Statistical learning in visual search: ‘contextual cueing’ reflects the acquisition of an optimal, ‘one-for-all’ oculomotor scanning strategy

doi:10.21203/rs.3.rs-2561645/v1

Download PDF

Article

Statistical learning in visual search: ‘contextual cueing’ reflects the acquisition of an optimal, ‘one-for-all’ oculomotor scanning strategy

https://doi.org/10.21203/rs.3.rs-2561645/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 20 Sep, 2023

Read the published version in Communications Psychology →

Version 1

posted

You are reading this latest preprint version

Detecting a target in visual search becomes more efficient over time when it is encountered repeatedly at a fixed location within a stable distractor arrangement (spatial ‘context’), compared to non-repeated contexts. The standard account attributes this contextual-cueing effect to the acquisition of display-specific long-term memories, which, when activated by the current display, ‘cue’ attention to the target location. Our alternative, ‘procedural-optimization’ account posits that contextual facilitation arises from the acquisition of generic oculomotor scanning strategies that are optimized with respect to the entire set of displays, with frequently searched displays accruing greater weight in the optimization. To decide between these alternatives, we examined novel measures of the similarity, across time-on-task, of the spatio-temporal sequences of fixations through repeated and non-repeated displays. We found scanpath similarity to increase generally with learning, but more for repeated versus non-repeated displays. This pattern contradicts display-specific guidance, but supports ‘one-for-all’ scanpath optimization.

Contextual Cueing

Selective Attention

Scanpath Theory

Visual Search

Procedural Learning

Statistical Learning

Training-related improvements of behavioral and cortical responses are a hallmark feature of almost every sensorimotor and cognitive task. To achieve this successfully, one hypothesis is that training leads to the build of stimulus-specific memories that aid performance when re-encountering these stimuli in future occasions. Conversely, experience with specific stimuli may also train rather abstract task procedures involved in the processing of these items. Here, we use novel eyetracking scanpath analysis to determine how stimulus experience unfolds using a well-established paradigm in cognitive neuroscience: visual search in repeated versus non-repeated target-distractor arrays. Our new data show that repeated arrays lead to reduced reaction times and fewer fixation numbers in detecting/ localizing the critical search-for target item. Critically, we show and quantify that these performance improvements are brought about by procedural training of task-general, visual search operations. Thus, training-related improvements in search tasks seem to depend on changes in task-general procedural memories and unspecific to individual displays.

Visual search for a target object among non-target, or distractor, objects can be facilitated by prior knowledge of the scene, including contextual long-term memory of co-occurring objects or the position of the target relative to an invariant (spatial) distractor arrangement (e.g., Chun & Jiang, 1998). Chun and Jiang, in their seminal study, presented participants with search arrays containing a target letter “T” amongst a set of distractor letters “L” (a relatively hard search task that affords little bottom-up or top-down guidance; cf. Moran et al., 2016). Critically, in half of the trials, the spatial arrangements of the distractor and target stimuli were repeated (i.e., repeated, ‘old’ contexts); in the other half, the distractor locations were generated anew on each trial, while keeping the target position constant (i.e., non-repeated, ‘new’ contexts). Thus, given that the absolute target positions were fixed in both types of trial, the only difference between them was whether or not the target location was predictable from the distractor context. The two sets of repeated and non-repeated displays were presented randomly intermixed within each block of trials. Chun and Jiang (1998) observed that the search reaction times (RTs) decreased with increasing block number for non-repeated as well as repeated arrays, attributed to general procedural learning of how to perform the task. Critically, however, this practice-dependent improvement was larger for the repeated arrays – an effect Chun and Jiang (1998) referred to as ‘contextual cueing’. Search RT facilitation by repeated contexts typically emerges rapidly, after just a few (2–4) encounters of the same visual arrangement (e.g., Geyer et al., 2010), and it appears to be implicit (and ‘automatic’) in nature ( Chun & Jiang, 2003). Further, in terms of oculomotor search performance, eye-tracking studies indicate that contextual learning leads to a reduction in the number of fixations required to reach the target in repeated, compared to non-repeated, displays (e.g., Peterson & Kramer, 2001; Tseng & Li, 2004; Manginelli & Pollmann, 2009; Kröll, Schlagbauer, Zinchenko, Müller, & Geyer, 2019; Harris & Remington, 2017).

One intriguing, and, as we will argue below, open question is how these savings in the number of fixations are actually produced in repeated search displays. The standard account attributes these savings to search being ‘cued’, or guided, more directly to the target location as a result of having acquired a (long-term) associative memory representation, or ‘template’, of a specific distractor-target arrangement. This template is activated upon re-encountering such an arrangement on a given trial, which then top-down increases the attentional priority of the target location (e.g., Chun & Jiang, 1998; Brady & Chun, 2007; Beesley et al., 2016) – thus enhancing the target’s potential to summon covert or overt attention. According to this account, the number of attention shifts required to detect a target in a repeated search array will decrease with increasing (re-)encounters of this array, due to the build-up of a search-guiding contextual memory template for this array (e.g., Chun & Jiang, 1998). Support for this comes from studies of contextual cueing that used fixation number as a dependent measure (e.g., Peterson & Kramer, 2001; Tseng & Li, 2004). These studies showed that finding the target in repeated arrays requires overall fewer eye movements – though with the ‘guidance’ effect emerging only after the first few fixations, suggesting that it may take some time for the template to come into play. That is, in Tseng and Li’s (2004) terms, search may involve some ‘inefficient’, unguided scanning of the array until an informative constellation of distractors represented in the template is encountered. Activation of the template would then lead an ‘effective’, guided search phase: a relatively direct ‘homing in’ of attention on the target location after a series of more exploratory fixations. Accordingly, the savings in the number of eye movements for repeated (vs. non-repeated) displays would arise from later fixations in the saccadic scanpaths – perhaps with the template-based priority signal pointing to the target location growing increasingly stronger as oculomotor scanning approaches the target item (Chun & Jiang, 1998). This ‘specific-template’ account is attractive, not least because it ties in seamlessly with the functional architecture assumed by general theories of search guidance, such as Guided Search (e.g., Wolfe et al., 1989; Wolfe, 1994; Wolfe, 2021).

