VISUAL IMPLICIT LEARNING OVERCOMES LIMITS IN HUMAN ATTENTION

Transcription

1 VISUAL IMPLICIT LEARNING OVERCOMES LIMITS IN HUMAN ATTENTION Y. V. Jiang*, L.-W. King, W. M. Shim, and T. J. Vickery Department of Psychology Harvard University Cambridge, MA ABSTRACT The human cognitive system is stunningly powerful in some respects yet surprisingly limited in others. We can recognize an object or a face in a single glimpse and type 70 words per minute, yet we cannot hold more than a few objects at a time in working memory or split our attention to several locations. Attention and working memory impose major capacity limitations on cognitive processing. This study focuses on implicit visual learning, a process that allows us to overcome cognitive limitations. We examined variability across individuals in their ability to learn spatial context, and the neural substrates underlying learning. Results showed that variations across normal adults in spatial context learning do not correlate with an individual s episodic memory or spatial ability, suggesting that limits in an individual s episodic memory or spatial abilities do not constrain visual implicit learning. The variations are inconsistent across testing sessions, with no single individuals persistently failing to learn. Brain regions include parahippocampal place area and the posterior parietal cortex are involved in different aspects of spatial context learning. 1. INTRODUCTION Humans process a visual display more efficiently when they encounter it for a second time. A previously perceived object, now presented briefly, is identified more accurately than a new object, showing visual priming (Tulving & Schacter, 1990). Such perceptual facilitation is seen not only for isolated shapes or words, but also for complex visual displays. When conducting visual search for a T target among L distractors, observers are faster at detecting the target when the same display is occasionally repeated, even though they are unaware of the repetitions (Chun & Jiang, 1998, 2003). Implicit learning of repeated visual display allows attention to be deployed to target regions defined by past experience. This learning, known as contextual cueing, is acquired rapidly. Its effect on search appears after only five or six repetitions and is preserved even when only half of the items are repeated (Song & Jiang, 2005). At least 60 repeated displays can be learned, suggesting that visual implicit learning is not constrained by low capacity limits (Jiang et al., 2005). Once acquired, learning persists for at least a week (Chun & Jiang, 2003; Jiang et al., 2005). Implicit visual learning may compensate for our limitations in moment-tomoment attentional capacaity Individual differences Given its potential importance in vision, one may expect every individual within the normal population to show spatial context learning. In fact, the degree of individual differences is generally narrower for implicit processes than explicit processes. Explicit mental processes vary considerably across individuals. For example, in normal adults, some individuals have better autobiographical memory than others (Schacter, 1996). Across different population groups, children and older adults have poorer source memory than young adults (Johnson, 1997), and amnesic patients are impaired at explicit memory tasks (Gabrieli, 1998). Implicit processes, however, show a more limited degree of individual variability. Reber (1993) has argued, on the basis of robust artificial grammar learning in children and older adults, that implicit learning is an evolutionarily and ontogenetically earlier system. It is preserved even when other cognitive functions are impaired. Individuals with low IQ, such as children with William s Syndrome, show relatively normal artificial grammar learning (Don et al., 2003). Patients with severe close-head injury also show preserved perceptual learning (Nissley & Schmitter- Edgecombe, 2002). Yet in our own research we often failed to see learning in a small subset of subjects (approximately 20%; see also Lleras & von Muhlenen, 2004). Such variability within a homogenous group of subjects normal adults (most of whom are college students) is puzzling given that implicit learning is considered a stable ability across a variety of population groups. An important goal of the present study is to examine the degree of individual variability in spatial context learning, what other cognitive abilities correlate with it, and its test-retest reliability. We note that studying variability across individuals is inherently a correlational method. Even if we may find, for example, that spatial context learning correlates with an individual s spatial navigation ability, it will still not identify the source that causes such individual differences. Conversely, a lack of correlation between contextual cueing and certain cognitive skills does not eliminate the possibility that other, as yet uninvestigated skills may correlate with it. 1

