Introduction

Our visual system is constantly being exposed to a rich and rapidly changing scenery. Rather than processing all available sensory inputs with an equal level of depth, we select a specific subset that is supposedly more relevant to the tasks at hand and process the selected stimuli in more detail (R Desimone, 1995; Moran & Desimone, 1985). Consequently, our perception of the environment needs to be a highly active process that is guided flexibly and dynamically based on the behavioral relevance (Reynolds et al., 2000; Treue & Martínez Trujillo, 1999) or possible occurrence of an upcoming stimulus (Rohenkohl et al., 2014). Two mechanisms control the allocation of resources in our perceptual system: involuntary bias towards features of the visual stimulus that stand out, and control processes that prioritize the sensory inputs based on collected information (Connor et al., 2004).

It is increasingly recognized that the anticipatory biases are not restricted to signals about locations (R Desimone, 1995; Reynolds et al., 2000) or object features (Müller et al., 2006; Treue & Martínez Trujillo, 1999). The behavioral bias also extends to the temporal domain (Nobre & Van Ede, 2018). The ability to extract temporal patterns and the regularity of events has long been known to improve action preparation and execution (Nobre et al., 2007). While many studies address the role of spatial and feature attention in visual working memory (Averbach & Coriell, 1961; Murray et al., 2011; Pertzov & Husain, 2014; Rajsic & Wilson, 2012; Sperling, 1960), the effects of temporal expectation on encoding and recognition of visual stimuli are not well established. In fact, little is known about how expecting a stimulus in time can enhance its retrieval from memory (Cravo et al., 2017).

Moreover, as a consequence of its intrinsic bias, our perception is not homogenous across the visual field (Carrasco et al., 2001). It is well established that performance not only decreases with eccentricity (Cannon, 1985; Carrasco et al., 1995; Legge et al., 2001; Rijsdijk et al., 1980), but also varies at iso-eccentric locations across the visual field as a function of polar angle (Barbot et al., 2021; Carrasco et al., 2001; Fuller et al., 2008). This asymmetric pattern of visual perception is sometimes referred to as the “visual performance field” (Altpeter et al., 2000; Cameron et al., 2002). The extent and direction of this field vary depending on the task (Himmelberg et al., 2020; Thomas & Nicholls, 2018). For instance, simple visual stimuli (such as Gabor patches) are more accurately identified in the lower compared with the upper visual field (Barbot et al., 2021; Himmelberg et al., 2020), whereas more complicated stimuli, like faces, are better categorized when they are presented in the upper visual field (Quek & Finkbeiner, 2016).

Differences in performance across the visual field are well documented in stimulus detection. But surprisingly, little empirical research has addressed spatial gating into working memory. A few studies show that the perceptual discrimination asymmetry (i.e., better performance in the lower and horizontal meridian compared with the upper visual field and vertical meridian) is carried over to tasks engaging short-term memory (Himmelberg et al., 2020; Montaser-Kouhsari & Carrasco, 2009). But other studies have demonstrated that the memory load has a drastic impact on the advantage of one visual hemifield over the other. For example, it has been shown that although single-feature items that are displayed on the left visual hemifield yield better performance, encoding and storing items including more than one feature are improved when they are presented on the right visual hemifield (S. Sheremata & Shomstein, 2014). Therefore, it is still unclear how these performance asymmetries around the visual field evolve: (1) with increasing set size, (2) when stimuli are presented simultaneously and are competing for resources, and (3) when stimuli are presented in several locations across the visual field and not just in the two visual hemifields (S. Sheremata & Shomstein, 2014) or four cardinal points (Montaser-Kouhsari & Carrasco, 2009).

Many features and attributes of stimuli that are encoded into working memory have been studied extensively in the past. However, it is less clear how location and temporal-onset anticipation of stimulus might affect its retrieval from memory. Working memory is conceptualized as a system that enables us to temporarily store and manipulate small amounts of sensory information which fades in the order of seconds (Baddeley, 2010). To support complex behaviors such as comprehension and learning, working memory operates as a mental workspace that maintains goal-relevant representations of perception in an active state although they may no longer be physically present (Baddeley, 2012; Fuster, 1990; Ma, 2014). While working memory serves as a core cognitive process that is vital for everyday functions, it has a limited capacity (Alvarez & Cavanagh, 2005; Luck & Vogel, 1997). Therefore, it is crucial for our perceptual system that resources are allocated such that only behaviorally salient information is stored in working memory.

Here we studied two key aspects of visual stimulus processing, namely: temporal attention and stimulus spatial location over the visual field, on working memory performance. For this purpose, we used colored cues to manipulate temporal expectations of an upcoming stimulus in a visual recognition task while a different number of sample stimuli were presented across the visual field. Consistent with the previous findings, we found a visual performance field effect for stimuli in working memory. Temporal attention had a positive effect on memory performance such that the stimuli with temporally predictable onsets were more accurately recognized than the ones with unpredictable onsets. Also, stimuli that were presented on the upper or right visual hemifield yielded better performances compared with the lower or left hemifields.

