Interference from Unconscious Distracters in Visual Working Memory Rik Janssen
Programmagroep Brein en Cognitie, Afdeling Psychologie Universiteit van Amsterdam
Student number: 10080619
Supervisor: Timo Stein
Second assessor: Yair Pinto
ABSTRACT. Traditionally, consciousness and Working Memory (WM) are assumed to be closely related: that what is processed by WM is processed consciously. Soto and Silvanto (2014) propose WM to be separate from consciousness because, amongst other reasons, unconsciously presented distracting information can impair WM-performance. Here, this notion is further researched using stricter criteria of consciousness. First, an objective
threshold is established and compared to the subjective threshold of visibility. Interference of distracters is then assessed by comparing performance of Visual Short-Term Memory (a sub-module of WM). No interference was found for incongruent distracters compared to
congruent and no distracter conditions. Therefore the notion that unconscious stimuli
Working memory and consciousness
In cognitive psychology and neuroscience, consciousness and working memory (WM) are often perceived as two related concepts, which has been a tradition for more than 40 years (e.g. Atkinson & Shiffrin, 1971). Working memory is a concept referring to the (short-term) retention, updating and manipulation of data. WM is seen as one of the three key components of memory, working alongside short term memory and long term memory. A common model of WM is clearly described by Baddeley (2003). Working memory is proposed to exist of the central executive and three short term information storage systems: the visuospatial
sketchpad, the phonological loop and the episodic buffer to combine different information streams. Within this model, consciousness reflects the complete content of working memory and this content is instantly reportable.
Soto and Silvanto (2014) recently challenged this view. They argue that WM and consciousness are two separate cognitive processes. In their proposed theory, processes in WM are not taking place consciously, but consciousness only holds a sloppy copy of that what is going on in WM. Soto and Silvanto have three arguments supporting their new framework for the relation between consciousness and working memory.
The first point that Silvanto and Soto make is that WM can be interfered by unconscious stimuli. In their 2012 research, they used a simple visual memory task to look at the influence of distracters onto performance of visual short term memory (VSTM), seen as a sub-module of WM. Distracters are stimuli that contain either congruent or incongruent information, and are presented during a memory task in between memorizing and recalling a test stimuli. Typically, these distracters are presented for a very short time and immediately followed by a masking stimulus, so that they are hard to perceive or even invisible. Soto and Silvanto found a drop in VSTM performance for the trials where the distracter consisted of incongruent
visual information. Surprisingly, this effect was still present when looking at only those trials where participants responded that they did not see a distracter. Therefore the authors
concluded that in these cases the distracter was presented subliminally, but still interfering with WM. In a similar set-up, Bona, Cattaneo, Vecchi, Soto & Silvanto (2013) replicated these results: for trials where distracters contained incongruent information, VSTM performance dropped, even when distracters were rated invisible.
Secondly, Soto and Silvanto argue that contents of WM are not always easily and correctly reported on. For instance, Bona et al. found that introspection accuracy for WM content was reduced only when distracters, both congruent and incongruent, were
unconsciously presented. Apparently, our judgement of how well we saw something is altered by both types of invisible stimuli, whereas accuracy of WM is only influenced by incongruent distracters, pointing towards different underlying mechanisms. Additionally, this shows that our WM contents are not completely open to introspection and that our insight into it is not flawless, meaning that consciousness and WM are not perfectly aligned.
Thirdly, WM is not only influenced by, but also seems to be able to operate on
unconscious information. For instance, priming can lead to facilitating of naming of words, numeric computations and objects (Dehaene & Changeux, 2011). Furthermore, Soto, Mäntylä & Silvanto (2011) saw that a masked memory cue could still evoke above chance memory performance. This held even for two masked memory cues in different positions in a task that required binding of orientation and location information. Pan, Ling, Zhao & Soto (2013) used continuous flash suppression to unconsciously administer visual information. If the
information matched the WM contents then the information gained access into awareness, even when the visual information that matched entered WM unconsciously. Even reading and arithmetic computations are being associated with unconscious processing by Sklar et al. (2012).
These results make a clear case for consciousness and WM being separate mechanisms. However, these conclusions might be somewhat too far-fetched. All of the above mentioned authors used subjective measures to assess whether a distracter in a specific trial was either conscious or unconscious. Specifically they used the Perception and Awareness Scale (PAS-scale), or variations on it. The PAS-scale is a four category scale on which participants rate the visibility of a specific stimulus on a trial to trial basis. The scale was developed by
Overgaard, Rote, Mouridsen & Ramsøy (2006) and the four options consist of (1) Did not see anything, (2) Brief glimpse, (3) Almost clear experience and (4) Clear experience. It is usual practice to count a rating of 1 on this scale as a trial where the stimulus was presented subliminally.