However, there may be an alternative, more ‘procedural’ account of contextual facilitation that does not rely on the notion that observers acquire memory representations that are specific to particular distractor-target arrangements – a conceptually new account that the present study set out to explore. In fact, procedural learning in contextual-cueing paradigms is a universal finding in virtually all pertinent studies (for reviews see, e.g., Goujon et al., 2015; Sisk et al., 2019): search speed improves, typically quite substantially, over the course of practice on the task, that is, across trial blocks (Chun & Jiang, 2003). Importantly, an improvement is evident for non-repeated – as well as repeated – displays, which is generally attributed to procedural learning, which optimizes, or ‘automatizes’, performance through the development/refinement of a task-appropriate (search) settings, akin to the development of a ‘skill’ (e.g., Fitts, 1964; Fitts & Posner, 1967; Shiffrin & Schneider, 1977; Anderson, 1982). Critically, though, the improvement is more marked for repeated (vs. non-repeated) displays, which constitutes the contextual-facilitation effect. The standard account attributes this effect to the acquisition of ‘specific’ search-guiding templates for repeated displays. However, rather than reflecting a qualitative difference in how repeated and non-repeated displays are processed, the alternative account conceives of contextual facilitation as arising from a ‘generic’ procedural optimization of search, in which repeated displays, simply by virtue of being repeatedly encountered, accrue greater weight in shaping the learning-dependent adjustments of the (skill-related parameters of) search performance; non-repeated displays, by contrast, would have only little individual influence because, by definition, they are encountered only once. Thus, in oculomotor terms, what is optimized in procedural task learning may be a search strategy which is increasingly generic in the sense that it is applicable to all search displays, repeated and non-repeated (rather than being specific for particular repeated displays). However, as a result of statistical learning, this strategy is more ‘tuned’ to, and so more effective for, those displays that are encountered frequently, rather than displays searched only once.

This presents a fundamentally different conceptualization of contextual facilitation of visual search: It replaces the dual-mechanism notion that, on top of some general practice effects, observers develop high-fidelity search-guiding memory templates for specific, repeated displays by a single mechanism of procedural learning – involving a generic (i.e., display-unspecific) optimization of oculomotor scanning in the sense of skill acquisition (Fitts, 1964) – explaining both the overall decrease in search RTs and number of fixations over time on task and the extra gains in RT and fixation number for repeated over non-repeated displays. The latter come about merely due to the increased number of exposures to repeated displays, which collectively are weighted more strongly in the development of the generic search strategy.

To decide between these alternative accounts, we set out to, first of all, establish (and thus replicate) contextual facilitation in terms of the standard summary RT and eye-movement measures (such as the average number of saccades/ fixations) that have informed theorizing in the extant contextual-cueing literature. Then, in order to investigate (i.e, to either corroborate or refute) the alternative account, we went on to examine novel oculomotor-scanpath–similarity measures – in particular, Dynamic Time Warping, Discrete Fréchet Distance, and Area Between Curves – that are diagnostic of similarity in the spatio-temporal sequence of fixations across individual (repeated and non-repeated) displays, as well as the sequences produced by individual participants. These analyses were designed to reveal detailed information about the proceduralization of search performance, which is lost in the ‘standard’ averaging of dependent measures (such as RT and fixation number) both across individual repeated and, respectively, non-repeated displays and across individual observers.

According to the procedural-optimization hypothesis, (1) scanpaths should become more homogeneous for individual displays across participants over trial blocks, with scanpaths for repeated displays becoming more similar compared to those for non-repeated displays; and (2) an increase in similarity with time on task should also be observable when scanpaths for different displays are compared within individual participants: similarity measures should be higher for pairs of (differently composed) repeated displays compared to pairs of (different) non-repeated displays. In contrast, the display-specific hypothesis of CC would predict that scanpaths become more dissimilar for pairs of repeated (relative to non-repeated) displays.

General approach

Our goal was to bring together established RT and oculomotor measures of the contextual-cueing effect, which focus on group mean values, with new, oculomotor-scanpath measures that quantitatively describe search behavior in a more fine-grained manner, in particular, at the levels of individual displays or individual participants (see Figure1A and B). Additionally, by replicating established measures from the literature (e.g., Peterson & Kramer, 2001; Tseng & Li, 2004; Brockmole & Henderson, 2006; Myers & Gray, 2010), we aimed to ensure the representativeness of our own data for contextual-cueing studies at large, thus increasing the confidence in the generality of our novel analyses and findings.

For our novel analysis approach to be feasible, we adjusted the experimental design in two respects: First, and motivated by a previous study of contextual cueing (Sewell et al., 2018), we reduced the number of learnable, repeated target locations, as well as the number of target locations in non-learnable, non-repeated displays to four each, with one target location per display quadrant; this was meant to ensure that the memory signals for the respective target location and the corresponding (possibly display-specific) scanpath would have as little interference from other repeated displays as possible and that allocation of attention over space and time would be maximally different. Second, we presented the same repeated and non-repeated display arrangements to all participants, in the same trial order. Using the same display arrangements allowed us to control the perceptual content of the display set throughout the experiment; in particular, using the same arrangements for non-repeated displays ensured a “fair” comparison between scanpaths, eliminating confounds originating from, across participants, variably composed distractor-target configurations in non-repeated displays. Methodologically, these adjustments made it possible to compare pairs of scanpaths at different levels and relating to (1) the similarity of fixation sequences through an individual display when viewed by pairs of different participants and (2) the similarity of scanpaths for an individual participant viewing (pairs of) different displays. These design measures enabled us to perform a thorough test of the contrasting predictions made by the specific and the generic procedural-optimization accounts.

Participants

The sample size was determined based on previous CC studies using relatively large sample sizes (e.g., Vadillo, Malejka, Lee, Dienes & Shanks, 2021; Peterson, Mead, Kelly, Esser-Adomako & Blumberg, 2022) in order to obtain reasonably stable estimates of contextual cueing. High statistical power was necessary since we analyzed contextual cueing at the level of individual displays. We recruited 46 participants for the present experiment (38 females, 3 left-handed, mean age = 23.28 [SD = 5.62, range = 19–43] years).

Apparatus and stimuli

The experimental routine was programmed in Matlab with Psychtoolbox extensions (Brainard, 1997; Pelli, 1997) and run on an Intel PC under the Windows-7 operating system. Participants were seated in a dimly lit laboratory booth in front of a 19-inch CRT monitor (AOC, Amsterdam; display resolution 1024 x 768 pixels; refresh rate: 85 Hz) at a viewing distance of 60 cm (controlled by a chin rest). The search displays consisted of 12 gray items (luminance: 1.0 cd/m²; 1 target and 11 distractors) presented against a black background (0.11 cd/m²). All stimuli extended 0.35° of visual angle in both width and height. As depicted in Figure 1C, the items were arranged on three (invisible) concentric circles around the display center (with radii of 1.74°, 3.48°, and 5.22° for circles 1, 2, and 3, respectively). In repeated displays, the locations and orientations of the distractors were held constant across trials; in non-repeated displays, all distractors (i.e., their locations and orientations) were generated anew on each trial. Note that in all presented displays, the location of the target was repeated but the (left/right) orientation of the target was determined randomly and was, thus, unpredictable. As a result, a repeated context could only be associated with a specific (repeated) target location, but not with a specific target identity. Following Chun and Jiang (1998), this approach is used in most CC studies to ensure that contextual facilitation of RTs is owing to the repeated context guiding attention/ the eyes, rather than facilitating the selection of the manual response (invariably) associated with a given repeated display. Importantly, both the set of (N=4) repeated displays and the set of (N=128) randomly generated non-repeated displays were kept constant across all 46 participants, so that each participant encountered identical repeated and non-repeated configurations. Note, however, that trial order was randomly chosen within each block of N=4 repeated plus N=4 non-repeated trials for individual participants. This enabled us to keep low-level, individual display properties constant across participants and thus compute dependent – scanpath – measures for each individual display (with variations between participants providing the error term).