Method

Participants

Thirty-three healthy volunteers (12 male; age range 20–33 years, median 26 years) took part in the experiments (13 for Experiment 1 and 20 for Experiment 2). The post hoc power analysis with G*Power (Erdfelder et al., 1996) revealed that while we recruited 13 participants for Experiment 1, the number of trials in that experiment was sufficient to detect the effects of temporal attention and stimulus location. Regarding the effect of temporal attention, the power to detect large (Cohen’s d = 0.8; cf. Cohen, 2013), medium (Cohen’s d = 0.5), and small-sized (Cohen’s d = 0.25) effects were estimated to be 0.99, 0.87, and 0.72, respectively. Furthermore, power analysis of the effects of stimulus location showed a 0.99, 0.98, and 0.91 probability of detecting a large, medium, and small-sized effect, respectively. Since the number of trials was large, we predetermined the sample size based on previous studies (Montaser-Kouhsari & Carrasco, 2009; Quek & Finkbeiner, 2016; Wilsch et al., 2020).

All participants reported normal or corrected-to-normal vision. All procedures in this study were performed in accordance with the Declaration of Helsinki. Approval was granted by the ethics committee of Shahid Beheshti University of Medical Sciences (IR.SBMU.MSP.REC.1396.757). Participants were fully aware of the nature of the study, provided written informed consent before testing, and were compensated for their time with course credit or money.

Experimental task and stimuli

All visual stimuli were created using MATLAB R2019b (Mathworks, Natick, MA, USA) and the Psychophysics Toolbox extensions (Brainard, 1997; Pelli, 1997) and presented on a 13-in. retina display.

A visual recognition task was used to measure the effect of temporal attention and stimulus location on visual working memory performance. Participants were seated in a dimly lit and quiet room at a 65-cm distance from the screen. They were presented with a stimulus-set followed by a probe and had to indicate whether the probe was included in the stimulus-set or not.

Each trial began with a colored disc with a diameter of ~2 visual degrees that acted as a temporal cue and was displayed for 100 ms. The color of the temporal cue indicated whether the stimulus-set would be presented following a short (650 ms: orange), long (1,400 ms: blue), or unpredictable (randomly selected between 650, 900, 1,150, or 1,400 ms: grey) delay. Four foreperiods were used in the unpredictable condition to increase the uncertainty of the temporal onset of stimuli in each trial. Then an array of different number of alphabetical stimuli (constituting four to ten letters; see Set size manipulations) was displayed and the participants were explicitly instructed to memorize as many letters as possible. Each letter had a diameter of around 1.2 visual degrees. When the letters were put together, they made-up an imaginary circle with 3.75° radius centered at the fixation point such that all letters were equidistant from the center of gaze. Stimulus-set remained on the screen for 100 ms to ensure saccades toward the stimuli cannot be executed (Fischer & Ramsperger, 1984; Montagnini & Chelazzi, 2005). After a delay of 2,600 ms, during which only the fixation point was displayed, a single letter (probe) was presented centrally to the participant. In half of the trials (where the probe was included in the stimulus-set) the probe was randomly selected from one stimulus location in each difficulty level with equal probability and remained on the screen until the subject responded by a keypress if the letter was included in the stimulus-set or not. We asked the subjects to keep fixation on the center of the screen throughout the trial and to respond as fast and as accurately as possible. Inter-trial intervals were around 1.5 s (jittered between 1 and 2 s) and did not contain any stimulation or responses.

Visual quadrants and set size manipulations

A visual quadrant is defined as one-fourth of the visual field; the letters in each visual stimulus were organized in a circular pattern and equally distributed between quadrants. Henceforth, we refer to the upper-right quadrant (0–90° in the Cartesian coordinate system) as Q1, upper-left (90–180°) as Q2, lower-left (180–270°) as Q3, and lower-right (270–360°) as Q4 (Fig. 1a).

Fig. 1
figure 1

Schematic of visual quadrants, experimental design, and timings for Experiments 1 and 2. Colored fixation disks cued the onset of the stimulus at 1,400 ms, 650 ms, or one random selection of four delays between 650 ms and 1,400 ms. Then the stimulus-set was displayed, followed by the presentation of a single letter. The participants responded whether the probed letter was included in the stimulus-set or not by a keyboard press. a For location analysis, the visual field is divided into four quadrants, Q1 to Q4 correspond to each visual quadrant. One example trial for b long and c short foreperiod trial from experiments one and two. The two experiments only differed in the set sizes and the location of letters in each stimulus-set (two set sizes for Experiment 1 and four for Experiment 2). All the other conditions were kept constant between the two experiments. Blue, orange, and grey disks indicate long, short, and unpredictable cues, respectively; the green circle around the key button shows the correct key response

Full size image

In Experiment 1, we focused on the advantage of temporal orienting of attention on visual working memory performance. In this experiment visual stimuli consisted of four or eight letters, i.e., one or two letters in each visual quadrant, respectively.