According to Schmidt (2015), subjective measures of consciousness like the PAS, are not valid or reliable. He argues that asking participants for judgements about the presence of a stimulus is surely informative, but it is however irresponsible to interpret these trials in
isolation and to conclude that these trials were invisible or even unconscious. One of the main problems is that this practice completely ignores response bias effects. Specific instructions, formulations of the question and amount of response options all influence the proportion of no consciousness at all responses. Besides, personal confidence about memory and perception can greatly differ between people.
Another problem that Schmidt proposes is that subjective measures are incompatible with important theories about human perception such as signal detection theory. This theory states that stimuli create internal representations and that one decides whether this
representation is more likely to be caused by random noise or by an actual signal. In this process, there will be correct responses (a hit, or a correct rejection) or incorrect responses (a miss or a false alarm).
Using the PAS-scale, one assumes that the different rating categories refer to different states of consciousness and with that assumes that within one category, all types of
consciousness are similar and refer to the same perceptual state. The most important argument in articles that argue WM and consciousness as separate processes (such as in Silvanto and Soto, 2012) is the Gap argument. This argument states that there is a gap between that there is still information recalled correctly, while people claim they did not see anything. Then it is concluded that due to this Gap, unconscious processing of information must have taken place.
However, in most cases this is m this argument could be the full consequence of the response bias: because dividing mental states between conscious and unconscious into a couple of categories requires rounding off on your experience. In this case, it could be that they respond did not see anything although in reality they saw something, but just not enough to go up one visibility category. The measure of information retainment is hopefully more sensitive than the subjectively categorized responses to a phenomenon (awareness) that might be continuous instead of an all-or-nothing dichotomy. For that reason, the gap argument is not a valid argument in combination with subjective visibility ratings.
Motivation and personal investment are two more factors that influence subjective
perceptual judgements, as Nieuwenhuis and De Kleijn argue in their 2011 research. They used a post-decision wagering task as a measure of consciousness to assess whether the transition between conscious and unconscious is either a gradual transition, or a continuous one. The main task was an attentional blink experiment, with an subjective measure of consciousness. Looking at the subjective measure results, there seemed to be a discontinuous transition. However, when people had to wager on the presence or absence of the stimulus presented during attentional blink, participants were highly accurate and hardly ever wagered
incorrectly or zero eurocents. Also, in by far the most instances, they wagered the maximum amount, indicating that they were pretty sure to have or have not seen the stimulus. In the
light of these arguments, arguably it is relevant to use other measures in addition to subjective measures of consciousness.
Objective versus Subjective invisibility
Where can we draw the line of invisibility if we cannot trust the subjective measure of self-reported invisibility ratings? We will need an objective measure to avoid these earlier mentioned problematic assumptions. The distinction between an objective and subjective threshold of visibility is often made (Merikle, Smilek & Eastwood, 2001; Snodgrass, Bernat & Shevrin, 2004) and clearly explained by Snodgrass (2004). Something is subjectively invisible if a person states that they did not see anything. The objective criterion for
invisibility is different. If something is objectively invisible to someone, a person will not be able to discriminate any information about the presented stimulus, in a forced-choice
paradigm. It is therefore valuable to assess the subliminal influences of a distracter while a participant is not able to discriminate the orientation of the stimulus. This orientation is the dimension of the stimulus that was found to be interfering WM in previous research. If we use an objective criterion of invisibility we will be less likely to rely on the earlier mentioned problematic assumptions. For instance, if someone performs at chance on a left-right decision paradigm, then we can be sure that no information about the orientation of the stimulus
entered consciousness. It would be interesting to do so, because we can then establish whether (1) a subjective threshold of consciousness would be at the same level of objective threshold (showing that self-report techniques do actually indicate that something was not seen and are therefore valid measure) and (2) if the interference from unconscious stimuli still hold if the objective criterion is used.
In this research, three separate experiments address these two questions. The first experiment will focus on the difference between objective and subjective thresholds of
invisibility, at different distracter presentation times and different mask settings. Using the results in experiment 1, thresholds for experiments 2 and 3 will be established. In the latter experiments, effects of distracter interference will be assessed at these different thresholds.
Methods
Materials. All three experiments took place with participants in separate cubicles. All cubicles contained the same type of computers (Dell Optiplex 9010, intel Core i5 processor) and the same Software was Matlab r20 (Mathworks). On-screen presentation times were checked with a photodiode measurement. Participants were rewarded with either course credits or money (15 euro for 1.5 hours).