There were overall 8 possible target locations, 4 of which were used for repeated displays (with constant distractor layouts) and the other 4 for non-repeated displays (with random distractor arrangements). All targets, in both repeated and non-repeated displays, were located on the second ring, controlling for the distance of the target from to the display center in all conditions. Furthermore, the targets were placed in all four quadrants with equal probability. Importantly, participants were not informed about the fact that some of the search arrays were presented repeatedly. The “T” target was rotated randomly by 90° to either the left or the right. The 11 remaining items were L-shaped distractors rotated randomly at orthogonal orientations (0°, 90°, 180°, or 270°). Figure 1C presents example display layouts (an illustration can be seen in Appendix III, Figure1).

To monitor and record eye movements, a video-based eye-tracker was used (EyeLink 1000; SR Research Ltd., Mississauga, Ontario, Canada; version 4.594). Eye-movement recordings were calibrated at the start of the experiment and after every 64 trials. Calibration was considered accurate when fixation positions fell within ~1° for all calibration points. The default psychophysical sample configuration of the eye-tracking system (i.e., saccade velocity threshold set at 35°/s, saccade acceleration threshold set at 9500°/s²) was adopted for identifying saccadic eye movements.

Trial sequence

A trial started with the presentation of a central fixation cross (0.10° x 0.10°, luminance: 1.0 cd/m2) for 500 ms. Next, the fixation cross was removed from the screen, and, following a blank interval of 200 ms, the search display was presented. Observers were instructed to find the target “T” and respond as quickly and accurately as possible to its (left vs. right) orientation, while being allowed to move their eyes freely. Each search display stayed on the screen until a manual response was elicited. If the “T” was rotated to the right (left), observers responded by pressing the right (left) arrow button on a computer keyboard with their right (left) index finger. Following a response error, the word “Wrong” appeared in the screen center for 1000 ms. Each trial was followed by a blank inter-trial interval of 1000 ms. The experiment consisted of 256 trials (32 blocks x 8 trials each, 50% repeated displays in each block). Participants were free to proceed to the next block at their own pace. The search task took some 30 minutes to complete.

Recognition test

At the end of the experiment, observers performed a yes/no (repeated/non-repeated display) recognition test, permitting us to assess whether they had acquired any explicit memory of repeated configurations presented in the preceding search task (a standard procedure in contextual-cueing experiments; see, e.g., Chun & Jiang, 1998). To this end, observers were presented with 4 repeated displays and 4 newly composed displays. The task was to indicate whether or not a given display had been shown previously, by pressing the left or the right mouse button, respectively. The 4 repeated and the 4 newly generated displays were presented in random order. Observers’ responses in the recognition task were non-speeded and no error feedback was provided.

Statistical analysis

Comparisons of scanpaths were carried out in Python (Van Rossum & Drake, 1995). Statistical analysis was performed using R (version 3.4.3; R Core Team, 2018). Because the same set of non-/repeated displays was used across participants with evenly distributed target positions across the four display quadrants, we analyzed our dependent measures (reaction times, error rates, oculomotor variables) using the lme4 package in R for mixed-effect modeling, to account for these dependencies by including target quadrant and participant as random factors in addition to the fixed factors of block and context.

Dependent measures

To establish comparability between the present investigation and previous contextual-cueing studies, as well as to validate our dataset as being representative for visual-search paradigms, we begin our analysis with an examination and replication of established measures of the contextual-facilitation effect: RT, fixation number, and saccade amplitude (see, e.g., Tseng & Li, 2004, or Peterson & Kramer, 2001). We then proceed to the presentation of scanpath measures of contextual facilitation, including the total length of the scanpaths (Brockmole & Henderson, 2006; Pollmann & Manginelli, 2009) and the standard deviation of the lengths of the saccades constituting each scanpath. The latter essentially provides a new measure of the variability of saccade lengths across individual observers and displays, where a decrease in variability can be considered a measure of automaticity (Logan, 1988). This is followed by new overlay-plot visualizations of individual participants’ spatiotemporal sequence of oculomotor behavior (which are also meant to demonstrate the usefulness of our scanpath approach to eye-tracking investigations of visual search in general). From these visualizations, a new quantitative measure of contextual cueing, namely: scanpath similarity or consistency, is derived.

Reaction times.

For the RT analyses, error trials and ‘extreme’ RTs three standard deviations below and above the mean were excluded from the data. This outlier criterion led to the removal of ∼3% of all trials. Individual observers’ mean RTs, and associated error rates, were calculated per experimental condition and submitted to a 2 × 32 repeated-measures linear model analysis of variance (ANOVA) with Context (repeated, non-repeated) and Block (1-32) as fixed factors and Participant (1-46) and Target Quadrant (upper-left, upper-right, lower-left, lower-right) as random factors.

Analysis of scanpath similarity

The similarity of scanpaths was computed using established measures in the field (cf. Fahimi & Bruce, 2020; Jekel et al., 2019), in particular, Dynamic Time Warping, Discrete Fréchet Distance, and Area Between Curves.

Dynamic Time Warping is a measure of similarity between two fixational series of different lengths. Two individual scanpaths may be highly similar with regard to the placing (i.e., the spatial coordinates) of individual fixations, but the temporal alignment of these sequences may be less consistent across individual trials. The strength of Dynamic Time Warping is that it can quantify the similarity of the shapes of scanpaths with distinct time series. Specifically, this metric compares two fixational series by aligning them in the time domain, thus minimizing the Euclidean distance between the aligned series. Concerning the Discrete Fréchet Distance: this metric can also deal with fixational time series of different length (and tempo). The Fréchet Distance considers both the location and ordering of the individual fixations along two scanpath curves and can be defined with regard to an analogy: a person that is walking a dog on a leash, with the person walking on one (scan-) path/curve and the dog on another path/curve. The discrete version of the Fréchet Distance only compares distances between fixations and not points in between. Hence, the Discrete Fréchet Distance corresponds to the length of the shortest leash possible for traversing both curves. Finally, we computed scanpath similarity based on Area Between Curves, which, like Dynamic Time Warping and the Discrete Fréchet Distance, permits comparisons of scanpaths with different lengths, although the particular scanpath measure is based on the area that falls between two scanpath curves. As Area Between Curves is well-suited to quantify hysteresis (Jekel et al., 2019), this measure should be particularly sensitive to capture scanpath similarity when trajectories have the same start and end points (initial fixation point and target location).