Experiment 2 was similar to Experiment 1 except for the selected set sizes. To confirm our findings regarding the performance of participants across the visual field, we added two additional set sizes to the ones in the previous experiment, such that the number of letters ranged between four and ten in four steps (i.e., 4, 6, 8, and 10) and they were located in all visual quadrants as well as cardinal points. This allowed us to demonstrate exact performance measures for 28 different locations around the visual field as well as to generalize our findings more accurately to all the iso-eccentric points in the Cartesian plane. For easy interpretation and comparison of the results, all the other conditions of the task were the same as Experiment 1; therefore, only the results of location analysis are reported for this experiment – the number of trials for analysing the effect of predictability was not sufficient due to the increased number of trials in the location condition. Figure 1b illustrates the time course and visual stimulus of one example trial from each experiment.

Procedure

Participants performed a 10-min training prior to the main task to acquaint themselves with the associations between temporal cues and stimuli onsets (which were explicitly indicated to them at the start of the experiment) and become familiar with the procedure. Data of four subjects who performed below 75% in the four-letter condition were excluded from further analyses.

Experiments 1 and 2 comprised 768 and 288 randomly presented trials and lasted ~3 h and ~1 h (including breaks), respectively. The number of trials in all set sizes was the same (one-half and one-fourth of the total number of trials in Experiments 1 and 2, respectively). Trials were also equally divided into predictable and unpredictable; for accurate comparison between predictable and unpredictable conditions, we analyzed trials with foreperiods comparable to the predictable trials (650 and 1,400 ms), therefore, the number of trials in every unpredictable condition is half of those in the predictable condition (Samaha et al., 2015). Set size was manipulated independently of the temporal cue (predictability) condition.

Data analysis

To control for speed-accuracy trade-offs, we used a measure called inverse efficiency score (IES) that combines reaction times (RTs) and the proportion of correct responses (C. D. Chambers et al., 2004; Romei et al., 2011; Snodgrass et al., 1985). It is defined as the average of RTs divided by the proportion correct (Rohenkohl et al., 2014) using the equation below:

$$ IES=\frac{Reaction\ Time}{Proportion\ of\ Correct\ Responses\ } $$

To measure whether the visual quadrant in which the stimulus was displayed affected subjects’ performance True Positive Rates (TPR) were calculated separately for each location and each participant as below:

$$ TPR=\frac{n. of\ True\ Positive\ trials\ (hits)}{n. of\ True\ Positive\ trials\ (hits)+ False\ Negative\ trials\ (misses)} $$

In the location analysis, TPR, that is the accuracy in trials where the probe was an item from the stimulus-set, is used as the performance measure. The proportion of correct responses, which is more common in behavioral studies, includes both true negative and true positive trials; however, when the location of the displayed stimulus is of interest, true negative rates (i.e., participants correctly responded “no” when the probe was not displayed) are irrelevant. Therefore, TPR or the rate of true positive responses (i.e., participants correctly responded “yes” when the probe was displayed in the stimulus-set) was utilized for this purpose.

To demonstrate the association between performance decline and the location of stimuli, we defined an index called the performance change index (PCI); it illustrates the difference between recognition performance in easy (four letters) and difficult (eight letters) conditions and is calculated as below:

$$ PCI=\frac{TPR_{4\_ letter}-{TPR}_{8\_ letter}\ }{TPR_{4\_ letter}+{TPR}_{8\_ letter}} $$

where TPR is the true positive rate for easy (4_letter) and difficult (8_letter) conditions. PCIs were then plotted for all participants in each visual field quadrant.

Accordingly, to eliminate the effects of superior performances in easier conditions we calculated the standardized TPR (i.e., z-score [TPR]) in each difficulty level, according to the following formula:

For each difficulty level:

$$ z- score\ \left[ TPR\right]=\frac{TPR\ in\ each\ position- average\ TPR}{standard\ deviation\ of\ TPR} $$

For each participant, z-score [TPR] was measured separately for each difficulty level and location (i.e., four to ten possible locations dependent on the set size) using the corresponding group averaged TPR. Hence, z-score [TPR] in each visual quadrant was calculated in 192 and 36 trials (768 divided by four visual quadrants and 288 minus 144 [trials in which probe was located on the cardinal points in the stimulus-set], divided by 4) for Experiments 1 and 2, respectively.