Experiment 1
Participants. Eleven (four men and seven women) participants completed the experiment (mean age: 21.6 years).
Stimuli. The experiment had a simple set-up: different mask settings were used and compared, to mask a target stimulus: a tilted grating (Gabor patch). The tilt of the patch was either 30, 40 or 50 degrees to the left or to the right. To make the stimulus harder to perceive, the patch was in low contrast (20%). The following mask was either totally black, or they were black and white concentered circles at either 10, 20 or 50% contrast. See Table 1 for the different masks.
Procedure. Participants were asked to judge the orientation of the target: they
responded whether it was tilted to the left or right compared to completely vertical. Using this set-up, five mask settings were tested with two target presentation times accounting for 10 conditions and 720 trials in total. Half of these trials were absent trials: no target was
presented at all. Participants were counterbalanced between starting the task at 60 Hz screen refresh rate (masks 1 through 5) or at 100 Hz (masks 6 through 10). Within these frame rate settings condition trials were randomized. See Table 1 for overview of the different
conditions.
Condition Target presentation time Mask type Mask
1 16.667 ms Blank (16.7 ms) > Black 2 16.667 ms Black 3 16.667 ms Circular 10% contrast 4 16.667 ms Circular 20% contrast 5 16.667 ms Circular 50% contrast 6 10 ms Blank (10 ms) > Black 7 10 ms Black 8 10 ms Circular 10% contrast 9 10 ms Circular 20% contrast 10 10 ms Circular 50% contrast
Table 1. Different mask settings for the different conditions.
Trials started with presentation of a fixation cross (black, centre of screen). Then the target was presented for 10 (one screen refresh at 100 Hz) or 17 ms (one screen refresh at 60 Hz), depending on condition. This was immediately followed by the mask, for 200 ms. Then there was a 100 ms offset before a left/right response was required, but not within a specific time frame. Participants answered using the left and right arrow keys on the keypad. The last screen of the trial was the PAS-scale, which was answered using keys 1 to 4 of the keypad, also untimed. The four options consisted of (1) Did not see anything, (2) Brief glimpse, (3) Almost clear experience and (4) Clear experience. See Figure 1 for a visual overview of the trial order.
Figure 1. standard trial order. In this example, the correct response would be Right ().
Results experiment 1
Subjective Visibility. Subjective visibility ratings according to PAS-scale responses. These PAS responses make sense according to mask type: Shorter distracter presentation times led to lower PAS responses and higher complexity of the mask led to lower PAS responses. Also a pause between distracter and mask showed highest PAS scores (conditions 1 and 6). To see the proportion of pressed PAS responses, see Figure 2. For the trials where the target was absent, the pressed responses were all similar, so we found that no mask evoked a false alarm effect more than the other.
D-prime (or discriminability index) is a statistic based on signal detection theory and refers to the measure of the strength of a signal (Heeger, D., 1998). Signal detection theory states that when one has to decide whether there is a stimulus present they compare two distributions: possible internal response (including noise) with stimulus absent or possible response when there is in fact a stimulus present. The observer then decides whether their internal response is more likely to result from the absent or the present distribution.
Calculating d’ is simple: it is the separation between the means of the distribution divided by their spread. Important note here is that d’ is not dependent on an internal criterion (that is variable from person to person) but incorporates information of both correct responses (hits and correct rejections) and incorrect responses (false alarms and misses) to estimate
discriminability of the stimulus. Therefore, the d-prime statistic is subjective but unbiased, because it is independent of someone’s internal criteria. In our case we calculated a PAS-scale response of 1 as an unseen judgement, and a response of 2, 3 or 4 as a seen judgement. If D-prime is higher, this means that presence of a stimulus is easier to discriminate (its mean of its distribution lies further away from the stimulus absent distribution). See Table 2 for D-prime statistics for each mask.
Figure 2. Proportion of PAS ratings (y-axis) for every masking condition (x-axis), for the target present trials.
Accuracy. Looking at proportion correct left-right discrimination responses, we see that for all presentation times at 17 ms proportion correct was well above the .5 threshold. One-way ANOVA tests were conducted for each mask, confirming that for all masks except mask 10, accuracy was significantly above .5. For statistical results see Table 2.