We chose to explore three scanpath metrics, rather than just one, in order to to provide a maximally precise and unbiased measurement of the effects of search task training on oculomotor behavior (assuming that not all scanpath metrics capture the underlying effects equally; cf. Fahimi & Bruce, 2020). Our specific trial schedule (see Figure 1A and B) allowed us to examine the similarity of scanpaths in multiple ways (for further details, see Appendix I). First, we compared each possible pair of gaze patterns arising from identical displays over different participants. This approach enabled us to compute scanpath similarity for each experimental block (each consisting of 4 repeated and 4 non-repeated arrays), thereby addressing the important question of how the consistency of viewing patterns changes as a function of practice on the task. Second, we computed similarity of oculomotor trajectories between each pair of different displays viewed by the same participant. This analysis was intended as a strong test of the display-specific vs. general-procedural accounts of contextual repetitions on search-task training.

To more formally examine whether these observations represent meaningful effects, we computed scanpath similarity for each experimental block (1-32). To recap our hypothesis: if participants are learning a generic search procedure that is increasingly effective, then similarity of scanpaths should increase over time for both repeated and non-repeated arrays, though this effect should be more pronounced for the former displays which, due to being repeated, accrue a greater weight in shaping the generic search procedure. But the prediction would be fundamentally different for the display-specific hypothesis of contextual cueing, according to which experience with individual repeated displays leads to the build-up of display-specific memories and associated scanning behavior. Accordingly, scanpath similarity obtained for pairs of individual repeated displays with different spatial composition should decrease with increased search-task training (and similarity measures should effectively be lower than those for non-repeated displays). Two analyses were conducted (see Figure 1A and B; also Appendix I). In the first, within-display analysis, similarity of eye-movement sequences was calculated from each pair of different participants when viewing a given, individual (repeated or non-repeated) display. Second, in the within-participant analysis, similarity measures were generated from eye-movement sequences in pairs of different displays when searched by an individual participant. Both analyses were conducted for three similarity measures: Dynamic Time Warping, Discrete Fréchet Distance, and Area Between Curves. Statistical inference was based on mixed-effect ANOVAs with the fixed factors of Context and Block and the random factors of Target Quadrant (in the analysis of within-display similarity) and Participant (within-participant scanpath analysis).

To preview our results: (1) we replicate previous findings of more efficient visual search in terms of expedited RTs and reduced fixation number (and other established oculomotor measures) for repeated vs. non-repeated search arrays. (2) Individual participants exhibit scanpath patterns that are increasingly similar across blocks of repeated and non-repeated displays – both within displays and within participants. Repeated displays nevertheless exhibit a higher consistency within displays and within participants, reflected in a higher degree of scanpath similarity across repeated displays.

Reaction times.

Overall, observers had an average error rate of ∼1.5%, without any indication of a speed-accuracy trade-off. Regarding error rates, no main or interaction effects involving the factor Context reached significance (all p’s > 0.05).

The analysis of the mean RTs revealed a main effect of Context: participants responded faster to repeated relative to non-repeated displays (1358.58 vs. 1574.28 ms; F(1, 8573.6) = 283.99, p < p < 2*10^-16). The main effect of Block was also significant, reflecting a decrease in RTs with increasing block number (block 1 = 1960.32 ms, block 32 = 1345.46 ms; F(31, 8579.4) = 13.67, p < 2*10^-16). The Context × Block interaction was non-significant (F(31,8579.7) = 0.88, p = 0.66), indicative of a stable contextual-facilitation effect across blocks (cf. Fig. 1E).

Recognition performance

Participants’ mean hit rate (repeated displays correctly identified as repeated) was 52.5%, which compares with a mean false-alarm rate (non-repeated display incorrectly judged as repeated) of 47.5%, t_pairwise(45) = 0.88, p = 0.38. Thus, there was little indication of explicit, above-chance recognition of displays that had been encountered repeatedly over the course of the search task.

Oculomotor performance

Figure 1F presents a series of oculomotor measures, derived from fixations and saccades, as a function of block number aggregated over trials, separately for repeated and non-repeated displays; Table 1 summarizes the respective descriptive and inference statistics when submitting the data to a linear mixed-effects model with Block and Context as fixed factors and Participant and Target Quadrant as random factors. The upper left subplot of Figure 1F illustrates the second ‘classical’ finding: a decline in the number of fixations across blocks, with overall fewer fixations made in repeated vs. non-repeated displays. The upper right plot of Figure 1F depicts scanpath length. Since targets positions were placed at equal distance from the screen center, the scanpath length also coincides with the so-called ‘scan-pattern ratio’ (Brockmole & Henderson, 2006): the total distance (in pixels) that the eyes traversed across a given search display until arriving at the target location, divided by the shortest distance possible between the initial fixation and the target location – essentially quantifying how directly the eyes approached the target (see, e.g., Manginelli & Pollmann, 2009). This measure turned out significantly smaller for repeated compared to non-repeated displays. Moreover, as can be seen in the lower left plot of Figure 1F, repeated configurations showed a smaller mean saccade amplitude. Of note, the average standard deviation of the saccade amplitudes was also significantly reduced for repeated vs. non-repeated displays – see lower right plot of Figure 1F.

Table 1

Oculomotor measures

Number of fixations

Main effect of Context

Mean repeated: 8.44

Mean non-repeated: 9.62

F(1,8576.5) = 142.71

p < 2*10^-16

Main effect of Block

Mean_{first block}: 11.29

Mean_{last block}: 8.36

F(31,8584.4) = 6.95

p < 2*10^-16

Interaction Block x Context

F(31,8584.8) = 1.08

p = 0.35

Scanpath length

Main effect of Context

Mean repeated: 1349.26

Mean non-repeated: 1611.78

F(1,8575.5) = 166.40

p < 2*10^-16

Main effect of Block

Mean_{first block}: 1903.37

Mean_{last block}: 1387.34

F(31,8582.2) = 5.72

p < 2*10^-16

Interaction Block x Context

F(31,8583.0) = 1.06

p = 0.38

Average saccade amplitude

Main effect of Context

Mean repeated: 151.72

Mean non-repeated: 160.01

F(1,8574.5) = 88.93

p < 2*10^-16

Main effect of Block

Mean_{first block}: 162.55

Mean_{last block}: 155.63

F(31,8580.7) = 1.50

p =0.036

Interaction Block x Context

F(31,8581.2) = 1.03

p = 0.41

Standard deviation of saccade amplitude

Main effect of Context

Mean repeated: 107.02

Mean non-repeated: 112.24

F(1,8575.2) = 44.60

p = 2.56*10^-11

Main effect of Block

Mean_{first block}: 118.17

Mean_{last block}: 109.95

F(31,8581.7) = 1.83

p = 0.0034

Interaction Block x Context

F(31,8582.5) = 0.93

p = 0.57

Table 1. The table presents summary and inference statistics for the four computed oculomotor measures of fixation number, scanpath length, average saccade amplitude, and standard deviation of saccade amplitude (the three latter in pixels).