In order to establish how recognition performance varied according to the location of the stimulus in the Cartesian coordinate system, we fitted the association between standard TPRs and the location of stimuli to a Fourier series using the equation below:

$$ y=b+a\ast \mathit{\cos}\left(w\ast x\right)+c\ast \mathit{\sin}\left(w\ast x\right) $$

where y is z-score [TPR] and x corresponds to stimulus location in degrees (0–360°). The parameters b, a, c, and w were calculated by the fitting procedure; where b was the intercept, w was the frequency and a, and c were the amplitude of cosine and sine waves, respectively. A MATLAB-based curve fitting tool was employed to determine the model fit equation; it deploys the least squares fitting and calculates the parameters by guessing and improving the guess iteratively conforming to how well the model fits the data. The initial parameters in our model were: b = 0, a = 0, c = 0 and w = 1.952.

Statistical analysis

All statistical analyses were performed using MATLAB R2019b. Prior to all the analyses, participants’ performances were calculated and those with an accuracy below 75% in the easy condition (two in Experiment 1 and two in Experiment 2) were excluded for further analysis; this resulted in 11 participants for Experiment 1 and 18 for Experiment 2.

For RT analyses, trials with RTs longer than 4 s and shorter than 200 ms were excluded (less than 1% of trials). Data were tested for normality using Kolmogorov-Smirnov, DAgostino Pearson, and Jarque-Berat tests and, therefore, met the assumptions of the analysis of variance (ANOVA and repeated-measures ANOVA). All tests were two-tailed, and the alpha level was set at 0.05, and where appropriate p-values were adjusted for multiple comparisons using Holm Bonferroni correction.

Moreover, while Experiment 1 was run on 11 participants, the number of trials each participant completed was high enough to treat each of them as a replication. For this purpose, we ran a generalized linear model (GLM) analysis on each participant’s data with the visual quadrant as the predictor and score (hit or miss) as the response variable using the formula below (with the ‘fitglm’ function in MATLAB):

$$ score={\beta}_0+{\beta}_1\times Q2+{\beta}_2\times Q3+{\beta}_3\times Q4 $$

where score is a logical variable indicating if the response to that trial was correct or not, β is the estimate for that regressor in the GLM and Q2, Q3, and Q4 represent visual quadrants; because of the categorical nature of the regressor variable, Q1 was used as the reference variable to which all the other beta estimates were compared.

Results

Task performance was affected by the temporal onset of the stimulus

To determine the difference between predictable and unpredictable temporal cues proportion of correct responses and inverse efficiency scores from Experiment 1 were submitted to a 2 × 2 × 2 repeated-measures ANOVA with temporal expectation (predictable vs. unpredictable), foreperiod (short vs. long), and difficulty level (four vs. eight stimuli) as within factors (n = 11). This resulted in 96 trials for predictable and 48 for unpredictable condition per cell design per participant.

We found a significant main effect of difficulty (F(1,10) = 91.23, p = 0.0000, η2 = 0.72) with higher proportion of correct responses for easier conditions, which confirms that our difficulty manipulations were indeed effective. We also found a significant interaction between predictability and difficulty (F(1,10) = 9.17, p = 0.0127, η2 = 0.0003) (Fig. 2a). This indicates that in the difficult condition there was a larger difference between predictable and unpredictable condition compared to the easy condition. Furthermore, the proportion of correct responses in individual participants also showed a similar effect (Online Supplementary Material (OSM) Fig. 1a).

Fig. 2
figure 2

The impact of temporal attention on performance measures in different set sizes. Plots show average accuracy and inverse efficiency score (IES) in short (left plots) and long (right plots) foreperiods. a Proportion of correct responses and b the average IES in Experiment 1. Performance decreases as the difficulty of stimulus-set increases in both proportions correct and IES. Furthermore, IES is significantly higher in predictable conditions. In both proportions correct and IES there is also a significant interaction between difficulty and predictability. Results are shown for short and long foreperiods, separately, and colors indicate the predictability conditions (blue: predictable, red: unpredictable). Error bars are standard errors of the mean

Full size image

We observed similar results with IES; there was a significant main effect of difficulty (F(1,10) = 74.11, p = 0.0000, η2 = 0.53 and a significant interaction between predictability and difficulty F(1,10) = 7.67, p-value = 0.0198, η2 = 0.0002). A main effect of predictability was also found significant (F(1,10) = 9.10, p-value = 0.013, η2 = 0.006) . This is in line with our hypothesis and demonstrates the precision of the IES measure (Fig. 2b). Moreover, these findings were consistent with IES measures in individual participants (OSM Fig. 1b).