For the subjectively invisible trials (PAS-response = 1: Did not see anything), we want to see whether accuracy is still above .5, which would indicate that even for subjectively invisible trials, information about the masked stimulus is still processed. This means that the objective threshold (= no information is processed at all) is different from the subjective threshold. See the second half of Table 2 two for these results. Accuracy for PAS = 1 is not
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 1 2 3 4 5 6 7 8 9 10 Per cen ta ge o f r es po ns es Condition
PAS 1: Did not see anything
PAS 2: Brief glimpse
PAS 3: Almost clear experience PAS 4: Clear experience
displayed, because for conditions 1,2,3 and 6, at least two participants never responded 1, resulting in too few trials in these conditions to reliably analyze. In the results we see that even for the trials that were subjectively invisible, information about the stimulus was still retained; significant positive deviations from .5 occurred in conditions 4, 5 and 8.
Condition Accuracy SD t Accuracy
(PAS=1) SD (PAS=1) t (PAS=1) D-prime SD t
1. Blank > Black 16ms 0.95*** 0.13 11.298 - - - 3.42*** 3.42 25.380 2. Black 16ms 0.93*** 0.16 8.756 - - - 3.03*** 3.03 10.170 3. Circ10perc 16ms 0.94*** 0.16 9.343 - - - 3.05*** 3.05 17.252 4. Circ20perc 16ms 0.89*** 0.15 8.801 0.73* 0.32 2.324 1.99*** 1.99 8.577 5. Circ50perc 16ms 0.73** 0.17 4.393 0.61* 0.13 2.724 1.02*** 1.02 4.521 6. Blank > Black 10ms 0.87*** 0.19 6.434 - - - 2.31*** 2.31 7.033 7. Black 10ms 0.61* 0.15 2.363 0.54 0.15 0.825 0.59** 0.59 3.554 8. Circ10perc 10ms 0.78*** 0.18 5.114 0.57* 0.13 1.847 1.46*** 1.46 4.920 9. Circ20perc 10ms 0.62** 0.13 3.001 0.54 0.11 1.285 0.57** 0.57 2.809 10. Circ50perc 10ms 0.53 0.09 1.184 0.51 0.10 0.500 0.17 0.17 1.243
Table 2. Accuracy and subjective visibility for each masking condition1.
We have found a really nice trend from high accuracy for weak masks to low (at chance performance) for the strongest mask, so probably we are searching in within the right range of masking settings. For experiments 2 and 3, a threshold for objective invisibility is set, using results from Experiment 1. For this see that only for mask type 10 (the strongest mask, where the stimulus is presented only for 10 ms and immediately followed by a high, 50% contrast circular mask) accuracy was at chance level, thus we can conclude that this is a proper way of masking this type of stimulus with the objective threshold of visibility in mind.
Experiment 2
Participants. In total 4 women and 7 men participated with a mean age of 21.9 years, three of whom had also participated in experiment 1. Of the initial 12 participants, one person was excluded from analysis due to misinterpreting instructions. The interference effect size
found in Silvanto & Soto (2012) was very large, Cohen’s d of 1.43. Based on this, the
minimum number of participants is 9 for .95 power. We tested 11 people, resulting in a power of .99 to find an effect of the size that was reported.
Stimuli. There were 4 types of stimuli. First there was a memory cue stimulus, this was a tilted Gabor patch, tilted 10, 40 or 70 degrees to the left or to the right. Distracter stimuli were the light contrast (20%) Gabor patch presented for only 10 ms. These were either congruent with the memory cue, or incongruent (tilted 40 degrees left or right from cue). Two masks were used: either the black mask or the circular 50% contrast mask, corresponding to conditions 7 (accuracy was .61) and 10 (accuracy was .53) in our first experiment. The last stimulus was the memory probe: this was a high contrast Gabor patch but tilted slightly (10 degrees) compared to the memory cue.
Procedure. This experiment consisted of two parts: a memory task and a short perception task, similar to experiment one but with only the two mentioned masks.
Participants were counterbalanced across starting the experiment with the black mask or the circular mask. In the memory task, participants remembered the orientation of a Cue: a tilted Gabor patch for about 3 seconds. During this retention period, a distracter stimulus was presented with a mask. The distracter was either absent (in 1/3 of the trials), or its orientation was congruent (1/3) or incongruent (1/3) with the orientation of the remembered cue. After the retention period, participants responded to a probe stimulus and indicated whether the cue was tilted to the right or left compared to the probe. Afterwards, they answered whether they had seen the distracter stimulus in that trial using the PAS rating system. Total number of trials was 288. Only 33% of trials did not contain a distracter, but all trials contained a mask. For a visual scheme of this order, see Figure 3. After the memory part of the experiment we wanted to check that the perception experiments results were also valid for this new group of participants. Therefore, following one memory task, participants completed one small
perception task as well (similar to experiment 1), before continuing to the second memory task.
Figure 3.Standard stimulus order within the memory task. In this example, the correct response would be Left ().
