Storing fine detailed information in visual working memory—Evidence from event-related potentials

Abstract

Visual working memory (VWM) maintains and manipulates a limited set of visual objects being actively used in visual processing. To explore whether and how the fine detailed information is stored in VWM, four experiments have been conducted while recording the contralateral delay activity (CDA), an event-related potential difference wave that reflects the information maintenance in VWM. The type of the remembered information was manipulated by adopting simple objects and complex objects as materials. We found the amplitude of CDA was modulated by object complexity: as the set size of memory array rose from 2 to 4, the amplitude of CDA stopped increasing for maintaining complex objects with detailed information, while continued increasing for storing highly discriminable simple objects. These results suggest that VWM can store the fine detailed information; however it can not store all the fine detailed information from 4 complex objects. It implies that the capacity of VWM is not only characterized by a fixed number of objects, there is at least one stage influenced by the detailed information contained in the objects. These results are further discussed within a two-stage storing model of VWM: different types of perceptual information (highly discriminable features and fine detailed features) are maintained in VWM via two distinctive mechanisms.

Introduction

Visual working memory (VWM) is a critical component of visual information processing (Baddeley, 1992). Though its capacity is highly limited, it allows us to integrate information presented in different eye fixations to get a coherent perception of the visual scene, as well as compare current information with information already stored in memory (Hollingworth, Richard, & Luck, 2008; Irwin & Andrews, 1996). Due to its significance, much research has been elicited to explore the VWM capacity limit over the past decade (see Jiang, Makovski, & Shim, 2009; Jonides et al., 2008, for a review). 

Currently, two contrasting views on VWM capacity have been proposed. The first considers that the capacity of VWM is limited by the number of visual objects, and approximately 3–4 object representations can be accurately maintained, independent of the number of features within an object and the complexity of objects. For example, Luck and Vogel (1997) found that the number of objects VWM can hold is equivalent between objects defined by a single feature (e.g., color, orientation) and multi-feature objects (e.g., tilted bars with color and orientation). The second argues that VWM capacity is a flexible while limited resource (Alvarez & Cavanagh, 2004, 2008; Bays & Husain, 2008), in which the number of objects that can be stored is reduced as object complexity or information load increases (e.g., Alvarez & Cavanagh, 2004, 2008; Eng, Chen, & Jiang, 2005). In one of their studies, Alvarez and Cavanagh (2004) used several different categories of complex shapes as research materials, which are difficult to search among distracters of the same category. They suggested that the search rate could be an index of visual complexity or information load per object, since the more visual information that must be analyzed per object, the slower the processing rate. They discovered that capacity estimates decrease monotonically as complexity increases. However, in a following research, Awh, Barton, and Vogel (2007) posited that in Alvarez's research, the poor change detection performance for more complex items was due to the high memory-test similarity which made the comparison between the representations stored in memory and the items presented in the test array much more difficult. After reducing the memory-test similarity by changing the stimuli into a different category under the test change condition, they found no statistical difference existing for capacity estimates of VWM between simple and complex objects. Thus, they concluded that VWM represents a fixed number of objects regardless of complexity. 

So far, Awh and his colleagues (2007, 2008) have provided convincing evidence supporting the fixed-number-storage view, even for complex objects. Notably, Awh et al. also suggested that those complex object representations have limited resolution in VWM. Jiang, Shim, and Makovski (2008) further claimed that these representations may only contain high discriminable features. Apparently, the fine detailed information is not maintained in the limited-resolution representations (see also Most et al., 2001; O'Regan, 1992; Simons & Chabris, 1999; but Hollingworth, 2004; Liu & Jiang, 2005). So whether and how the fine detailed information is stored in VWM remains unclear. 