To summarize, for both the proportion correct and the IES measures, recognition performance was better following temporally predictable, compared with unpredictable stimuli. These performance benefits were present in difficulty levels that yielded about 60-80% performances but not for the easier condition.

The location of the stimulus in the visual field affects the task performance

Another objective that we pursued was to evaluate the effects of stimulus location on performance. To this end, we adopted four different approaches to the outcome of experiments one and two.

  1. 1.

    To ascertain whether the visual quadrant in which the stimulus was displayed affected the subjects’ performance, average TPRs over all participants in experiments one and two (n = 29) were plotted for each location across the visual field to form the “visual performance field” (Fig. 3).

    Fig. 3
    figure 3

    Radar plots showing performance as a function of stimulus location in the visual field (a.k.a visual performance field). The shadowed area represents the average True Positive Rate (TPR) over all subjects. Central crosses and outer limits denote chance-level performance (50%) and 100% correct, respectively, and dashed lines indicate a 10% increase in performance (i.e., the closer the black point to the outer limit, the higher the performance for that specific location). Each radar plot corresponds to one difficulty level; plots on the top and bottom row indicate TPR for each stimulus location in Experiments 1 and 2, respectively. Grey numbers around each radar plot indicate the location of each stimulus in degrees. The subset represents one sample stimuli in each sample size

    Full size image
  2. 2.

    PCI was analyzed to demonstrate the relationship between the location of the stimulus in the visual field quadrants and the performance drop due to increased difficulty (see Methods). We completed this analysis on data from Experiment 1 due to the greater number of trials in each set size (one-half of the total trials, as compared with Experiment 2 where each set size contained one-quarter of the total number of trials) and because each letter could be attributed to one visual quadrant (no stimulus on cardinal points). A one-way ANOVA was then performed on PCI from all visual field quadrants as within factor. As is depicted in Fig. 4, performance change showed significant effect of location (F(3,40) = 8.04, p = 0.0003, η2 = 0.37). Tukey’s HSD Test for multiple comparisons found a significant difference between Q1 and both Q3 and Q4 (p-value= 0.0005, 0.95% CI = [-0.39, -0.09] and p-value = 0.0273, 0.95% C.I. = [-0.31, -0.01], respectively), as well as between Q2 with Q3 (p-value = 0.0041, 0.95% CI = [-0.35, -0.20]). This indicates that while in the easier condition stimuli were recognised almost perfectly regardless of their location in the visual field, increasing the stimulus set size had a significantly higher impact on targets presented in the lower half of the visual field compared to the upper half.

    Fig. 4
    figure 4

    Performance change as a function of stimulus location. The violin plot shows the performance change index (PCI – the difference between performance in low and high difficulty conditions) in each visual field quadrant in Experiment 1. Filled dots represent participants, white dots show the median and the horizontal line shows the mean of each group. PCI demonstrates a significant difference between visual field quadrants; such that increasing the difficulty from four letters to eight letters has a significantly higher impact on targets on the lower or left half of the visual field compared with the upper or right, respectively. *p < 0.05

    Full size image
  3. 3.

    Therefore, we removed the effects of superior performances in easier condition by standardizing TPR and subsequently submitted z-score TPRs to a one-way ANOVA to compare visual quadrants. The results showed a significant main effect of visual quadrants (F(3,84) = 17.85, p < 0.0000, η2 = 0.59). Moreover, post hoc t-test comparisons revealed significant differences between both visual quadrants on the upper compared with either of the two lower quadrants (Q1-Q3: T(28) = 5.93, p-value = 0.0000134, Q1-Q4: T(28) = 4.93, p = 0.0002, Q2-Q3: T(28) = 3.85, p = 0.003781, and Q2-Q4: T(28) = 3.07, p = 0.028013) (Fig. 5).

    Fig. 5
    figure 5

    Standardized performance across visual quadrants. Bar plots show average standardized True Positive Rates (TPRs) over all participants across visual field quadrants from both experiments. The performance in visual quadrants is standardized for each difficulty level. Standardized TPR in upper visual quadrants was significantly higher than either quadrant on lower hemifield. *p < 0.05

    Full size image

    Additionally, we ran a GLM on each subject’s data separately. All the GLMs could explain a significant amount of variance in the outcome variable (see supplementary information for statistical measures). Moreover, using Q1 as the reference predictor variable, we showed that the beta estimates of the three other quadrants were significantly negative in most of the cases, i.e., performance increases as the location of stimulus changes from Q1 to Q4 (OSM Fig. 2).

  4. 4.

    To express the association between accuracy and the location of the stimulus in the Cartesian coordinate system, we fitted the standardized TPRs to a Fourier series (Fig. 6). Because of our standardization approach, we were able to treat all the locations similarly in the cartesian plane regardless of the set size (i.e., difficulty) each stimulus comes from.