Results Experiment 2
Working memory accuracy. We expected to find a drop in memory performance accuracy for trials where there was an incongruent distracter presented. Generally, our memory task was not really hard: baseline WM accuracy for the distracter absent trials was 0.80 (SD = 0.08), see Figure 4. In contradiction with earlier research, no lower memory performance scores were appeared in the incongruent distracter trials compared to the distracter absent trials or the congruent distracter trials. This means that we found no
interference for the trials were there was an incongruent distracter (see Figure 4): accuracy of working memory was stable across conditions for both the black mask, F(1, 10) = 0.07, p = .97, and the circular 50% contrasts mask, F(1, 10) = 0.13, p =.73. Also when looking at only the subjectively invisible trials (PAS = 1), there was no significant effect of the distracter onto working memory performance for either the black mask: F(1, 10) = 0.96, p = .35, or the circular mask: F(1, 10) = 0.33, p =.58 (see Figure 5). Even the trials where the distracter was rated visible no interference occurred for the black mask; F(1, 4) = 0.21, p = .67 or the circular mask; F(1, 4) = 0.69, p = .53. See figures 4, 5 and 6 for WM accuracy across conditions.
Figure 4 (upper left). Mean accuracy for the memory task.
Figure 5 (upper right). Mean accuracy for subjectively invisible trials (PAS=1).
Figure 6 (lower left). Mean WM accuracy for visible trials.
Visibility of distracter. Visibility of the distracter was extremely low for both masks in the perception part of the experiment. Participants rated visibility of distracter as PAS=1 in almost all trials. Also, the accuracy was at chance level, so both masks were now at the objectively invisible threshold. All participants verbally mentioned to have had difficulty seeing any of the distracters and keeping themselves motivated. Therefore there are less participants in the analysis for the visible trials, there were only 5 people that had >10 trials rated as visible. See Table 3 for all visibility results for the different conditions of the perception experiment.
Condition Accuracy SD t Accuracy
(PAS=1) SD (PAS=1) t (PAS=1) D-prime SD t
1. Black Mask 0.51 0.04 0.745 0.50 0.04 0.226 -0.01 0.29 -0.072
2. Circular Mask 0.50 0.04 0.338 0.50 0.04 0.328 -0.18 0.24 -2.513
Table 3. Perception results for Experiment 2. 0,5 0,6 0,7 0,8 0,9 1
Black Mask Circular Mask
W or kin g M em or y A ccu ra cy
Congruent Incongruent Absent
0,5 0,6 0,7 0,8 0,9 1
Black Mask Circular Mask
W or kin g M em or y A ccu ra cy
Congruent Incongruent Absent
0 0,2 0,4 0,6 0,8 1
Black Mask Circular Mask
Besides the main analysis, we want to compare visibility ratings for the perception part of the experiment to the visibility ratings inside the memory task of the experiment. This could tell us whether WM-engagement on a more complex task influences the ability to see a shortly presented stimulus. This is an interesting notion that is the opposite of our main investigation: instead of working memory being interrupted by a visual stimulus, we want to make sure it is not the working memory task that negatively influences the visual perception process. See Table 4 for the results.
Mask Type Mean PAS
Memory Task Mean PAS Perception Task Mean difference t 1. Black Mask 1.30 1.07 0.23 ns 1.445 2. Circular Mask 1.22 1.03 0.19 ns 1.111
Table 4. Subjective visibility differences.
Experiment 3
Participants. For a more sensitive analysis, the number of participants was increased from 11 in previous experiments, to 19 in this experiment. These were seven males and twelve females with a mean age of 24,42 years. Of these, 3 people also participated in experiment 2 and of them, one also participated in experiment 1.
Stimuli. This experiment was similar to Experiment 2, but there was a third, easier mask setting condition included. Thus, there was the black mask, the circular mask and a 10 ms pause and then black mask condition. We did this to keep participants motivated during the experiment. In our last experiment we saw that distracters were too hard to perceive and participants mentioned giving up during the experiment. The blank screen and black mask setting led to an orientation discrimination accuracy of .87 in Experiment 1, highly significant above chance level.
Secondly, a hopefully more sensitive measurement of accuracy was used: instead of just a left right response, participants were asked to indicate exact orientation of the memory
cue, using the mouse to adjust a white bar in the right position. This adjustment bar appeared in a random orientation each trial.
Procedure. The experiment consisted of two parts. Firstly, participants completed the memory task. See figure 7 for the order within one memory trial. A memory cue is presented, that people were instructed to remember for about 2 seconds. In those two seconds a masked distracter stimulus was presented. Distracter presentation time was always 10 ms, and in 33% of trials no distracter was presented at all. After a small delay, the adjustment bar prompted participants to respond to the orientation of the remembered stimulus.