2 Thus, compared with an experimental approach in which researchers directly manipulate an experimental variable, an individual differences approach has limited inferential power. However it has been successfully applied to other domains of research, such as working memory (Baddeley, 1986). With respect to spatial context learning, an individual differences approach can be insightful for at least two reasons. First, it can potentially dismiss the notion that implicit learning in general is a basic ability that pretty much everyone possesses. Second, by finding out which cognitive functions correlate with contextual cueing, we can guide future experimental research to isolate possible causes of the correlation Neural substrates Recent development in functional neuroimaging has promoted a flourish of brain imaging studies that explored the neural basis for implicit learning. While some maintains that different brain regions are associated with implicit and explicit learning (e.g., Reber et al., 2003, on category learning), others propose that the activation of certain regions, such as the medial temporal lobe, is determined by the nature of the learned materials rather than by awareness (Rose et al., 2002). Learning to search faster on repeated displays relies on at least two processes: encoding the repetition, and associating each repeated display with a consistent target location. Behavioral studies, including some of our own, have shown that repeating displays alone without repeating target location does not enhance visual search. This observation led some to conclude that no memory is established from repetitions (Wolfe et al., 2000). Yet subjects may have acquired perceptual familiarity from repeated displays, even though no behavioral gain is shown. Functional Magnetic Resonance Imaging (fmri) provides a good tool to test such memory. Recent neuropsychological and neuroimaging studies have revealed a reliance of spatial implicit learning on the hippocampus and surrounding medial temporal lobe regions (Chun & Phelps, 1999; Preston et al., 2001, SFN conference abstract). It is unclear though, whether these regions are important for encoding display repetition, or for associating a repeated display with a target location. It is possible that the medial temporal lobe is important for associative learning, while other more perceptual brain regions are important for encoding display repetition. One goal of this study is to identify the neural correlates of spatial context learning. Specifically, are the same brain regions involved in different aspects of learning, or do different brain regions underlie different components of learning? 2. STUDY 1. INDIVIDUAL DIFFERENCES Given that one can test the correlation between spatial context learning and only a few tasks out of an infinite number, the selection of other tasks to test for individual differences is key to this research. Our choice of tasks is guided by studies that show a reliance of contextual cueing on normal medial temporal lobe functions. Amnesic patients with extensive medial temporal lobe damage are impaired in contextual cueing (Chun & Phelps, 1999; Manns & Squire, 2001), even though they are known to do well on sequence learning, motor pursuit, perceptual priming, and other implicit tasks (Schacter & Buckner, 1998). Given that the medial temporal lobe is important for spatial context learning, and that this region is also involved in episodic memory (Vargha-Khadem et al., 1997) and spatial navigation (Cohen & Eichenbaum, 1993), we reason that tasks that tap into episodic memory and spatial abilities will be good candidates to correlate with contextual cueing. In this study, subjects were tested in several tasks, including: contextual cueing, word memory (as a measure of episodic memory), spatial IQ test (as a measure of subjects spatial ability), visual working memory (another task that relies on the medial temporal lobe, Olson et al., 2006), and task switching (a difficult task that depends on frontal lobe but not the medial temporal lobe). Subjects also filled in a questionnaire to rate their spatial navigation ability and the clarity of their autobiographical memory Method Participants: Eighteen students from Harvard University (13 females and 4 males, mean = 22.7 years) volunteered to participate in this experiment for payment. They all had normal or corrected-to-normal visual acuity. Tasks and procedures: Each participant was tested in six tasks in random order. The tasks are specified next. (1) Contextual cueing. Each subject participated in 24 blocks of visual search trials. Each block included 9 repeated and 9 non-repeated displays. Subjects searched for a rotated T among rotated Ls and reported the orientation of the T (either left or right). The display consisted of 1 white T rotated 90º to the left or right, and 11 white Ls rotated 0º, 90º, 180º, or 270º. Items were presented at randomly selected locations from an invisible 10 x 10 grid matrix. Subjects pressed the spacebar to initiate each block. On each trial, after a 500ms fixation period, the search display was presented until a response was made. Accuracy feedback immediately followed each response. The design of this task was similar to Chun and Jiang (1998, Experiment 1). Within a block of 18 unique trials, 2