Scanpath analysis

In a first, qualitative analysis, we visualized scanpaths across blocks and target positions. In more detail, we plotted the scanpath representations for each of the four target positions in repeated and, respectively, non-repeated displays (i.e., the four repeated and non-repeated trials within a block). As can be seen from Figure 1D (for a high-resolution image, see Appendix II: Figure 2), in the last block of the search task – after ample opportunity for contextual adaptation – there was a higher degree of similarity between scanpaths for repeated relative to non-repeated contexts. Specifically, in the repeated condition for each display, saccades tended to be more often and more closely executed in parallel direction across observers, indicative of systematic biases in observers’ oculomotor behavior. Also, in the repeated compared to the non-repeated condition, the color of the lines connecting successive fixation locations (with green denoting the initial and blue the final saccade on a given trial, and intermediate colors denoting saccades in between) tended to be more clustered spatially, indicating that saccades were executed in a more systematic order as well.

Similarity analysis

The main finding of this study is that scanpath similarity is increasing throughout the experiments, with repeated displays gaining a significant advantage early on which remains throughout the experiment. These findings are indicative of the convergence towards an optimal search strategy on the level of the set of displays (see discussion). For both analyses, within displays and, respectively, within participants (see Figure 1G and Table 2), we found significant main effects of Context and Block (all p < 0.001); the Context × Block interactions were non-significant (all p > 0.3). The pattern reflects a steady increase in scanpath similarity (corresponding to smaller numerical values in Dynamic Time Warping, Discrete Fréchet Distance, and Area Between Curve) over the course of the experiment (main effect of Block), which was however higher in repeated vs. non-repeated displays (main effect of Context) with a stable context effect emerging early on (no significant interaction). Of note, the three similarity measures yielded qualitatively similar results, despite being sensitive to slightly different aspects of the scanpaths, attesting to a high reliability of our findings.

Thus, our scanpath similarity measures provide strong support for a procedural-optimization hypothesis, according to which participants, over time on task, learn a generic search procedure that is increasingly effective for all – repeated and non-repeated – displays, though repeated displays weigh more highly in the optimization as a result of being searched more often.

Table 2

Similarity within display

Similarity within participant

DTW

Main effect of Context

Mean _repeated 1613.82

Mean _non-repeated 1987.74

F(1,2760.0) = 154.62

p < 2*10^-16

Main effect of Block

Mean_{first block}: 2329.61

Mean_{last block}: 1657.38

F(1,2760.1) = 6.48

p = 2*10^-16

Interaction Block x Context

F(1,2760.1) = 1.10

p = 0.32

DTW

Main effect of Context

Mean _repeated: 2247.55

Mean _non-repeated: 2628.75

F(1,189) = 99.32

p < 2*10^-16

Main effect of Block

Mean_{first block}: 3119.54

Mean_{last block}: 2180.57

F(31,189) = 2.80

p = 9.54*10^-6

Interaction Block x Context

F(31,189) = 0.58

p = 0.96

DFD

Main effect of Context

Mean _repeated: 291.7

Mean _non-repeated: 314.64

F(1,2760.0) = 14.90

p = 1.16*10^-4

Main effect of Block

Mean_{first block}: 337.8

Mean_{last block}: 292.78

F(1,2760.1) = 2.11

p = 3.36*10^-4

Interaction Block x Context

F(1,2760.0) = 0.71

p = 0.88

DFD

Main effect of Context

Mean _repeated: 419.14

Mean _non-repeated: 427.24

F(1,189) = 91.22

p < 2*10^-16

Main effect of Block

Mean_{first block}: 458.01

Mean_{last block}: 415.03

F(31,189) = 2.84

p = 3.78*10^-4

Interaction Block x Context

F(31,189) = 0.64

p = 0.93

ABC

Main effect of Context

Mean _repeated: 149860.73

Mean _non-repeated: 202522.02

F(1,2760.0) = 193.54

p < 2*10^-16

Main effect of Block

Mean_{first block}: 252733.5

Mean_{last block}: 153740.48

F(1,2760.1) = 5.80

p < 2*10^-16

Interaction Block x Context

F(1,2760.0) = 1.06

p = 0.37

ABC

Main effect of Context

Mean _repeated: 188744.43

Mean _non-repeated: 242078.25

F(1,189) = 107.64

p < 2*10^-16

Main effect of Block

Mean_{first block}: 297143.1

Mean_{last block}: 190412.32

F(31,189) = 2.87

p = 5.66*10^-6

Interaction Block x Context

F(31,189) = 0.61

p = 0.95

Table 2 Summary inference statistics for the two levels of analysis – within display and, respectively, within participant – for the three scanpath similarity measures of Dynamic Time Warping (DTW), Discrete Fréchet Distance (DFD), and Area Between Curves (ABC).

To gain an in-depth understanding of contextual facilitation, we analyzed established measures of the contextual-facilitation effect that focus on aggregate oculomotor indices, as well as new measures based on spatio-temporal scanpath sequences. Concerning individual eye-movement patterns: replicating prior reports, we found that detecting a target involves fewer fixations in repeated compared to non-repeated target-distractor arrangements (e.g., Peterson & Kramer, 2001; Tseng & Li, 2004; Zang, Jia, Müller, & Shi, 2015), as well as a shorter scanpath length and, accordingly, a smaller scan-pattern ratio (Brockmole & Henderson, 2006; Manginelli & Pollmann, 2009). We also found the saccade amplitudes to be significantly shorter for repeated displays (in contrast to Tseng & Li, 2004). In addition, we established a novel aggregate oculomotor index of contextual facilitation (that hitherto had not been reported in the literature): a reduced standard deviation of the saccade amplitudes for repeated vs. non-repeated displays.