After the memory task, participants again completed a perception task, similar to experiment 1 and at the end of experiment 2, but now for the three mask types: Blank then black, black mask and circular 50% contrast mask.
Figure 7. standard trial order. In this example, the correct response would be to tilt the adjustment bar somewhat more to the right.
Results Experiment 3
Working Memory Accuracy. Note that in this case, accuracy is not measured in left-right response (which has a .5 at chance level), but in degrees off-target. So a higher level of memory accuracy will show a lower number of degrees. Working memory accuracy did not differ in the distracter type (congruent, incongruent, absent) conditions, F (2, 36) = 0.141, p = .86. See Figure 8 for working memory performance across conditions. Even when looking at
only the subjectively invisible trials, no interference effects of the distracter type were found, F (2, 36) = 0.853, p = .44 (see Figure 9).
For the trials that were rated visible, only 9 participants responded more than 10 times that they had seen the distracter stimulus and these were included in analysis. Even for these visible trials, no interference of Working Memory was found for the incongruent distracter condition compared to any other condition; F (2, 16) = 1.45, p = .26.
Figure 8. Working Memory accuracy for the memory task.
Figure 9. Working Memory accuracy for the subjectively invisible trials.
0 5 10 15 20 25
Blank+Black Mask Black Mask Circular 50% Mask
Deg rees o ff-ta rg et
Congruent Incongruent Absent
0 5 10 15 20 25
Blank+Black Mask Black Mask Circular 50% Mask
Deg rees o ff-ta rg et
Figure 10. Working Memory accuracy for the visible trials.
Perception of distracter. In the perception part of the experiment, there were significant positive deviations from chance level with the blank then black mask, and the black mask conditions (See table 5). For the subjectively invisible trials, there were no deviations from chance level.
Condition Accuracy SD t Accuracy
(PAS=1) SD (PAS=1) t (PAS=1) D-prime SD t
1. Blank > Black Mask 0.78*** 0.21 5.721 0.54 0.24 0.757 0.97*** 1.14 3.725
2. Black Mask 0.59** 0.12 3.358 0.51 0.09 0.626 0.14 0.41 1.479
3. Circular Mask 0.48 0.06 -1.212 0.48 0.06 1.305 0.33 0.33 -1.197
Table 5. Perception accuracy results for Experiment 3.
Just like in experiment 2, we want to make sure that the visibility ratings were not different within the memory experiment, than in the perception only part of the experiment. Table 6 shows how visibility ratings in the memory part and the perception part of the experiment compared. For the 50% contrast circular mask, the PAS ratings in the memory part of the experiment were higher than in the perception part of the experiment. This indicates that the memory task does not negatively influence the perception process. However, it cannot be concluded that in the perception task, perceiving the target was harder, because the perception task was last in the experiment so other factors like tiredness cannot be eliminated to have influenced this result.
0 5 10 15 20 25
Blank>Black Black Circular
Deg ees o ff-ta rg et
Mask Type Mean PAS
Memory Task Mean PAS Perception Task Mean difference t
1. Blank > Black Mask 1.63 1.84 -0.21 ns -1.496
2. Black Mask 1.38 1.33 0.05 ns 0.511
3. Circular Mask 1.19 1.04 0.15* 2.631
Table 6. Subjective visibility ratings in both parts of the experiment.
Analyses over all experiments
Interestingly, subjective and objective visibility of shortly presented stimuli varied across experiments for the same masks. The perception results from experiment 3 are more in line with the results in experiment 1 than those of experiment 2 are. This is true for especially a stimulus presentation time of 10 ms, followed by a black mask. If these differences are significant, it would be interesting to see what factors could have influenced the results in this way. Both its objective and subjective visibility varied across experiments, while its physical appearance did not. Actual on-screen time at 100 Hz and at 60 Hz were verified by a
photodiode measurement and turned out to be highly accurate. D-prime was therefore compared across experiments for the same mask. The d-prime data are most suitable to compare between experiments, because it is not affected by the participants internal thresholds. It is different from the accuracy results because this statistic is not based on the left-right discrimination capacities of the participant, but purely on the judgements of there being a stimulus present at all. The black mask and the circular mask after 10 ms stimulus was compared across all 3 experiments, and the black mask after black screen was compared between experiments 1 and 3.