    Fig. 6
    figure 6

    Scatter plot showing accuracy versus letter positions. Points are the average of all subjects’ standardized True Positive Rates (TPRs) overall difficulty levels in both experiments. Letters’ positions are illustrated with their degree in Cartesian Plane. Data are fitted with a one-term Fourier model. The visual field displayed variable characteristics in iso-eccentric locations. As the stimulus moves further from zero-degree, performance increases and reaches the peak at ~60°; recognition performance then decreases until it reaches the trough at ~260°

    Full size image

The goodness of fit measures (adjusted r-squared = 0.67 and RMSE = 0.18) confirmed that variation in recognition performance across the visual field could be explained by a Fourier model; such that the performance increased as the location of the stimulus moved further from 0° until it reached a peak at ~60°. Then the performance gradually dropped from ~60° to ~260° where it reached a trough and started rising again.

In summary, in both experiments, the location of the stimulus in the visual field influenced the task performance, such that stimuli on the upper right and left visual quadrants were detected better than the stimuli on the lower visual quadrants.

Discussion

In this study, we investigated two key aspects of visual stimulus processing on recognition performance, namely: temporal attention and spatial location across the visual field. We used a visual recognition task with stimuli presented at different locations across the visual field. Colored disks were used to cue the variable temporal onset of the upcoming stimulus-set. Our study revealed four main findings: (1) Cuing the temporal onset of the upcoming stimulus improved performance. (2) We described a performance field in the milieu of visual working memory and showed that stimuli in both upper visual quadrants (Q1 and Q2) yield significantly better outcomes (True Positive Rate or TPR) than stimuli in the lower visual quadrants (Q3 and Q4). (3) The performance change for stimuli presented on the lower half of the visual field was more prominent for more difficult tasks. (4) Moreover, when the impact of difficulty in each individual was removed by standardizing the participants’ performance in all difficulty levels the spatial location bias could be explained by a model with one peak (at ~60°) and one trough (at ~260°). In the following sections, we discuss the main findings of this study.

Temporal attention enhances visual recognition performance regardless of the location of the stimulus

Using a colored cue, we tested the effect of temporally orienting attention to the onset of to-be-remembered stimuli. We depicted that participants’ accuracy and inverse efficiency score (IES) was enhanced in predictable compared with the unpredictable condition.

These findings were most evident in the more difficult condition that yielded around 60–80% accuracy and are consistent with previous findings that showed participants with mid-to-high performances had superior accuracies in predictable compared with unpredictable trials, but not the participants with near perfect or very low performances (Samaha et al., 2015). This may be due to the fact that when the task is easy and participants have near-ceiling performances, deploying attentional resources won’t be necessary as compared with the conditions where participants struggle to respond correctly to the probe due to the increased number of stimuli that should be kept in working memory (Kraft et al., 2007).

Furthermore, similar to other experiments on temporal attention (Cravo et al., 2017; Wilsch et al., 2018) there was a high inter-subject variability, and therefore low effect sizes in our data; nevertheless, we could show that orienting attention to the onset of the to-be-remembered stimuli improves both accuracy and IES. Previous studies have established the enhancing effect of temporal attention on many cognitive tasks such as orientation detection (Rohenkohl et al., 2014; Rohenkohl & Nobre, 2011), long-term (Cravo et al., 2017), and short-term memory (van Ede et al., 2016; Wilsch et al., 2018), but this is to our knowledge the first to focus on temporal attention in a visual working memory task with stimuli distributed iso-eccentrically around the visual field. Our findings are also consistent with the results of studies that measured the impact of attention (both spatial and temporal) on visual performance fields and demonstrated that attention improves performance homogenously around the visual field and does not depend on the location of the stimulus (Donovan & Carrasco, 2018; Fernández et al., 2019; Purokayastha et al., 2021). The behavioral benefits of temporal attention on working memory could be addressed in future research by systematically manipulating the delay between stimulus-set and probe; additionally, a higher number of participants might make up for the inter-subject variability and result in larger effect sizes.

Performance bias across the visual field depends on the task and stimulus properties

Parallel to our findings on the effect of temporal orienting of attention, we demonstrated a visual performance field with enhanced recognition performance on both upper visual quadrants compared with lower-left and lower-right. While a large body of literature corroborates performance asymmetry around the visual field, their results are not necessarily consistent and are dependent on the cognitive task and visual properties of the stimulus. Current study is one of the firsts addressing location-dependence of working memory performance for stimuli with mid-level complexity which were presented simultaneously and were all, therefore, potential targets.