Having established comparability of the present findings with those reported in prior studies of contextual cueing, we went on to analyze oculomotor scanpaths in order to differentiate between a display-specific and a procedural, display-generic account of acquired contextual facilitation and thus characterize the memory mechanism that drives the gains in the aggregate eye-movement indices (such as the reduced total fixation number) for repeated displays. Using two distinct approaches – of comparing eye movement sequences between pairs of identical displays when viewed by different participants, and, respectively, individual participants viewing different displays – and three metrics of scanpath similarity (Dynamic Time Warping, Discrete Fréchet Distance, and Area Between Curves), we found that, while the consistency of the scanpaths increased overall with increasing time on task for both repeated and non-repeated displays, these practice-dependent gains were more strongly driven by displays sampled repeatedly (vs. displays sampled only once) over the course of the task. This effect pattern is indicative of common regularities shared between the (statistical) sample of search displays, which influence scanpaths in a way that is independent of the particular arrangement encountered on a trial or even the individual participant – akin to generic, ‘one-for-all’ procedural learning. Had there been display-specific learning of individual repeated displays, the similarity measures obtained from any two such displays should have been reduced, relative to non-repeated (baseline) displays, as each repeated array should have come to elicit its unique scanning pattern. However, at variance with this prediction from display-specific learning accounts of contextual cueing, we found the similarity measures to be actually.

Of note, this pattern of scanpath-similarity effects does not rule out that the scanpaths become tuned to specific displays at some point along their progression, for instance, when the eye finally ‘homes in’ on the target location (Tseng & Li, 2004); nor do we take this to rule out the possibility of display-specific learning under all circumstances (considered further below). Rather, we take our findings to demonstrate that, over the course of a hard search task, the notion of display-general learning provides an apt account of the mechanism of contextual facilitation. In line with this, Tseng and Li (2004; see also Pollmann & Manginelli, 2009, or Kröll et al., 2019) have proposed that the (efficient) phase, in which the eyes come closer to the target with each successive fixation, does not differ significantly between repeated and non-repeated displays; but the two display types differ with respect to the number of fixations in the preceding inefficient phase (with reduced fixation numbers in repeated displays). However, in contrast to Tseng and Li (2004), our new scanpath-similarity analyses, which take into account the entire spatio-temporal series of fixation events, show that eye movements in the so-called ‘ineffective’ phase are, in fact, not (in terms of Tseng & Li’s, 2004, p. 1371) “wasted”. Instead, our findings of increasing scanpath homogeneity with extended time on task suggest that what may look an ‘ineffective’ phase actually constitutes an important period during which procedural learning of a general scanning scheme becomes functional.

Contextual-cueing of visual search: procedural and display-unspecific

Our new results suggest that display-unspecific, procedural learning of saccadic trajectories plays an important role in oculomotor contextual facilitation. A possible explanation is that, upon task instruction, participants explore or adjust individual-specific, idiosyncratic scanning strategies (Clarke, Nowakowska & Hunt, 2019) towards the current sensory input and, over time, settle on and shape the search strategy that is best suited to solve the task at hand (Fitts, 1964), here: a hard (letter) search task. It is important to note that if oculomotor scanning is allowed or encouraged, these developing strategies might look different from conditions in which observers are instructed to search the display covertly/passively (e.g., Lleras & von Mühlenen, 2004) or even without eye movements. In line with this, electrophysiological (EEG) studies report evidence of display-specific contextual cueing – more precisely: target-side-specific lateralizations of event-related potentials indicative of attentional resource allocation – as early as around 200 ms post display onset (Zinchenko, Conci, Töllner, Müller & Geyer, 2020; see also Schankin & Schubö, 2010). Of note, however, the search displays in these studies were presented for relatively short times (e.g., 700 ms in Zinchenko et al., 2020), coupled with an increased size and a reduced density of the display items to make it possible for participants to perform the task at all without eye movements (which would produce artifacts in EEG studies). These short presentation times may have ‘forced’ display-specific learning of distractor-target contextual relations, due to the need to buffer (at least parts of) the item arrangements in working memory (WM), in order to be able solve the task once the display arrays are no longer physically available (assuming that 700 ms are not sufficient for full processing of the items, as indexed by mean RTs being typically in excess of 1000 ms in T vs. L-type letter search tasks). This maintenance of item configurations in WM would lead to more display-specific encoding of search arrays into contextual long-term memory (e.g., Hartshorne & Makovski, 2019; Axmacher, Elger, & Fell, 2009). Moreover, even small changes in task requirements can lead to differences in WM engagement (see also Draschkow, Kallmayer, & Nobre, 2021).

In terms of skill acquisition (Fitts & Posner, 1967), when participants perform a new task of searching displays with novel, as yet to-be-discovered statistical properties, one would expect that they learn to adapt and optimize their oculomotor scanning behavior with respect to the scene statistics in rather generic terms – as opposed to acquiring search-guiding memory representations tailored to specific, individual display arrangements, as assumed by standard accounts of contextual cueing. These scanning strategies are optimal in the sense that they save processing ‘energy’: they do not require memorizing arbitrary distractor-target configurations and expensively checking a given display arrangement against a set of representations in contextual memory. As an unavoidable side effect, developing a strategy that is optimally adjusted to the statistical search environment at large (with repeated displays having a greater weight in shaping this strategy due to their increased frequency of occurrence) would also optimize the scanning of non-repeated displays (which, due to their random variation, have little weight in determining the strategy). Finally, based on evolutionary considerations, learning the overall statistics of a scene environment would not only be more efficient, but also be more robust to environmental changes (in ‘old’ scenes as well as the addition of new repeated scenes, which could be more easily incorporated in an environment-generic strategy) compared to learning highly specific display-target configurations.

Consistent with these functional considerations, there is also evidence from an fMRI study (including eye movements), by Manelis and Reder (2012), in line with a procedural-learning account of contextual cueing: When comparing the first with the last epoch of the experiment, Manelis and Reder found a significant decrease in functional connectivity between hippocampus and sensory-procedural areas (which they did, however, not expressly attribute to mechanisms related to procedural oculomotor learning). Of note, hippocampal activity is not only involved in motor-sequence learning (e.g., in finger-tapping tasks, Albouy, King, Maquet & Doyon, 2013), but also in other statistical learning tasks, such as serial RT tasks (e.g., Hazeltine, Grafton & Ivry, 1997; Müller, Kleinhans, Pierce, Kemmotsu & Courchesne, 2002), particularly in the initial stages, with a decrease in activity later on (Albouy et al., 2008). This is consistent with a critical, but (over the course of practice) diminishing role of the hippocampus in procedural motor learning and would go some way to explain why patients with damage to the hippocampus do not display a contextual-cueing benefit (Chun & Phelps, 1999). Thus, a procedural-learning account involving the hippocampus could explain the phenomenon of contextual cueing without assuming the acquisition of display-specific contextual representations.