Mask Type D-prime
Exp 1 SD Exp 1 D-prime Exp 2 SD Exp 2 D-prime Exp 3 SD Exp 3 Test df Significance
1. Blank > Black Mask 2.31 1.09 X X 0.97 1.14 T-test 28 .004
2. Black Mask 0.59 0.55 0.01 0.29 0.14 0.41 ANOVA 2, 38 .005
3. Circular Mask 0.17 0.46 0.18 0.24 0.09 0.33 ANOVA 2, 38 .056
Looking at the perception data for the blank > black mask (used in experiments 1 and 3), we see that the D-prime is higher in Experiment 1, than for the same mask in Experiment 3, t(28) = 3.16, p = .004. This shows people were more often correct to judge whether there was a stimulus presented before the mask or not in experiment 1 than in experiment 3. Also D-prime for the black mask differed significantly across experiments (see Table 7). It was easiest to perceive the stimulus with this mask in Experiment 1, easier than in experiments 2 and 3. For the circular masks, these differences were non-significant. This was actually expected seeing that it’s D-prime was very low and accuracy at chance in all experiments, showing that it is indeed close to completely invisible in all cases.
Discussion
In this paper, two main questions were addressed. The first question concerned the border between visible and invisible stimuli. Where in previous articles, a subjective threshold of visibility was used, this article was focused at finding an objective threshold of visibility. Experiment 1 was designed to find this threshold and also, to make the objective threshold directly comparable to the subjective threshold. What we found was indeed an objective threshold of visibility, that in this case was a distractor presentation time of 10 ms together with an intense circular mask at 50% contrast. This mask setting led participants to perform at chance for the perception task: indicating that no visual information of the stimulus was retained. When all trials where participants responded that they did not see anything where isolated (adhering to the subjective threshold of visibility), there was still an above chance performance on the perception task, showing that for these subjectively ‘invisible’ stimuli
information was still retained and acted upon. This is a very important finding: that the subjective threshold was not as strict as the objective threshold.
The second question was whether invisible stimuli could interfere with working memory, whether it was at an subjectively or objectively invisible level. Experiments 2 and 3 were aimed at finding this interference. However, in both experiments we did not find any interference of invisible distracters on performance on a working memory task, and neither for visible distracters. These findings were surprising; and not really in line with the expected results based on the earlier findings in articles by Soto and Silvanto (2012, 2014) or results in Bona and colleagues (2013), who did find interference for the subjectively invisible trials. Note that they did use a quite loose masking setting in these experiments: for instance Bona used a 17 ms distracter in his experiments, together with a black mask, a setting that turned out to be very visible in our first experiment. For the Silvanto and Soto 2012 research, a 13 ms black mask setting was used. This is an interesting setting: in our first experiment this setting was pretty visible (even though we used a stricter distracter setting of 10 ms instead of 13), but in the latter experiments this setting turned out to render the stimulus very invisible, both subjectively and objectively. One question that arises during interpretations of the results was as to how accuracy and visibility of the same masks could yield so very different
outcomes in the different perception experiments. Is it possible that in our first sample of participants we had an extremely accurate and well-observing sample? Were they somehow significantly younger or fitter than the other two experiments? In terms of age, there were no significant differences. Otherwise, it is hard to establish that the first sample was fitter. Experiment 1 took place on the same times of the day as the other two experiments, and other circumstances (room, lighting, rewards) were constant across experiments. Only differences between experiments were total duration, procedure and overall masking intensity.
Duration. The first experiment contained only short trials, and these 720 trials in total resulted in an on-screen experiment run time of around 45 minutes. For the other two
experiments, the time participants spent looking at the screen was around 75 minutes. This factor might have influenced the overall results: effects of tiredness or boredom might have had a larger negative effect on accuracy and visibility in the latter two experiments.
Procedure. The first experiment only contained perception trials, there were no memory tasks involved. As for the other two experiments, the perception trials followed the memory trials. This could have led to, again, tiredness, boredom and overall decreased motivation. At the beginning of the experiment we thought that it would actually be easier overtime to observe the masked distractor, that some kind of learning effect would arise. But in our study, that was not the case. When we compared results of the first part of the
experiment to the second part: no worse or better results were found either way. Overall Masking Intensity. Looking at different masking settings, in the first experiment there were a couple of very ‘easy’ settings to mask the distractor. For instance, masks 1, 2, and 3 accuracy was over 90% and for masks 4 and 6 it was still over 85%.
Conceivably, it is easier to spot the distractor if it is once in a while very clear, to remind you of what you are looking for. Also these ‘hits’ could keep people motivated and self-assured. Cheesman and Merikle found similar results in their 1984 research: if presentation times were gradually lowered, participants responded better to the shorter presentation times as well. In our second experiment, most participants doubted there being a distractor present at all in the experiment, and mentioned getting really sleepy during the experiment. Generally having more success spotting the stimulus seems to increase performance, just like in our first and third experiment.
