the data we found that subjects were able to indicate whether a stimulus was present or not. Although further research is necessary for conclusions about the exact meaning of the differences in performance and visibility ratings between the memory and perception task, the present study has robust results for the non-replication of Soto et al. (2011) due to the high power of this study. These results are supporting the traditional view on the relationship of WM and consciousness rather than the novel views.
References
Baars, B.J. & Franklin, S. (2003). How conscious experience and working memory interact.
Trends in Cognitive Sciences, 7, 166-172.
Bona, S., Cattaneo, Z., Vecchi, T., Soto, D. & Silvanto, J. (2013). Metacognition of visual short- term memory: dissociation between objective and subjective components of VSTM.
Front Psychology, 4, 62.
Dehaene, S (2014). Consciousness and the brain. Deciphering How the Brain Codes Our
Thoughts. New York, NY: Pinguin Books.
Dehaene, S. & Changeux, J. (2011). Experimental and Theoretical Approaches to Conscious Processing. Neuron, 70, 200-227
Field, A. (2013). Discovering statistics using IBM SPSS statistics. London: Sage.
Jacobs, C. & Silvanto, J. (2015). How is working memory content consciously experienced? The ‘conscious copy’ model op WM introspection. Neuroscience and
Biobehavioral Review, 55, 510-519.
Overgaard, M., Rote, J. Mouridsen, K. & Ramsøy, T. Z. (2006). Is conscious perception gradual or dichotomous? A comparison of report methodologies during a visual task.
Conscious and Cognition, 15, 700-708.
Schmidt, T. (2015). Invisible Stimuli, Implicit Thresholds: Why Invisibility Judgements Cannot be Interpreted in Isolation. Advances in Cognitive Psychology, 11, 31-41. Soto, D., Mäntylä, T. & Silvanto, J. (2011). Working memory without consciousness. Current
Biology, 21, 912-913.
Soto, D. & Silvanto, J. (2014). Reappraising the relationship between working memory and conscious awareness. Trends in Cognitive Sciences, 18, 520-525.