Studies addressing simpler cognitive processes such as contrast sensitivity (Abrams et al., 2012; Cameron et al., 2002; Carrasco et al., 2001; Corbett & Carrasco, 2011; Robson & Graham, 1981), motion perception (Zito et al., 2016), memory tasks with one-feature objects (S. L. Sheremata et al., 2010), and global processing (S. D. Christman & Niebauer, 1997; Efron et al., 1987, 1990b, 1990a; Previc, 1990; Previc & Blume, 1993), have either shown a lower or left visual field advantage, whereas an upper or right visual field superiority was found in studies that focused on more complex perceptual dimensions such as visual search (S. D. Christman & Niebauer, 1997; Efron et al., 1987, 1990b, 1990a; Pflugshaupt et al., 2009; Previc, 1990; Previc & Blume, 1993; Previc & Naegele, 2001), Uncrossed (far) disparity detection (Previc, 1990), local and categorical processing (S. D. Christman & Niebauer, 1997; Navon, 1977; Previc, 1990), object recognition (K. W. Chambers et al., 1999; S. D. Christman & Niebauer, 1997; Felisberti & McDermott, 2013; Palanica & Itier, 2017; Quek & Finkbeiner, 2014b), memory task for two-feature objects (S. L. Sheremata et al., 2010), and face processing (Felisberti & McDermott, 2013; Kessler & Tipper, 2004; Quek & Finkbeiner, 2014a; Schmidtmann et al., 2015).

Additionally, it is well established that the type of stimulus is another key factor affecting the visual performance fields; such that low-level stimuli (e.g., Gabor patches (Carrasco et al., 2001, 2002; Grigorescu et al., 2004)) and stimuli with low temporal or spatial frequency information (S. Christman, 1989; Efron et al., 1987, 1990b, 1990a; Previc & Blume, 1993) show enhanced performance in either lower or left visual hemifield. However, higher-level stimuli (e.g., faces (Carlei et al., 2017; Quek & Finkbeiner, 2016; Schmidtmann et al., 2015), complex shapes (Zito et al., 2016), and stimuli with high spatial and temporal frequency information (S. D. Christman & Niebauer, 1997; Previc, 1990; Sergent & Holzer, 2013)) are better detected in either upper or right visual quadrant. More related to our project, studies that used letters mostly resulted in better performances on the right or upper visual hemifields (Chelazzi et al., 2014; Hagenbeek & Van Strien, 2002; Hirata & Bryden, 1976; Rutkowski et al., 2002). It is important to note that although letters are fundamental components of the written language, they are not processed in language-related areas of the brain as shown in (Hagenbeek & Van Strien, 2002; Hirata & Bryden, 1976; Young et al., 1984) and are processed as mid-level visual rather than language-related stimuli (S. Christman, 1989). Therefore, it is clear that not only does the cognitive function that is engaged in the task affect the performance asymmetries around the visual field, but also the type of stimulus per se is a crucial factor.

Stimuli on the upper and right visual fields have better access to perceptual processes

The performance bias we found in our study could be attributed to the enhanced attentional processes that were then carried over to working memory (Ester et al., 2009; Serences et al., 2009; see also Tsubomi et al., 2013). Since working memory is a storage with limited capacity, stimuli are degraded within working memory relative to their perceptual priority (Bays et al., 2009; Fougnie et al., 2010; Van Den Berg et al., 2012; Zhang & Luck, 2008). Those with higher behavioral relevance or inherent attentional bias will be selected by our perceptual system and gated into working memory.

One other unique aspect of our study was that unlike most studies investigating visual performance fields (Barbot et al., 2021; Chelazzi et al., 2014; Montaser-Kouhsari & Carrasco, 2009), we presented all stimuli simultaneously to see which locations grab more attention and therefore result in better recognition performance when competing for attentional resources. A vast body of literature demonstrates that attention has a limited accessible size (He et al., 1996); for a stimulus to “pop out” or become consciously available, it must be processed more efficiently and win the competition between stimuli (Robert Desimone, 1998; Egeth & Yantis, 1997; Frith, n.d.; Lamme, 2004).

We showed that in a competition where stimuli in all visual quadrants are potential targets, the upper and right visual quadrants are better remembered. This notion is also supported by our performance change analysis that showed by increasing the number of items to remember, performance decreases more drastically in locations with lower initial performance, i.e., in left and lower visual quadrants. In other words, when the difficulty and therefore, the competition for attentional resources increases, the stimuli on the lower half of the visual field and more so on the left lower quadrant lose performance more drastically than the ones on the upper or right quadrants. This performance drop in lower visual hemifield continues to the point of chance-level and only increasing the set size above that point will result in decreased performance in upper visual hemifield. The concept of pronounced inhomogeneity by increasing the difficulty has been shown in other studies (Cameron et al., 2002) and the contrast between our findings and theirs can be attributed to the reasons explained above (task and stimulus difference).