As known in previous studies, a change detection task can be decomposed into encoding, maintenance, and comparison phases (Pessoa, Gutierrez, Bandettini, & Ungerleider, 2002; Todd & Marois, 2004). In Awh et al.'s behavioral research, they also found that the behavioral performance dropped when the change signal was subtle in the test array, which is also repeated by other studies (Alvarez & Cavanagh, 2004, 2008; Eng et al., 2005). Awh et al. attributed the complexity effect to the comparison phase. However, there are actually two possible explanations to this result. On one hand, VWM may only consist of a fixed number of objects with low resolution. The drop in performance is purely due to the comparison errors, since the low-resolution object representations are not sufficient for comparisons involving fine detailed information. On the other hand, it is equally plausible that beyond the fixed number of coarse representations, VWM also contains a limited amount of fine detailed information, which may also contribute to the comparison process. However, since only a very limited amount of fine detailed information can be successfully stored, the performance can not be as good as when the comparison only requires low-resolution information. 

Since ERPs can provide a measure of stimulus-related processing with a high temporal resolution, we attempted to investigate the storage of fine detailed information by recording event-related potentials (ERPs) during the maintenance phase of object representations in VWM. Specifically, Vogel and Machizawa (2004), Vogel, McCollough, and Machizawa (2005), and Vogel, Woodman, and Luck (2006) found that there is a waveform of the event-related potential with a sustained negative voltage over the hemisphere that is contralateral to the memorized hemifield and persisted throughout the memory retention interval in a VWM task. Importantly, the amplitude of this contralateral delay activity (CDA) is higher for 4 simple objects than that for 2 simple objects. They explained that the amplitude of CDA reflected the maintenance of object representations in VWM. In this case, we chose CDA as an index of representations for the maintenance phase of VWM in the present study. 

In the following four experiments, two kinds of stimuli with different levels of complexity, i.e., simple objects and complex objects, were used as materials. The term complexity used here does not refer to the intrinsic physical property of an object, but refers to the amount of details necessary to perform the task. For the simple objects, retaining highly discriminable simple features (i.e., low-resolution information) is sufficient to detect the change; while for the complex objects, the change signal is subtle, thus storing fine detailed features is necessary to detect the subtle change. Pilot behavioral studies indicated that only about 2 complex objects can be maintained in VWM. Therefore, if the fine detailed information can't be stored in VWM, and VWM capacity is set by a fixed number of objects regardless of complexity, to anticipate, the amplitude of CDA for 4 objects will be always higher than that for 2 objects; otherwise, if fine detailed information can be stored in VWM, then the amplitude of CDA may be modulated by the object complexity. 

Experiment 1

Previous research showed that the search rate of random polygons is slow (about 70–80 ms/item) and only 2 of them can be remembered (Alvarez & Cavanagh, 2004), indicating they are a kind of complex stimuli. In Experiment 1, we chose random polygons (adapted from Alvarez & Cavanagh, 2004) as complex shapes, and basic shapes as simple shapes, to explore the effect of complexity on VWM capacity. 

Methods

Participants

Twelve right-handed students (6 females) from Zhejiang University were paid to participate in this experiment. Participants reported no history of neurological problems, all with normal or corrected-to-normal vision. 

Stimuli

Two types of shapes were used ( Figure 1): 6 random polygons and 6 basic shapes. Each object subtended a visual angle of 1.23° × 1.23°. All stimuli were black, and presented on a gray background. 

Figure 1

 

Objects used in Experiment 1.

Experimental design

Participants were seated in an electrically shielded and sound-attenuated recording chamber at a distance of 70 cm from a 17-inch monitor. Stimulus arrays were presented within two 4° × 7.3° rectangular areas, centered 3° to the left and right of a central fixation cross on a gray background. The memory array consisted of 2 or 4 different shapes in each hemifield. Stimulus positions were randomly selected in each trial, while the distance between items within a hemifield was at least 2° (center to center). The basic shapes or random polygons were used as materials in different blocks. The shape of each item in the memory array was selected in random from the same category of shapes without repetition. 