3 the target was at a different location on each trial. But across blocks, the target locations were repeated (i.e., target locations shown on block 1 were also presented on block 2). The distractor locations were repeated across blocks in the old condition, but were newly generated in each block in the new condition. The new and old displays were randomly intermixed within a block. All subjects were tested on the same new and old displays, ensuring that individual variability cannot be attributed to differences in search stimuli. (2) Word memory. Subjects read aloud a list of 40 sequentially presented words. The words were presented at the center of the display, at a pace of 1 word per second. Subjects were informed that their memory for the words would later be tested. After a 5-10 minutes filled delay during which subjects completed the spatial IQ tests and mental rotation, a recognition test was administered. Two words, one read a moment ago and one novel, were presented side by side on the computer screen. Subjects were asked to identify the word presented before. We recorded recognition accuracy. (3) Spatial abilities. Seven questions taken from a spatial IQ test and 10 questions of mental rotation were administered. In the spatial IQ test, subjects were allowed 30 seconds to view the display and respond. The questions entailed mentally folding boxes, selecting a piece to fill in a jigsaw puzzle, and choosing a shape that s the odd one out. In the mental rotation task, subjects were allowed 10 seconds per trial to view the display and respond. Half of the mental rotation trials used Shepard and Metzler s (1971) 3-D stimuli, the other half used Tarr and Pinker s (1989) 2-D stimuli. Subjects were shown a sample object above fixation, and two choice stimuli side by side below fixation. One of the choice stimuli was rotated from the sample, while the other was rotated from the sample s mirror image. The angle of rotation was between 100º and 165º. Subjects were asked to select the rotated sample. We measured accuracy. (4) Visual working memory. A change-detection task tested one s visual working memory for spatial locations (Jiang et al., 2000). Subjects were shown two displays of dots separated by a 1 second interval. Each dot subtended 0.8º and was printed in green against a gray background. The dots were presented at randomly selected locations within an invisible 10 x 10 matrix (22º). The first display contained 11 dots and was presented for 200msec. The second display contained 12 dots, all but one matched the first display. Subjects were asked to find the newly added dot. We measured response accuracy. (5) Task switching. We thought that it was important to test control tasks that are not likely to rely on the medial temporal lobe. This ensured that any correlation between contextual cueing and word memory (or spatial ability etc.) was specific to a given cognitive function. In this task, subjects were presented with four colored digits in a horizontal array, one of which was the target. Subjects were asked to report either the target s identity or its spatial position among the four. On each trial a cue informed subjects which digit was the target and which task to perform. Subjects had 2 seconds to make a response before the next trial started. We measured the total number of trials subjects performed before they made 10 consecutively correct responses. Because the cue changed randomly from trial-to-trial and each trial was time-limited, this was a difficult task that brought out a wide range of individual differences. (6) Questionnaire. There were 7 questions on the questionnaire. They asked subjects to rate the clarity of their autobiographical memory as well as their spatial navigation ability. Subjects also rated how organized they were, a characteristic that is unlikely dependent on the medial temporal lobe Results This experiment generated a large array of data, of which only results relevant to the purpose of this study are reported below. FIGURE 1. Learning of repeated spatial context. Search speed became faster for repeated displays (old) than unrepeated displays (new). The error bars show standard error of the difference between old and new. Contextual cueing. An ANOVA on condition (old vs. new) and block (blocks 1 to 24) revealed a significant main effect of condition, F(1, 18) = 18.78, p <.001, a significant main effect of block, F(23, 391) = 7.77, p <.001, and a significant interaction effect, F(23, 391) = 1.68, p <.03. The new and old conditions did not differ initially, but as the experiment progressed, subjects became faster in the old condition. In the second half of the experiment, the average contextual cueing effect was 137msec. Figure 1 shows the results. 3