Such a display-generic, procedural account of oculomotor contextual cueing would also explain why participants usually do not explicitly recognize repeatedly encountered display arrangements (Chun & Jiang, 1998): hippocampal involvement would revolve around procedural aspects of search performance, rather than explicit (episodic) memory. This view is not necessarily incompatible with existing functional accounts of long-term memory, according to which hippocampus contributes to the formation of inter-element associations: in the current context, these associations would involve the binding of individual eye fixations – thus making procedural memory a specific instance of a more general, associative hippocampal memory system (e.g., Chun & Phelps, 1999; Henke, 2010).

Contextual cueing is an important predictive-coding mechanism characterized by facilitation of search performance in repeated search arrays. As such, this facilitatory effect can be accommodated equally by accounts assuming associative learning of individual target locations in individual distractor arrangements (and the reproduction of individual scanpaths for these arrays) or, respectively, the acquisition of generic oculomotor patterns that optimize the scanning of the (for the participants initially overall new) set of search displays. The current study was designed to test the latter (display-generic) against the former (display-specific) account, by systematically investigating participants’ eye-movement trajectories in repeated displays and comparing them against non-repeated displays. Our findings are in line with a display-general scanning account of contextual facilitation: over time on task, scanpath sequences became increasingly similar across participants and displays, with total scanpath similarity being higher for repeated displays. We propose that contextual facilitation derives from the acquisition of ‘procedural’ oculomotor scanning programs, which become operational quite early during a given search trial. Conceptualizing contextual cueing as procedurally optimized oculomotor trajectories also offers new ways (1) for understanding the difficulty to update established ‘contextual memories’ following consistent target-position changes within otherwise unchanged distractor arrangements (Chun & Jiang, 1998, Experiment 3; Geyer, Seitz, Zinchenko, Müller, & Conci, 2021; Manginelli & Pollmann, 2009; Zellin, von Mühlenen, Müller, & Conci, 2013), as well as (2) for understanding the apparent ‘high capacity’ of ‘contextual memory’ (see, e.g., Jiang, Song, & Rigas, 2005; after being presented with 12 different target-distractor arrangements per day over a 5-day period, Jiang et al.’s participants showed contextual facilitation for the total number of 60 arrangements when tested at the end); and (3), it provides a possible explanation as to why contextual cueing leads to fMRI activations in sensory brain areas contributing to procedural learning (e.g., Manelis & Reder, 2012; see also Chaumon, Schwartz & Tallon-Baudry, 2009, for confirmatory evidence using MEG). Moreover, a display-generic account of contextual learning would provide a coherent and, in terms of Occam’s razor, the most simple explanation not only of the search advantage for repeated versus non-repeated displays, but also of the practice-dependent improvement of search in novel displays (Chun & Jiang, 1998) in terms of procedural learning or skill acquisition (Fitts & Posner, 1967). Finally, given our new evidence that oculomotor search is optimized independently of a particular configuration as a skill of performing a visual search task in general, with repeated displays merely having a stronger influence in the optimization process and therefore leading to overall better performance: we propose to use the more neutral term ‘contextual facilitation’ (instead of display-specific ‘contextual cueing’) to describe the effects of procedural spatial learning in visual search.

Author contributions

WS and AZ conceived and designed the experiments; AZ performed the experiments: WS analyzed the data; HJM contributed to discussion; WS and TG wrote the paper.

Additional information

Competing Interests: The authors declare that they have no competing interests.

Acknowledgement

This research was supported by grants of the German Research Council (DFG) to Thomas Geyer (GE 1889/4-2 & RTG 2175).