Procedure

Each trial began with a 200 ms arrow cue presented over a fixation point, pointing either to the left or right. After a variable delay, which ranged from 250 to 350 ms, a 500 ms memory array was displayed, followed by a 900 ms blank period and then, a 2000 ms test array ( Figure 2). Participants were required to keep their eyes fixated while to remember the shapes in the hemifield indicated by the arrow cue. One shape in the test array in the cued hemifield was different from the corresponding shape in the memory array on 50% of trials; the shapes of the two arrays were identical on the remaining trials. When a shape changed between the memory and test arrays, a new shape was selected at random from shapes that hadn't appeared in the memory array. The subject's task was to indicate whether the memory and test arrays were the same or different, with the accuracy rather than the speed of the response being stressed. Each of the participants was tested in two sessions: one was the basic shape condition, and the other was the random polygon condition. The two sessions were counterbalanced in displaying order. Each session had 4 blocks, and each trial block lasted about 6 minutes with 2-minute break in between. Each subject performed 160 trials per set size. 

Figure 2

 

Example of a change memory trial for the right hemifield in Experiment 1.

Electrophysiological recording and analyses

The EEG was recorded from 32 scalp sites using Ag/AgCl electrodes mounted in an elastic cap, with the reference on the left and right mastoids. Vertical electrooculogram (VEOG) and horizontal electrooculogram (HEOG) were recorded with two pairs of electrodes, one pair placed above and below the left eye, and another pair placed beside the two eyes. All inter-electrode impedance was maintained below 5 ÊΩ. The EEG and EOG were amplified by SynAmps (NeuroScan Inc. Sterling, Virginia, USA) using a 0.05–100 Hz bandpass and continuously sampled at 1000 Hz/channel for off-line analysis. 

Eye movements and blinks were corrected using an ICA procedure (Jung et al., 2000). Remaining artifacts exceeding ±75 μV in amplitude were rejected. Artifact free data were then segmented into epochs ranging from 200 ms before to 1400 ms after memory onset for all conditions. Five pairs of electrode sites at posterior parietal, lateral occipital and posterior temporal areas (P3/P4, CP3/CP4, P7/P8, TP7/TP8, and O1/O2) were chosen for analysis. The contralateral waveforms were computed by averaging the activity recorded at left hemisphere electrode sites when participants were cued to remember the right side of the memory array with the activity recorded from the right hemisphere electrode sites when they were cued to remember the left side. CDA was constructed by subtracting the ipsilateral activity from the contralateral activity. The averaged CDA waveforms were smoothed by applying a low-pass filter of 17 Hz (24 dB). 

Pilot studies showed that the contralateral activity began to diverge from ipsilateral activity at about 200 ms after the memory array onset and persisted during the whole maintenance period, therefore a measurement window of 300–1400 ms after the onset of the memory array was adopted in the present study. For factors that had more than two levels, the Greenhouse-Geisser Epsilon was used to adjust the degrees of freedom. 

Results and discussion

Behavioral data

As shown in Figure 3, the change detection accuracy declined as object complexity and set size increased. A two-way analysis of variance (ANOVA) with the factors of complexity (simple vs. complex) and set size (2 vs. 4) yielded significant main effects of complexity, F(1,11) = 169.24, p < 0.001, and set size, F(1,11) = 382.61, p < 0.001, yet no significant interaction of these factors, F(1,11) = 2.414,p = 0.149. Memory capacity under each set size was estimated with Cowan's K formula (Cowan, 2001) for simple object condition and complex object condition, separately.¹ Capacity estimates (K) showed that 1–2 random polygons could be remembered (K = 1.15 for set size 2, K = 1.05 for set size 4), and 2–3 items for basic shapes (K = 1.74 for set size 2, K = 2.55 for set size 4). 

Figure 3

 

Averaged behavioral results in Experiment 1.