4 Correlation between contextual cueing and other tasks. Of the 18 individuals, only one subject showed a negative cueing effect (-24ms), all others were on the positive side, with the largest cueing effect of 265ms. However, the size of the contextual cueing effect did not correlate with any other factors: word memory, r = 0.083; spatial IQ: r = 0.22; mental rotation: r = 0.09; visual working memory, r = -0.19; and task switching, r = 0.20, all p-values >.35. Nor did contextual cueing correlate with self-reported autobiographical memory and spatial navigation ability, all ps > Discussion If individual differences in spatial context learning reflect stable cognitive skills, then they ought to correlate significantly with other measures of those skills. What skills would those be? Guided by the knowledge that a normal medial temporal lobe is necessary for spatial context learning (Chun & Phelps, 1999; Manns & Squire, 2001), we tested subjects in word recognition, spatial IQ and mental rotation, and visual working memory tasks, purported to tap into medial temporal lobe functions. We reasoned that these tasks were likely to correlate with contextual cueing. Our results, however, failed to show significant correlations. Individuals who showed a large contextual cueing effect did not necessarily show a good episodic memory or visual working memory, and they did not necessarily perform well on spatial IQ and mental rotation tasks. What contributed to this lack of correlation? We found that it was not because the other tasks we selected were unreliable. When the subjects were tested in a second session on the same tasks (but with different questions or materials), we found a high degree of testretest correlation in the tasks we selected: mental rotation, r =.72; spatial IQ, r =.44; visual working memory, r =.69; and task switching, r =.54. In addition, mental rotation, spatial IQ, and visual working memory are significantly correlated with one another, all r values >.51. Thus, the tasks we selected to correlate with contextual cueing were reliable and showed consistent individual differences: those who can hold more spatial locations in visual working memory tend to do so reliably on multiple occasions, for instance. Contextual cueing, however, poses an important exception. In a separate study (Jiang et al., 2005), we tested a group of 12 individuals on contextual cueing tasks across 5 sessions. We measured the size of the contextual cueing effect in each session for each individual and rank ordered an individual s placement in the group for each session. The rank order was highly inconsistent: individuals who ranked high in one session did not necessarily rank high in another session. The correlation across sessions was not significantly different from zero. There was also no correlation in raw RT data. Contextual cueing thus appears to provide an important exception to the observation that consistent individual differences are revealed across many cognitive tasks. An individual who does not show learning in one testing session is not necessarily someone who cannot learn from repeated spatial context. If tested again that individual may show as much learning as other participants do. Indeed, a person who fails to show learning in one session may nonetheless show it a week later, even when the same displays are tested (Jiang et al., 2005). Thus, the potential to learn from repeated spatial context most likely exists in every normal adult, but whether learning is revealed in a given testing session is less predictable. Experimental factors contributing to the variability of learning may include task difficulty, set size, and the mode of attention (Lleras & von Muhlenen, 2004). Two caveats must be noted with regard to the lack of consistent individual differences in contextual cueing. First, given that people with damaged medial temporal lobe do not learn repeated spatial context, learning is not retained for everyone. By testing college students only, however, we may have missed an opportunity to observe wider individual differences in the normal population. It remains an important question to identify individuals who consistently fail to learn, and to pinpoint the source of that failure. Second, a correlational approach taken by the current study has limited inferential power. Given that we did not, and could not, exhaust all possible tasks, the lack of correlation between contextual cueing and the tasks tested here does not eliminate the possibility that contextual cueing will be reliably correlated with other tasks. Factors that we have steered away for lack of a clear theoretical motivation, such as an individual s cognitive style, gender, handedness, and personality traits, may turn out to be correlated with contextual cueing. While we cannot preclude the possibility that contextual cueing may be correlated with some, as yet untested factors, such as motor learning or procedural learning, we note that such correlation will necessarily be difficult to find. This is because the contextual cueing task lacks testretest reliability. Unless future studies increase the testretest reliability, any search for cognitive skills correlated with an individual s spatial context learning will be futile. Whether the same conclusions hold true for other types of visual statistical learning remain to be tested in the future. 