Albouy, G., Sterpenich, V., Balteau, E., Vandewalle, G., Desseilles, M., Dang-Vu, T., ... & Maquet, P. (2008). Both the hippocampus and striatum are involved in consolidation of motor sequence memory. Neuron, 58(2), 261-272.
Albouy, G., King, B. R., Maquet, P., & Doyon, J. (2013). Hippocampus and striatum: Dynamics and interaction during acquisition and sleep‐related motor sequence memory consolidation. Hippocampus, 23(11), 985-1004.
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological review, 89(4), 369.
Axmacher, N., Elger, C. E., & Fell, J. (2009). Working memory-related hippocampal deactivation interferes with long-term memory formation. Journal of Neuroscience, 29(4), 1052-1060.
Beesley, T., Vadillo, M. A., Pearson, D., & Shanks, D. R. (2016). Configural learning in contextual cuing of visual search. Journal of Experimental Psychology: Human Perception and Performance, 42(8), 1173.
Brady, T. F., & Chun, M. M. (2007). Spatial constraints on learning in visual search: modeling contextual cuing. Journal of Experimental Psychology: Human Perception and Performance, 33(4), 798.
Brainard, D. H., & Vision, S. (1997). The psychophysics toolbox. Spatial vision, 10(4), 433-436.
Brockmole, J. R., & Henderson, J. M. (2006). Recognition and attention guidance during contextual cueing in real-world scenes: Evidence from eye movements. The Quarterly Journal of Experimental Psychology, 59 (7), 1177–1187.
Chaumon, M., Schwartz, D., & Tallon-Baudry, C. (2009). Unconscious learning versus visual perception: dissociable roles for gamma oscillations revealed in MEG. Journal of Cognitive Neuroscience, 21(12), 2287-2299.
Chun, M. M., & Jiang, Y. (1998). Contextual cueing: Implicit learning and memory of visual context guides spatial attention. Cognitive Psychology, 36(1), 28-71.
Jiang, Y., & Chun, M. M. (2003). Contextual cueing: Reciprocal influences between attention and implicit learning.
Chun, M. M., & Phelps, E. A. (1999). Memory deficits for implicit contextual information in amnesic subjects with hippocampal damage. Nature neuroscience, 2(9), 844-847.
Clarke, A. D., Nowakowska, A., & Hunt, A. R. (2019). Seeing beyond salience and guidance: the role of bias and decision in visual search. Vision, 3(3), 46.
Draschkow, D., Kallmayer, M., & Nobre, A. C. (2021). When natural behavior engages working memory. Current Biology, 31(4), 869-874.
Fahimi, R., & Bruce, N. D. (2021). On metrics for measuring scanpath similarity. Behavior Research Methods, 53(2), 609-628.
Fitts, P. M. (1964). Perceptual-motor skill learning. In Categories of human learning (pp. 243-285). Academic Press.
Fitts, P. M., & Posner, M. I. (1967). Human performance. brooks. Cole, Belmont, CA, 5, 7-16.
Geyer, T., Rostami, P., Sogerer, L., Schlagbauer, B., & Müller, H. J. (2020). Task-based memory systems in contextual-cueing of visual search and explicit recognition. Scientific Reports, 10(1), 1-13.
Geyer, T., Seitz, W., Zinchenko, A., Müller, H. J., & Conci, M. (2021). Why are acquired search-guiding context memories resistant to updating?. Frontiers in Psychology, 12, 564.
Geyer, T., Zehetleitner, M., & Müller, H. J. (2010). Contextual cueing of pop-out visual search: When context guides the deployment of attention. Journal of vision, 10(5), 20-20.
Goujon, A., Didierjean, A., & Thorpe, S. (2015). Investigating implicit statistical learning mechanisms through contextual cueing. Trends in cognitive sciences, 19(9), 524-533.
Harris, A. M., & Remington, R. W. (2017). Contextual cueing improves attentional guidance, even when guidance is supposedly optimal. Journal of Experimental Psychology: Human Perception and Performance, 43(5), 926.
Hartshorne, J. K., & Makovski, T. (2019). The effect of working memory maintenance on long-term memory. Memory & cognition, 47(4), 749-763.
Hazeltine, E., Grafton, S. T., & Ivry, R. (1997). Attention and stimulus characteristics determine the locus of motor-sequence encoding. A PET study. Brain: a journal of neurology, 120(1), 123-140.Henke, K. (2010). A model for memory systems based on processing modes rather than consciousness. Nature Reviews Neuroscience, 11(7), 523-532.
Jekel, C. F., Venter, G., Venter, M. P., Stander, N., & Haftka, R. T. (2019). Similarity measures for identifying material parameters from hysteresis loops using inverse analysis. International Journal of Material Forming, 12(3), 355-378.
Kroell, L. M., Schlagbauer, B., Zinchenko, A., Müller, H. J., & Geyer, T. (2019). Behavioural evidence for a single memory system in contextual cueing. Visual Cognition, 27(5-8), 551-562.
Lleras, A., & Von Muhlenen, A. (2004). Spatial context and top-down strategies in visual search. Spatial vision, 17(4), 465-482.
Logan, G. D. (1988). Toward an instance theory of automatization. Psychological review, 95(4), 492.
Manelis, A., & Reder, L. M. (2012). Procedural learning and associative memory mechanisms contribute to contextual cueing: Evidence from fMRI and eye-tracking. Learning & Memory, 19(11), 527-534.
Manginelli, A. A., & Pollmann, S. (2009). Misleading contextual cues: How do they affect visual search? Psychological Research, 73(2), 212-221.
Moran, R., Zehetleitner, M., Liesefeld, H. R., Müller, H. J., & Usher, M. (2016). Serial vs. parallel models of attention in visual search: accounting for benchmark RT-distributions. Psychonomic Bulletin & Review, 23, 1300-1315.
Müller, R. A., Kleinhans, N., Pierce, K., Kemmotsu, N., & Courchesne, E. (2002). Functional MRI of motor sequence acquisition: effects of learning stage and performance. Cognitive Brain Research, 14(2), 277-293.
Myers, C. W., & Gray, W. D. (2010). Visual scan adaptation during repeated visual search. Journal of Vision, 10(8), 4-4.
Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: Transforming numbers into movies. Spatial Vision, 10, 437-442.
Peterson, M. S., & Kramer, A. F. (2001). Attentional guidance of the eyes by contextual information and abrupt onsets. Perception & Psychophysics, 63 (7), 1239–1249.
Peterson, M., Mead, P., Kelly, S., Esser-Adomako, E., & Blumberg, E. (2022). Spatial context target relearning following a target relocation event: Not mission impossible. Attention, Perception, & Psychophysics, 84(2), 474-489.
Pollmann, S., & Manginelli, A. A. (2009). Anterior prefrontal involvement in implicit contextual change detection. Frontiers in Human Neuroscience, 3, 28.
R Core Team (2018). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.URL http://www.R-project.org/.
Sewell, D. K., Colagiuri, B., & Livesey, E. J. (2018). Response time modeling reveals multiple contextual cuing mechanisms. Psychonomic Bulletin & Review, 25(5), 1644-1665.
Sisk, C. A., Remington, R. W., & Jiang, Y. V. (2019). Mechanisms of contextual cueing: A tutorial review. Attention, Perception, & Psychophysics, 81(8), 2571-2589.
Schankin, A., & Schubö, A. (2010). Contextual cueing effects despite spatially cued target locations. Psychophysiology, 47(4), 717-727.
Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84(2), 127.
Tseng, Y.-C., & Li, C.-S. R. (2004). Oculomotor correlates of context-guided learning in visual search. Perception & Psychophysics, 66 (8), 1363–1378.
Vadillo, M. A., Konstantinidis, E., & Shanks, D. R. (2016). Underpowered samples, false negatives, and unconscious learning. Psychonomic Bulletin & Review, 23(1), 87-102.
Vadillo, M. A., Malejka, S., Lee, D. Y., Dienes, Z., & Shanks, D. R. (2021). Raising awareness about measurement error in research on unconscious mental processes. Psychonomic Bulletin & Review, 1-23.
Van Rossum, G., & Drake Jr, F. L. (1995). Python reference manual. Centrum voor Wiskunde en Informatica Amsterdam.
Wolfe, J. M., Cave, K. R., & Franzel, S. L. (1989). Guided search: an alternative to the feature integration model for visual search. Journal of Experimental Psychology: Human perception and performance, 15(3), 419.
Wolfe, J. M. (1994). Guided search 2.0 a revised model of visual search. Psychonomic bulletin & review, 1(2), 202-238.
Wolfe, J. M. (2021). Guided Search 6.0: An updated model of visual search. Psychonomic Bulletin & Review, 1-33.
Zang, X., Jia, L., Müller, H. J., & Shi, Z. (2015). Invariant spatial context is learned but not retrieved in gaze-contingent tunnel-view search. Journal of Experimental Psychology: Learning, Memory, and Cognition, 41(3), 807.
Zellin, M., von Mühlenen, A., Müller, H. J., & Conci, M. (2013). Statistical learning in the past modulates contextual cueing in the future. Journal of Vision, 13(3), 19-19.
Zinchenko, A., Conci, M., Töllner, T., Müller, H. J., & Geyer, T. (2020). Automatic guidance (and misguidance) of visuospatial attention by acquired scene memory: evidence from an N1pc polarity reversal. Psychological Science, 31(12), 1531-1543.

There is NO Competing Interest.

Appendix.docx

Download PDF

Journal Publication

published 20 Sep, 2023

Read the published version in Communications Psychology →

Version 1

posted

You are reading this latest preprint version

Statistical learning in visual search: ‘contextual cueing’ reflects the acquisition of an optimal, ‘one-for-all’ oculomotor scanning strategy

Status:

Journal Publication

Version 1

Abstract

Figures

Significance Statement

Introduction

Methods

Results

Discussion

Summary And Conclusions

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1