ERP data

Consistent with previous research (McCollough, Machizawa, & Vogel, 2007; Vogel & Machizawa, 2004) and our behavior result, increasing set size from 2 to 4 resulted in a substantial increase in the amplitude of CDA in the basic shape condition (Figure 4a). However, the amplitude no longer kept increasing for random polygons (Figure 4b). Taking set size and electrodes as factors, a two-way ANOVA on the mean amplitude of CDA was conducted to test this effect for each shape condition. The results in the basic shape condition revealed a significant main effect of set size, F(1,11) = 6.652, p = 0.041, suggesting the amplitude of remembering 4 basic shapes was higher than that of remembering 2 basic shapes. The main effect of electrodes was non-significant, F(4,44) = 2.613, p = 0.091, and so was the interaction between the two factors, F(4,44) = 0.808, p = 0.488. Importantly, the two-way ANOVA in the random polygon condition yielded no main effect of set size, F(1,11) = 0.771, p = 0.40, indicating that the amplitude of remembering 4 polygons was no higher than that of remembering 2 polygons. Besides, the main effect of electrodes was significant, F(4,44) = 3.502, p = 0.025, and the interaction between the two factors was non-significant, F(4,44) = 1.719, p = 0.205. Hence, the ERP results in Experiment 1 showed that the amplitude of CDA can be modulated by the complexity of materials, suggesting that the fine detailed information can be stored in VWM, and the detailed information from 2 complex objects has already used up VWM storage resource. 

Figure 4

 

Averaged ERP results in the basic shape (a) and random polygon condition (b) of Experiment 1.

Experiment 2

In Experiment 1, the complexity of memory items was manipulated by adopting two different kinds of stimuli, random polygons and basic shapes. Here, we attempted to manipulate the complexity within objects by modulating top-down task requirement. Specifically, the same set of random polygons in Experiment 1 but with different colors were used as materials, and the participants were asked to remember the different dimension of the objects. According to Luck and Vogel (1997), 3–4 objects can be remembered regardless of the complexity of features they contain. In contrast, according to Alvarez and Cavanagh (2004, 2008), VWM capacity is different for complex features and simple features. Thus, our goal in Experiment 2 was to explore whether the amplitude of CDA can be modulated by the complexity of object features to be encoded. 

Methods

Participants

Twenty-four right-handed students from Zhejiang University were paid to participate in this experiment. Each experiment session had 12 participants (7 females). All participants reported no history of neurological problems, all with normal or corrected-to-normal vision. 

Stimuli

Random polygons in Experiment 1 with different colors were used as materials. Each polygon had 7 possible colors, namely, red, green, blue, violet, yellow, black, and white. The color and shape of each item were randomly selected without repetition for the memory array. 

Experimental design

The experimental design was the same as in Experiment 1. 

Procedure

All aspects of Experiment 2 were the same as in Experiment 1 except that there were two sessions of trials and each group of participants only participated in one of them. In the color session (i.e., simple feature group), the participants only needed to remember the color of polygons while the shapes of polygons between the memory array and test array were identical. The other session was random polygon session (i.e., complex feature group), wherein the participants only needed to remember the shapes of polygons and the colors between the memory array and test array were identical. In each session, participants were instructed to report whether the memory and test arrays were the same or different in the corresponding target feature dimension. 

Electrophysiological recording and analyses

Recording and analyses were the same as in Experiment 1. 

Results and discussion

Behavioral data

As shown in Figure 5, the accuracy for color was much higher than that for random polygons, t(22) = 11.038, p < 0.001. In addition, the change detection accuracy declined as the set size kept increasing from 2 to 4 in both conditions (color: F(1,11) = 36.250, p < 0.001; random polygons: F(1,11) = 63.367, p < 0.001). Memory capacity under each feature condition was estimated, and the results showed that about 1 random polygon could be remembered ( K = 0.86 for set size 2, K = 0.69 for set size 4), while 2–3 for colors ( K = 1.83 for set size 2, K = 2.53 for set size 4). 

Figure 5

 

Averaged behavioral results in Experiment 2.

ERP data

The ERP results of Experiment 2 ( Figure 6) were similar to those of Experiment 1. A two-way ANOVA in the color condition yielded a significant main effect of set size, F(1,11) = 5.914, p = 0.033, suggesting the amplitude of remembering 4 items was higher than that of remembering 2 items; the main effect of electrodes was non-significant, F(4,44) = 2.275, p = 0.120, and the interaction between set size and electrodes was non-significant, F(4,44) = 1.061, p = 0.359. However, as to the random polygon condition, the ANOVA found a non-significant main effect of set size, F(1,11) = 0.861, p = 0.373, suggesting there was no difference on the amplitude of CDA between retaining 2 and 4 polygons. Besides, the main effect of electrodes was marginally significant, F(4,44) = 2.737, p = 0.067, and the interaction was non-significant, F(4,44) = 1.185, p = 0.330. Therefore, by modulating the top-down task requirement, the ERP results of the present experiment further implied that the fine detailed information can be stored in VWM. More importantly, even though the features belong to the same set of objects, the VWM resource has been exhausted by the detailed information from 2 objects, contrast to 4 objects in the simple feature condition. 