3. STUDY 2: NEURAL SUBSTRATES What is the neural basis of spatial context learning? This simple question has turned out to be a challenge to answer. In pilot studies we scanned subjects while they performed a standard visual search task, searching for a T among Ls. We found very few regions of the brain sensitive to the repetition of spatial context, even though 4

5 subjects were much faster searching from repeated displays. To study the neural substrates of spatial context learning, we introduced natural scenes into the search display (see also Brockmole & Henderson, 2006). On each trial subjects viewed a circular array of 16 elements, one of which was a letter T rotated to the left or to the right. The elements were presented against a natural scene background. In the old condition, the background scene was repeated across blocks, and the target T was always presented at a fixed location against a particular background scene. Thus for example, a supermarket scene always signified that the target was at the 12 o clock position. In the new condition, neither the background scene nor the target location was repeated. To differentiate scene familiarity from associative learning, we also included a shuffled condition where the background scene was repeatedly presented, but the target location could be at any of the 16 locations, randomly determined. Thus, although a particular scene was repeated many times in the experiment, it provided no information about where the target would be. By comparing the three conditions, it was possible to separate associative learning from scene familiarity. First, neither the shuffled nor the new conditions allowed subjects to build consistent association between a scene and the target location. The difference between these two conditions is therefore limited to perceptual familiarity: the shuffled condition had repeated pictures, whereas the new condition had unfamiliar pictures. On the other hand, both the old condition and the shuffled conditions used familiar scenes. However, only the old condition had specific target locations assigned to each scene. Thus, shuffled and old conditions were comparable in perceptual familiarity, but differed in associative learning. Figure 2 illustrates the three conditions. FIGURE 2. Schematic illustration of conditions tested in study 2 and pilot behavioral data. Behavioral data from 16 subjects collected outside of the neuroimaging scanner showed that subjects searched 5 faster in the old condition than the shuffled condition and the new condition. The latter two did not differ from each other, suggesting that performance was not enhanced by perceptual familiarity alone. Our brain imaging study thus adopted the natural scene-based learning task Method Participants: Twelve normal adults from Harvard University and its community participated in the brain imaging experiment for payment. They all had normal or corrected-to-normal visual acuity. The study protocol was approved by Partners IRB (Massachusetts General Hospital) and by Harvard University IRB. Visual search stimuli: Each visual search trial of 2.5 seconds consisted of a 200msec fixation period followed by the search display for 2300msec. The search display contained a background scene, selected from personal collection and online web sources, overlaid with visual search items. The scenes were photographs of natural scenes or city scenes, without humans as their central figures. Search items consisted of 1 black T and 15 distractor L s arranged in a circular array (8º radius). Each T or L appeared on a gray circle (1º diameter) to enhance visibility. The T was rotated 90º either to the left or to the right. Subjects task was to search for the T and press one of two keys to report its orientation. Visual search design: Subjects were tested in three conditions old, shuffled, and new in two fmri experiments, one using blocked-design and the other using rapid event-related design. In the old condition the background scene and the target location remained constant from one repetition to another. In the new condition, a novel background scene was used. In the shuffled condition, the background scene repeated, but the target location