Manipulation of set size not only increases the load of memory but also results in perceptual or working memory crowding. Tamber-Rosenau et al. (2015) demonstrated that although visual working memory has a spatial resolution comparable to that of perceptual processes, they employ distinct mechanisms. Also, the relative location of stimuli might result in different crowding effects in perception and working memory (Yörük et al., 2020). Here, to diminish the effect of difficulty as well as crowding in higher set sizes we standardized each participant’s performance in each difficulty level.

This allowed us to generate a more general description of the performance field around the centre of gaze and plot all 28 locations (four difficulty levels in two experiments) together in the Cartesian plane to find the locations with the highest (60°) and lowest (260°) accuracies. Many studies have investigated performance bias around the visual field so far, most of which only focused on the visual quadrants, meridians, or hemifields (Abrams et al., 2012; Carrasco et al., 2001; Montaser-Kouhsari & Carrasco, 2009; S. Sheremata & Shomstein, 2014). A few studies examined higher number of locations around the visual field (Barbot et al., 2021); however, to our knowledge, this is the first study that provides a more general description of performance the around visual field. This analysis, in turn, invigorates our previous findings on the biased distribution of attentional resources around the visual field.

Asymmetric distribution of spatial attention and functional distinction between visual hemifields can explain enhanced recognition performance for right and upper stimuli

As discussed earlier, many perceptual processes (Hellige et al., 2010) as well as cognitive functions (Hugdahl & Westerhausen, 2010) are biased towards one visual hemifield or quadrant. The upper and right superiority we detected in our study can be explained by the hemispheric structural organization as well as functional dissociations.

Studies on patients with hemineglect (Heilman & Valenstein, 1979; Mesulam, 1981) as well as on healthy participants in cognitively loaded conditions (Takio et al., 2014) have supported the proposal of “hemispatial” theory (Heilman & Van Den Abell, 1980). This theory suggests that attentional mechanisms in the cortical hemispheres are not symmetric. While the right hemisphere directs attention to both contralateral and ipsilateral visual hemifield, the left hemisphere only directs attention to the right hemifield (Heilman & Van Den Abell, 1980; Reuter-Lorenz et al., 1990). This hemispheric asymmetry has also been confirmed in neuroimaging studies that exhibited more contralateral bias in the left attentional network compared with the right (S. L. Sheremata et al., 2010; Szczepanski et al., 2010; Szczepanski & Kastner, 2013). Therefore, the enhanced memory performance in the right visual hemifield can be due to the endogenous rightward bias in the organization of spatial attention that even precedes the appearance of the visual stimulus (Loughnane et al., 2015).

In addition to the hemispheric asymmetries, functional distinction between visual hemifields can be attributed to ecological constraints. As Previc proposed (Previc, 1990), lower visual hemifield represents peri-personal (near) space and therefore is specialized for global processing of images in near vision, whereas extra-personal (far) space is mainly represented in the upper visual field (see Danckert & Goodale, 2003). This dissociation might be the reason why we observed improved performance in encoding visual stimuli, which is normally executed in the extra-personal space, for those that were presented in the upper hemifield.

Accordingly, not only is more cortical area allocated to processing the upper visual field for parafoveal visual stimuli (Benson et al., 2021), but also hemineglect research enforces the idea that in addition to the right visual field bias, we might be more attentive to the stimuli in upper rather than lower visual hemifield to constantly stay aware of our extra-personal space. For instance, several spatial attention tasks illustrated an upward bias, such that the left lower visual quadrant (Q4) yielded the lowest performance measures in these patients (Aimola et al., 2012; Cappelletti et al., 2007; Cazzoli et al., 2011; Halligan & Marshall, 1989; Pitzalis et al., 1997).

The recognition asymmetry we describe here could arise from an inherent (and therefore obligatory) bias in attentional resources towards targets with increasing number of stimuli. Conversely, it could be a result of the participant’s strategic planning to retain as much information as possible in a fading working memory storage (Sperling, 1960) before the probe appears. In this study we did not aim to distinguish between these two phenomena that are very close in nature; however, the extent to which each of these two plays a role could be addressed in a future study using differently weighted targets around the visual field (Hu et al., 2014, 2016).

Conclusion

In this study, we demonstrated that temporal orienting of attention to the onset of the to-be-remembered stimuli improves performance, and this effect is consistent for all locations around the visual field. We also showed that in a working memory task, stimuli that are presented on upper and right visual quadrants yield better performance compared with those on lower or left visual quadrants. This performance difference was demonstrated in a task where all the stimuli were presented simultaneously and hence were all potential targets.

Our findings reveal that working memory, like many other cognitive functions such as perception and stimulus discrimination, is impacted by the location of stimulus around the visual field and is advantaged by temporal orienting of attention. These results are consistent with previous findings and extend our understanding of temporal attention and performance asymmetry to visual recognition.