Figure 6

 

Averaged ERP results in the color (a) and random polygon condition (b) of Experiment 2.

Experiment 3

Experiment 3 had two aims. Firstly, we intended to investigate the nature of fine detailed information. Secondly, we explored whether the complexity effect found in Experiments 1 and 2 could be extended to other kinds of stimuli. We hypothesized that the fine detailed information was the one required serial, attentive perceptual processing. To test this hypothesis, we adopted the landolt rings as materials. Though a landolt ring looks like a kind of simple object at the first glance, previous research suggested that the process of a gap's orientation of landolt needs focal attention, with a search rate of about 100 ms/item in the visual search task (Gao, Shen, Gao, & Li, 2008; Shen et al., 2007; Woodman & Luck, 2003). Therefore, according to the above hypothesis, gap's orientation is a kind of fine detailed information. We predicted that there would be no increase between the CDA amplitude of 4 gap's orientations and 2 gap's orientations for retaining fine detailed information. 

Methods

Participants

Fourteen right-handed students (6 females) from Zhejiang University were paid to participate in this experiment. All participants reported no history of neurological problems, all with normal or corrected-to-normal vision. 

Stimuli

Eight landolt rings with different colors were used ( Figure 7). Each of them subtended a visual angle of 1.01° × 1.01°. The same set of colors used in Experiment 2 was adopted. There was a gap in each item, whose orientation was selected from a set of 8 orientations: 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. The colors and orientations of the rings in the memory array were selected randomly with the constraint that no more than 2 items shared the same color or orientation. 

Figure 7

 

Colored landolt rings with eight possible gap orientations used in Experiment 3 and red landolt ring with 0° orientation was illustrated.

Experimental design

The experimental design was the same as in Experiment 1. 

Procedure

All aspects of Experiment 3 were identical with Experiment 1 but with the following exception. Each participant was tested in two sessions. One was the color session, in which the participants only needed to remember the color of landolt rings while the gaps of landolt rings between the memory and test arrays were identical. The other was the orientation session, in which the participants only needed to remember the orientations of the gaps and the colors between the memory and test arrays were identical. Here, the complexity level in the first session was low while high in the second session. Participants were instructed to report whether the memory and test arrays were the same or different in the corresponding target feature dimension. Each session had 2 blocks, each trial block lasting about 6 minutes with 2 minutes break in between. Each subject performed at least 84 trials per set size. 

Electrophysiological recording and analyses

Recording and analyses were the same as in Experiment 1. 

Results and discussion

Behavioral data

The accuracy for color was much higher than that for orientation ( Figure 8). A two-way ANOVA with factors of complexity (simple vs. complex) and set size (2 vs. 4) found significant main effects of complexity, F(1,13) = 27.673, p < 0.001, and set size, F(1,13) = 203.492, p < 0.001, yet a non-significant interaction, F(1,13) = 2.181, p = 0.164. Capacity estimates showed that about 1–2 gap's orientations could be remembered ( K = 1.50 for set size 2, K = 1.52 for set size 4), 2–3 for colors ( K = 1.76 for set size 2, K = 2.34 for set size 4). 

Figure 8

 

Averaged behavioral results in Experiment 3.