changed with each appearance of the background. There were 16 different trials in each condition. Procedure: The experiment was separated into several tasks on two different days. On the training day, participants were tested in a behavioral experiment involving 24 blocks of training trials. In each block of 32 trials, subjects saw old and shuffled trials equally often. The training session familiarized subjects with the repeated scenes and trained them to associate an old scene with a particular target location. The next day subjects received five blocks of warm-up trials, using the same images as they saw during the training day. Subsequently subjects were tested in an fmri session where they saw old, shuffled, and new displays. Brain imaging session. Subjects were scanned in a 3.0T scanner at the Matinos Center for Brain Imaging in Charlestown, MA. We first collected one scan of high

6 resolution T1 structural images. Standard T2* weighted EPI sequences were used for functional scans (TR = 2000msec, TE = 30msec, flip angle = 90º, slice thickness = 4mm, in-plane resolution = x mm). We sampled 28 axial slices that covered the entire brain except for the bottom of cerebellum. Visual stimuli were back-projected onto a magnetic-compatible screen and were reflected to subjects eyes through a 45º mirror. Each subject participated in 8 scans, half involved blocked-design where the three conditions were tested in separate blocks. Each scan of blocked-design contained 6 blocks. Each block lasted 40 seconds including 16 trials each presented for 2.5 sec each. The old, new, and, shuffled conditions were presented in separate blocks, each appearing in two blocks of a scan. The task blocks were separated by 16 seconds of fixation. Fixation also preceded the first block and followed the last block. The order of the conditions was counterbalanced within a scan and across scans. The duration of a scan was 5 minutes 52 seconds. The other four scans involved event-related design where the three conditions were randomly intermixed in presentation, each presented in 32 trials, along with 32 blank fixation trials (each trial lasted 2.5 seconds), in a continuous scan of 5 minutes 20 seconds. In addition to the visual search task, subjects also participated in two localizer tasks to localize parahippocampal place area (PPA), the hippocampus, and the posterior parietal cortex. The PPA localizer involved blocks of trials where subjects viewed natural scenes or scrambled images. The scrambled images were created by chopping a natural scene image into 768 small pieces and re-arranging the pieces randomly. Brain regions activated more by natural scenes than scrambled scenes (at p <.001) were localized in each subject s brain. This allowed us to identify the parahippocampal place area in each subject (see Epstein & Kanwisher, 1998). The parietal/hippocampal localizer involved blocks of trials where subjects viewed 4 different images and had to (1) learn a hidden rule that mapped the four images to four response keys (rule-learning condition), or (2) press a key corresponding to the location of a red box (control condition) while viewing the image. The rule-learning task was found to activate hippocampus and posterior parietal cortex in previous studies (e.g., Law et al., 2005). However, actual brain activation observed in our study failed to localize hippocampus reliably. Only the posterior parietal cortex was reliably localized. Data analysis: fmri data were analyzed using SPM 99 and in-house software. Each subject s data were motion corrected and normalized onto a common brain space (the Montreal Neurological Institute template). Data were smoothed using a Gaussian filter with a Full Width Half Maximum of 8mm, and high-pass filtered during analysis. For the blocked-design data, whole brain analysis across the 12 individuals was carried out for all pair-wise contrast among the three visual search conditions (old, new, and shuffled). For the event-related design data, a regions of interest (ROI) analysis centered on parahippocampal place area (PPA, defined by the PPA-localizer) and superior parietal lobule (SPL, defined by the rule-learning localizer) was carried out. The ROI analysis was used to improve statistical power (Saxe et al., 2006) Results FIGURE 3. Brain regions significantly activated more by search from novel displays than repeated displays. A whole brain analysis showed that parahippocampal place area and superior parietal lobule were both more active when subjects searched from novel displays than from repeated, old displays. Figure 3 shows the brain activation images at p <.001 uncorrected for multiple comparisons. Independent regions of interest defined by the PPAlocalizer showed that PPA was affected primarily by perceptual familiarity, but not by associative learning. In the blocked-design experiment, PPA activity was higher for the new condition than old or shuffled conditions (ps <.009), but was equally low for the old and the shuffled conditions (p >.10), suggesting that it was sensitive to scene-repetition regardless of whether the scene was consistently associated with a target location. This pattern of results also holds for data collected in the event-related design (ps <.02 for new compared with old and shuffled; p >.10 for old compared with shuffled). Figure 4 shows activation pattern in the PPA. In contrast, the superior parietal lobule (SPL), localized independently by the rule-learning task, showed reduced activation in the old condition compared with the other conditions (p <.05), suggesting that this region was sensitive to associative learning but not to perceptual familiarity. These results make sense given that the SPL was known to be involved in visual attention (e.g., Corbetta & Shulman, 2002). Attention can be quickly guided to the target location in the old condition. This guidance was absent in the new and the shuffled conditions, leading to differences in attention-related brain regions such as the SPL. The involvement of SPL in memory-based attentional cueing is consistent with the 6

7 idea that perception-based and memory-based attentional guidance share common neural substrates (Summerfield et al., 2006). FIGURE 4. Activation in the Parahippocampal place area (PPA) for the three visual search tasks. PPA was localized by a separate localizer that compared scenes with scrambled images. (PPA MNI coordinates: [ ] and [ ]) that common neural substrates may underlie memorybased attention and perception-based attention (Summerfield et al., 2006). FIGURE 5. Activation in the superior parietal lobule (SPL) in scene-based contextual cueing. SPL was localized by a separate task that compared a difficult rule-learning task with a simple rule-learning baseline. (SPL MNI coordinates: [ ] and [ ]). 3.3 Discussion Human attention and working memory pose severe limitations on cognitive processing. These limitations are major causes of human errors in driving, aircraft control, and combat. An important mechanism that allows us to overcome cognitive limitations is implicit visual learning. Studies on contextual cueing showed that there did not appear to be a clear capacity limitation in implicit learning. As many as 60 repeated displays can be learned quickly without proactive or retroactive interference (Jiang et al., 2005). At least two processes underlie this learning: perceptual familiarity with repeated visual input, and associative learning between repeated display and target information. Our brain imaging study has shown that both aspects of learning change the way visual input is processed. Perceptual familiarity with repeated visual information reduces activation in the parahippocampal place area, indicating priming or adaptation to the repeated input (Grill-Spector & Malach, 2001). Although perceptual familiarity alone does not enhance behavioral performance (Wolfe et al., 2000), its influence on the brain activity suggests that repetition is registered, and therefore, the cognitive system is not amnesic to repetition (Shen & Jiang, 2006). Behavioral gain emerges when the repeated information is predictive of the target location, allowing attention to be quickly directed to the target. Such change in attentional guidance is reflected in activation in the superior parietal lobule. Because slow, serial allocation of attention is less needed on repeated displays than on new displays, activation in superior parietal lobule is reduced on repeated displays. The modulation of attention by associative learning suggests Although this study has clarified neural substrates underlying implicit visual learning, it has two limitations that must be overcome in the future. First, as a correlation measure, functional MRI cannot reveal causal relationship between brain function and cognition. Thus, for example, the parietal involvement in the learning task may reflect an outcome of learning, rather than the cause of learning. Complementary methods, such as neuropsychological data collected from brain-damaged patients, will be needed to inform us brain regions needed for visual implicit learning. Second, results from a scene-based learning task may not fully generalize to non scene-based learning tasks. An important difference between scenebased learning and the standard T-among-L search task is the involvement of explicit learning and strategy. The scene-based learning often leads to explicit knowledge of the repetition (Brockmole & Henderson, 2006; King, Shim, & Jiang, 2005, VSS abstract). The underlying mechanism for scene-based learning may be different from the more abstract implicit learning task. CONCLUSION Humans process a visual display more efficiently when they encounter it for a second time, showing implicit learning of visual context. Such learning has high capacity and can proceed without selective attention. It is closely related to spatial navigation and may underlie airplane pilots learning of complex control panels, or airport security personnel s learning of the visual environment they monitor. Delineating individual differences and brain mechanisms of learning will potentially assist the selection of soldiers, enhance the design of human-machine interface, and aid the selection of suitable aircraft or security personnel. 7