As mentioned in the introduction, there are factors that influence the subjective visibility ratings such as personal investment (e.g. wagering money), but seeing that in this
research we found flexibility of the objective threshold as well across experiments. In further research, it would be valuable to assess what other factors influence the objective as well.
Consciousness and Working memory. The new framework that is proposed by Soto and Silvanto is that consciousness and working memory are separate: consciousness is only a copy of that what is in working memory but does not have full access to it. One of their main arguments is that unconscious stimuli can impair WM performance. In this research there was no interference found from distracters onto Visual Short-Term Memory, at least not with these strict masking settings to achieve invisibility. Therefore, the results are not in line with the statement that Consciousness and WM are separate processes.
Consciousness all-or-nothing? In most cases we assume something to be either conscious, or unconscious. But is this assumption actually valid to make? What if
consciousness is more a gradual transition between somewhat more conscious and somewhat less conscious states? The Soto and Silvanto theory assumes that consciousness is an all-or-nothing dichotomy, seeing that they categorize trials only in conscious and unconscious trials. But how can we be sure that this is in fact how consciousness works? Can’t we be somewhat conscious, without being entirely sure? If consciousness is a gradual transition, categorization into all-or-nothing can be extremely misleading, because it requires rounding off on an personal experience.
Objective and Subjective visibility. In previously mentioned articles (Silvanto & Soto, 2012; Bona et al., 2013) an invisible distracter was always a subjectively invisible distracter. Our research was unique in the sense to have worked with an objective threshold of
invisibility. However, it came as a surprise to see that even the objective threshold of visibility was subject to change, according to how it’s presented amongst other masks.
In future research it would be interesting to test whether this objective threshold is for sure influenced by overall masking intensity of all conditions, in that case using the same set of participants across different experiments, so that it is surely not the participant sample that caused the results found here. Additionally it would be interesting to assess if the objective threshold of visibility is also influenced by factors that influence the subjective visibility such as personal investment and motivation (Nieuwenhuis and De Kleijn, 2011).
Literature
Atkinson, R. C., & Shiffrin, R. M. (1971). The control processes of short-term memory. Stanford: Stanford University.
Baddeley, A. (2003). Working memory: looking back and looking forward. Nature Reviews. Neuroscience, 4(10), 829.
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. Frontiers in psychology, 4.
Cheesman, J., & Merikle, P. M. (1984). Priming with and without awareness. Attention, Perception, & Psychophysics, 36(4), 387-395.
Dehaene, S., & Changeux, J. P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70(2), 200-227.
Merikle, P. M., Smilek, D., & Eastwood, J. D. (2001). Perception without awareness: Perspectives from cognitive psychology. Cognition, 79(1), 115-134.
Nieuwenhuis, S., & de Kleijn, R. (2011). Consciousness of targets during the attentional blink: a gradual or all-or-none dimension?. Attention, Perception, &
Psychophysics, 73(2), 364-373.
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. Consciousness and cognition, 15(4), 700-708.
Pan, Y., Lin, B., Zhao, Y., & Soto, D. (2014). Working memory biasing of visual perception without awareness. Attention, Perception, & Psychophysics, 76(7), 2051-2062. Schmidt, T. (2015). Invisible stimuli, implicit thresholds: Why invisibility judgments cannot
Silvanto, J., & Soto, D. (2012). Causal evidence for subliminal percept-to-memory interference in early visual cortex. Neuroimage, 59(1), 840-845.
Sklar, A. Y., Levy, N., Goldstein, A., Mandel, R., Maril, A., & Hassin, R. R. (2012). Reading and doing arithmetic nonconsciously. Proceedings of the National Academy of
Sciences, 109(48), 19614-19619.
Snodgrass, M. (2004). The dissociation paradigm and its discontents: How can unconscious perception or memory be inferred?. Consciousness and cognition, 13(1), 107-116. Snodgrass, M., Bernat, E., & Shevrin, H. (2004). Unconscious perception: A model-based
approach to method and evidence. Attention, Perception, & Psychophysics, 66(5), 846-867.
Soto, D., Mäntylä, T., & Silvanto, J. (2011). Working memory without consciousness. Current Biology, 21(22), R912-R913.
Soto, D., & Silvanto, J. (2014). Reappraising the relationship between working memory and conscious awareness. Trends in Cognitive Sciences, 18(10), 520-525.
Endnotes
1
The number of asterisks in the tables represent the statistical significance of the tests.
P value Wording Summary
< 0.001 Very significant ***
0.001 to 0.01 Very significant **
0.01 to 0.05 Significant *