ERP data

The ERP results of Experiment 3 ( Figure 9) replicated the response profiles we got in Experiments 1 and 2. A two-way ANOVA on the mean amplitude of CDA in the color condition yielded a significant main effect of set size, F(1,13) = 9.459, p = 0.009, suggesting the amplitude of retaining 4 objects was higher than that of retaining 2 objects. Besides, the main effect of electrodes was significant, F(4,52) = 4.572, p = 0.010, yet there was a non-significant interaction of the two factors, F(4,52) = 0.822, p = 0.486. As to the gap's orientation condition, importantly, the main effect of set size was non-significant, F(1,13) = 1.179, p = 0.297, indicating that there was no difference on the amplitude of CDA between retaining 2 and 4 gap's orientation. The main effect of electrodes was significant, F(4,52) = 3.130, p = 0.046, yet the interaction between the two factors was non-significant, F(4,52) = 0.407, p = 0.738. Therefore, the current experiment supported our hypothesis about the nature of fine detailed information, which needs focal attention to process. Furthermore, the complexity effect that the amplitude of CDA is modulated by object complexity can be extended to the landolt rings, further supporting our conclusion about the storage of fine detailed information in VWM. 

Figure 9

 

Averaged ERP results in the color (a) and gap condition (b) of Experiment 3.

Experiment 4

Some may argue that the use of blocked trials of simple and complex objects could lead participants to adopt different strategies in separate blocks. Specifically, as we have voluntary control over the number of objects we choose to store in VWM, participants might simply choose to store 2 objects in the case of the complex objects, thus leading to the above results that the amplitude of CDA didn't raise from set size 2 to 4. To testify this alternative possibility, we intermixed the simple objects and complex objects used in Experiment 1 within a single block. In this case, it was less likely that the participants would choose to retain a smaller number of items for the complex objects condition, because there were 50% of trials in which participants who retained as many objects as possible would be rewarded by accurate change detection. To anticipate, if the strategy indeed influenced the number of objects that participants chose to hold in VWM, there would be no difference on the response profiles of CDA between simple objects and complex objects. 

Methods

Participants

Twelve right-handed students (6 females) from Zhejiang University were paid to participate in this experiment. All participants reported no history of neurological problems, all with normal or corrected-to-normal vision. 

Stimuli

The stimuli were the same as in Experiment 1. 

Electrophysiological recording and analyses

All sites were recorded with a left-mastoid reference, and the data were re-referenced offline to the algebraic average of the left and right mastoids. The other aspects were the same as in Experiment 1. 

Results and discussion

Behavioral data

The change detection performance ( Figure 10) was similar to that in Experiment 1. A two-way ANOVA with the factor of complexity (simple vs. complex) and set size (2 vs. 4) found significant main effects of complexity, F(1,11) = 121.821, p < 0.001, and set size, F(1,11) = 216.514, p < 0.001, yet no significant interaction of the two factors, F(1,11) = 1.035, p = 0.331. Capacity estimates ( K) showed that only 1 random polygons could be remembered ( K = 0.98 for set size 2, K = 0.80 for set size 4), and 2–3 items for basic shapes ( K = 1.73 for set size 2, K = 2.13 for set size 4). 

Figure 10

 

Averaged behavioral results in Experiment 4.

ERP data

Visual inspection of Figure 11 suggested the amplitude of CDA for 4 basic shapes were higher than that for 2 basic shapes; however, the amplitude of CDA for 4 random polygons was still no higher than that for 2 random polygons. To evaluate the significance of set size, a two-way ANOVA was conducted separately for each shape condition on the mean amplitude of CDA. They showed a significant main effect of set size in basic shape condition, F(1,11) = 14.976, p = 0.003, but no such significant main effect of set size in the random polygon condition, F(1,11) = 3.332, p = 0.095, and the amplitude of CDA for 2 random polygons was even higher than that for 4 random polygons in some electrodes, suggesting the complexity of object modulated the amplitude of CDA, moreover, VWM can not always store all the information from 4 objects regardless of complexity. The other aspects were all non-significant, all ps > 0.1. Overall, the results of Experiment 4 suggested that participants didn't adopt different strategies in the different sessions of trials of Experiments 1, 3, and 3. 

Figure 11

 

Averaged ERP results in the basic shape (a) and random polygon condition (b) of Experiment 4.