8 ACKNOWLEDGEMENT Research reported here was supported by Army Research Office grant LS to Y.V.J. We thank Albert Leung and Sidney Burks for data collection and discussions. Correspondence should be directed to Y.V. Jiang by at REFERENCES Baddeley, A., 1986: Working memory. Oxford, England: Oxford University Press. Brockmole, J.R., and Henderson, J.M., 2006: Using realworld scenes as contextual cues during search. Visual Cognition, 13, Chun, M.M., and Jiang, Y., 1998: Contextual cueing: Implicit learning and memory of visual context guides spatial attention. Cognit. Psychol., 36, Chun, M.M., and Jiang, Y., 2003: Implicit, long-term spatial contextual memory. J. Exp. Psychol. Learn Mem. Cogn., 29, Chun, M.M., and Phelps, E.A., 1999: Memory deficits for implicit contextual information in amnesic subjects with hippocampal damage. Nat.Neurosci.,2, Cohen, N.J., and Eichenbaum, H., 1993: Memory, amnesia, and the hippocampal system. Cambridge, MA: MIT Press. Corbetta, M., and Shulman, G.L., 2002: Control of goaldirected and stimulus-driven attention in the brain. Nat. Rev. Neurosci., 3, Don, A.J., Schellenberg, E.G., Reber, A.S., DiGirolamo, K.M., and Wang, P.P., 2003: Implicit learning in children and adults with Williams syndrome. Dev. Neuropsychol., 23, Epstein, R., and Kanwisher, N., 1998: A cortical representation of the local visual environment. Nature, 392, Gabrieli, J.D.E., 1998: Cognitive neuroscience of human memory. Annu. Rev. Psychol., 49, Grill-Spector, K., and Malach, R., 2001: fmri-adaptation: A tool for studying the functional properties of human cortical neurons. Acta Psychol., 107, Jiang, Y., Olson, I.R., and Chun, M.M., 2000 : Organization of visual short-term memory. J. Exp. Psychol. Learn Mem. Cogn., 26, Jiang, Y., Song, J.H., and Rigas, A., 2005: High-capacity spatial context memory. Psychon. Bull. Rev., 12, Johnson, M.K., 1997: Source monitoring and memory distortion. Philos. Trans. R. Soc. Lond. B. Biol. Sci., 352, Law, J.R., Flanery, M.A., Wirth, S., Yanike, M., Smith, A.C., Frank, L.M., Suzuki, W.A., Brown, E.N., and Stark, C.E., 2005: Functional MRI activity during the gradual acquisition and expression of pairedassociate memory. J. Neurosci, 25, Lleras, A., and von Muhlenen, A., 2004: Spatial context and top-down strategies in visual search. Spat. Vis., 17, Manns, J., and Squire, L.R., 2001: Perceptual learning, awareness, and the hippocampus. Hippocampus, 11, Nissley, H.M., and Schmitter-Edgecombe, M., 2002: Perceptually based implicit learning in severe closehead injury patients. Neuropsychol, 16, Olson, I.R., Page, K., Moore, K.S., Chatterjee, A., Verfaellie, M., 2006: Working memory for conjunctions relies on the medial temporal lobe. J. Neurosci, 26, Reber, A.S., 1993: Implicit learning and tacit knowledge: An essay on the cognitive unconscious. New York: Oxford University Press. Reber, P.J., Gitelman, D.R., Parrish, T.B., Mesulam, M.M., 2003: Dissociating explicit and implicit category knowledge with fmri. J. Cogn. Neurosci., 15, Rose, M., Haider, H., Weiller, C., Buchel, C., 2002: The role of medial temporal lobe structures in implicit learning: An event-related fmri study. Neuron, 36, Saxe, R., Brett, M., Kanwisher, N., 2006: Divide and conquer: A defense of functional localizers. NeuroImage, 30, Schacter, D.L., 1996: Searching for memory: The brain, the mind, and the past. New York: BasicBooks. Schacter, D.L., and Buckner, R.L., 1998: Priming and the brain. Neuron, 20, Shen, Y.J., and Jiang, Y.V., 2006: Interrupted visual searches reveal volatile search memory. J. Exp. Psychol. Hum. Percept. Perform., 32, xxx-xxx. Shepard, R.N., and Metzler, J., 1971: Mental rotation of three-dimensional objects. Science, 171, Song, J.-H., and Jiang, Y., 2005: Connecting the past with the present: How do humans match an incoming visual display with visual memory? J Vis, 5, Summerfield, J.J., Lepsien, J., Gitelman, D.R., Mesulam, M.M., and Nobre, A.C., 2006: Orienting attention based on long-term memory experience. Neuron, 49, Tarr, M.J., and Pinker, S., 1989: Mental rotation and orientation-dependence in shape recognition. Cognit. Psychol, 21, Tulving, E., and Schacter, D.L., 1990: Priming and human memory systems. Science, 247, Vargha-Khadem, F., Gadian, D.G., Watkins, K.E., Connelly, A., Van Paesschen, W., and Mishkin, M., 1997: Differential effects of early hippocampal pathology on episodic and semantic memory. Science, 277, Wolfe, J.M., Klempen, N., and Dahlen, K., 2000: Postattentive vision. J. Exp. Psychol. Hum. Percept. Perform., 26,