General discussion

VWM holds information that is actively being used in cognitive performance. Due to the processing limitation of brain system, only a small amount of information can be selected and consolidated into VWM for further processing. To avoid the interference of comparison errors on the estimation of VWM capacity (Awh et al., 2007) while investigating the maintenance phase directly, the storage mechanism of VWM for the fine detailed information was explored by taking the amplitude of CDA as an index. Our ERP results revealed that for maintaining simple objects, the amplitude of CDA for 4 objects was higher than that for 2 objects; however, for maintaining fine detailed information, there was no difference. So far as we know, this is the first time that clear ERPs evidence has been presented, suggesting that VWM does not merely represent a fixed number of objects, but also is affected by the fine detailed information contained in the memory materials. 

One of the impressive results from Awh et al. (2007) is that behavioral performance did not drop with the increment of object complexity, as long as the difference between memory array and test array was salient. It suggests that the comparison between memory and test arrays plays an important role in the complexity effect reported in previous studies (Alvarez & Cavanagh, 2004; Olson & Jiang, 2002; Wheeler & Treisman, 2002; Xu, 2002). However, from this result alone, it is still unclear how the on-line maintenance of object representation is influenced by the fine detailed information of each object. The present study took the amplitude of CDA as the dependent measure, which can directly track the maintenance process itself and exclude any effect from the comparison phrase. Indeed, on each trial, recording of the CDA signals has been finished before the presentation of test array. Our findings clearly indicate that the amplitude of CDA is modulated by fine detailed information, thereby providing strong evidence for the existence of a stage of processing in VWM, which is affected by fine detailed information contained in the objects. 

To provide a comprehensive interpretation for the effect of fine detailed information on CDA, it is necessary to briefly review other recent models of VWM which have emphasized object complexity. To reconcile the evidence that on one hand, storage of simple features reveals strong object-based processing (Awh et al., 2007; Luck & Vogel, 1997), on the other hand, the complexity of memory array also significantly influences the performance (Alvarez & Cavanagh, 2004; Olson & Jiang, 2002; Wheeler & Treisman, 2002; Xu, 2002), some researchers suggested that storage in VWM is not a unitary process, but consists of two dissociable stages whose capacities are limited by different types of visual information. Alvarez and Cavanagh (2008) proposed that VWM involves at least two stages of processing. In the first stage, about 4 objects' low resolution boundary features (e.g., the shape of the outer contour in a Gabor patch) are extracted, regardless of the complexity of each object. Those low-resolution boundary features serve as the indexing features for retrieving other information. In the second stage of processing, based upon those boundary features, more fine detailed surface features (e.g., the striped texture in a Gabor patch) can be encoded and maintained. However, storage of high-resolution surface features in this stage is limited by object complexity: the more surface feature details that an object contains, the smaller the number of objects that can be stored. Similarly, Gao et al. (2008) suggested that there are dissociated mechanisms in VWM for storing information extracted at different stages of perceptual processing. The object-based storage in VWM is not based on the coherent object representation assembled by serial attentive processing; on the contrary, it is originated from highly discriminable simple features, which have already been segmented into individual low-resolution objects at the end of the parallel perceptual processing. The output of serial, attentive perceptual processing (e.g., color-color conjunction), which is a kind of fine detailed information, can not be added to the object representations in VWM for free, but in need of extra resources. The two-stage processing hypothesis has also received support from recent neuroimaging results. Xu and Chun (2006, 2007, 2009) found two dissociable neural mechanisms mediating VWM: 4 objects at most are firstly selected and encoded by the inferior intraparietal sulcus (IPS) by their spatial locations; then a subset of these selected objects are encoded into the high-resolution ones by the superior IPS depending on complexity. 

All the findings taken together (Alvarez & Cavanagh, 2008; Awh et al., 2007; Xu & Chun, 2006; the current research), the present results confirm that the amplitude of CDA has substantial connection with VWM capacity. However, it isn't correlated with the processing of a fixed number of object representations containing only low-resolution boundary features (Alvarez & Cavanagh, 2008) or the low-resolution output of parallel perceptual processing (Gao et al., 2008). On the contrary, it reflects the second stage of storing, i.e., the maintenance of fine detailed information, which is not only influenced by the number of objects, but also limited by the encoding level of information resolution of each object. Accordingly, just as Awh et al. revealed, there is indeed a stage in VWM that can hold a fixed number of objects, which actually reflects the property of the first stage of storing low-resolution information. 

The main focus of the current research is on the mechanism for storing fine detailed information. Therefore, we adopted complex objects as materials, since it is a widely adopted and much convenient, intuitive way to manipulate fine detailed information (e.g., Alvarez & Cavanagh, 2004, 2008; Eng et al., 2005; Makovski & Jiang, 2008). However, further research needs to be carried out to explore the nature of this fine detailed information. Firstly, to ensure the fine detailed information to be remembered, the current research adopted a within-category change task, while two kinds of complex objects were tested. It may be possible that the specific task and stimuli used in the current research lead the participants to adopt a kind of strategy, in which only about 2 complex objects were remembered. So the generalizability of our conclusion should be further tested by using some other different tasks and complex objects. Secondly, simple objects can also own fine detailed information. For example, we need to encode the fine detailed information of the simple objects (e.g., orientated bars) when the change is subtle (see Jiang et al., 2008 for an example). It is thus an intriguing and theoretically important question to know the storage mechanism of the fine detailed information for the simple objects. Besides, despite that the current research provides evidence supporting the hypothesis that in a certain stage of processing, storage in VWM is limited by fine detailed information contained in the object, the mechanism of how fine detailed information influences storage remains largely incomplete. For instance, how are the objects in a previous lower level stage selected into that stage? How is the object represented in that stage? All these questions need further exploration. 

The current findings about CDA have not only enriched the theories of VWM, but also provided methodological implications on how to use CDA appropriately as a powerful tool to explore VWM in future. To our knowledge, CDA is the only known neural signature of VWM by which one can track the on-line maintenance of visual information dynamically (see a review, Drew, McCollough, & Vogel, 2006). In Vogel and Machizawa (2004), low resolution simple objects were adopted as memory material and the results revealed a strong correlation between the amplitude of CDA and the number of objects held in VWM. Driven by their results, we had initially expected that the amplitude of CDA may only correlate with the number of objects, regardless of object complexity. In that case, we could take CDA as an index to diagnose what counts as an ‘object’ in VWM. However, after a series of experiments reported in the current research, we found quite an opposite result: the amplitude of CDA can be strongly influenced by the encoding level of information resolution of the memory materials. This unexpected result indicates that in future research, while employing CDA as a tool to measure the number of objects being held in VWM, researchers should be cautious to exclude any potential confounding effect from object complexity. Nevertheless, it still promises as an alternative way in future studies. As discussed above, how the fine detailed information is processed in VWM is still far from clear. Suppose CDA is reliably correlated with the degree of complexity for the memory materials, one can thus safely adopt it to uncover the algorithm for computing ‘complexity’ in VWM. 

Acknowledgments

This research is supported by the National Natural Science Foundation of China (No. 30870765; No. 30570604), Key Project of Humanities and Social Sciences, Ministry of Education (No. 07JZD0029), Key Project of National Social Science Foundation of China (No. 07AY001), Fund of the Ministry of Education for Doctoral Programs in Universities of China (No. 20060335034), the National Foundation for Fostering Talents of Basic Science (No. J0630760) and the Research Center of Language and Cognition, Zhejiang University. We are grateful to Tao Gao, Edward Awh, Lun Zhao, and an anonymous reviewer for insightful comments. We are also indebted to Yisheng Dong and Wenjun Yu for assistance with figure production. 

Commercial relationships: none. 

Corresponding author: Mowei Shen. 

Email: [e-mail]. 

Address: Department of Psychology, Xixi Campus, Zhejiang University, Hangzhou, 310028, P.R. China. 

Footnote

Footnotes

¹  Though K may be underestimated under complex object condition (Awh et al., 2007), here we calculated K for the following reasons. Firstly, K is an index sensitive to the fine detailed information contained in the object. Secondly, K can be adopted to test the validity of object complexity manipulation and to make a direct comparison with previous work.