Eye movements and the integration of visual memory and visual

Perception amp Psychophysics2005 67 (3) 495-512

The world contains more information than an observercan sample and process at any one moment Physicalproperties of the eye limit the scope of perception byconstricting the proportion of the environment from whichinformation is received and the quality of the informa-tion extracted over different spatial regions Foveal vi-sion corresponding to the center of gaze resolves highspatial frequency and color components of a scene butcovers only 2ordm of the visual world In contrast peripheralvision is tuned to lower spatial frequencies and derivesdegraded color information In addition to these physicallimitations that restrict sampling space cognitive limita-tions on memory and attention set bounds to information-processing capabilities Visual short-term memory (VSTM)is a limited capacity store estimated to hold three to fiveitems (Irwin 1992 Irwin amp Andrews 1996 Luck ampVogel 1997) Although the scope of attention may vary(eg C W Eriksen amp St James 1986) it has been esti-mated to have a maximum span of as little as 1ordm of visual

angle (B A Eriksen amp C W Eriksen 1974) with morediffuse states of attention leading to less refined pro-cessing (Shulman amp Wilson 1987) To view a scene inits entirety then observers shift their attention and theirgaze from place to place

Given the temporal extent of visual processing infor-mation extracted from the world at one moment must beanalyzed in conjunction with that which was obtainedpreviously What mechanism enables this analysis A re-cent proposal is that visual percepts can be integratedwith the contents of VSTM (Brockmole Irwin amp Wang2003 Brockmole amp Wang 2003 Brockmole Wang ampIrwin 2002) That is when an observer generates a rep-resentation of a visual stimulus in memory a subse-quently perceived stimulus can be directly incorporatedinto the existing representation Thus discontinuous butnevertheless related visual information can be repre-sented as a single unit in memory thereby cognitivelylinking once independent pieces of information

Support for the memoryndashpercept integration hypothe-sis has been obtained using a temporal integration para-digm (eg Di Lollo 1980 C W Eriksen amp Collins1967) Two arrays of dots were serially presented withina square grid Together the two arrays filled all but onespace in the grid subjects had to report the unfilled gridlocation As one would expect the time that separatedthe arrays was critical If the delay was very short(50 msec) sensory memory supported the perceptual

495 Copyright 2005 Psychonomic Society Inc

JRB is partially supported by NSF Grant DGE-0114378 DEI issupported by NSF Grant BCS 01-32292 We thank Aaron BenjaminArthur Kramer Brian Ross Frances Wang Robert Logie and twoanonymous reviewers for their insightful comments on this research Wealso thank Nicholas Cassavaugh for technical assistance and ShawnBolin for help in collecting data Address correspondence to J R Brock-mole Department of Psychology Michigan State University East Lan-sing MI 48824 (e-mail jimeyelabmsuedu)

Eye movements and the integration of visualmemory and visual perception

JAMES R BROCKMOLEMichigan State University East Lansing Michigan

and

DAVID E IRWINUniversity of Illinois at Urbana-Champaign Champaign Illinois

Because visual perception has temporal extent temporally discontinuous input must be linked inmemory Recent research has suggested that this may be accomplished by integrating the active con-tents of visual short-term memory (VSTM) with subsequently perceived information In the present ex-periments we explored the relationship between VSTM consolidation and maintenance and eye move-ments in order to discover how attention selects the information that is to be integrated Specificallywe addressed whether stimuli needed to be overtly attended in order to be included in the memory rep-resentation or whether covert attention was sufficient Results demonstrated that in static displays inwhich the to-be-integrated information was presented in the same spatial location VSTM consolidationproceeded independently of the eyes since subjects made few eye movements In dynamic displayshowever in which the to-be-integrated information was presented in different spatial locations eyemovements were directly related to task performance We conclude that these differences are relatedto different encoding strategies In the static display case VSTM was maintained in the same spatial lo-cation as that in which it was generated This could apparently be accomplished with covert deploy-ments of attention In the dynamic case however VSTM was generated in a location that did not over-lap with one of the to-be-integrated percepts In order to ldquomoverdquo the memory trace overt shifts ofattention were required

496 BROCKMOLE AND IRWIN

integration of the arrays and they appeared to be pre-sented simultaneously If the delay was extended to100ndash300 msec however accuracy was poor since thesecond array masked the first In the past this findinghas been taken as evidence that visual integration is im-possible once sensory memory has decayed (see Colt-heart 1980 Phillips 1974) However given that in theabsence of sensory memory successful performance re-quires both the formation and the retention of a longer-lasting representation integration might fail in theseshort delay situations because of a failure to consolidatethe initial representation into a stable and durable for-mat Indeed Brockmole et al (2002) demonstrated thatwhen the delays are longer than 300 msec performancebegins to improve and reaches an asymptotic level be-tween 1000 and 1500 msec The improvement in accu-racy over time was attributed to the generation of a men-tal representation of the lead array because the increasein accuracy was accompanied by a reduction in errors as-sociated with the lead array (ie erroneously selecting aposition previously occupied by the first array as the po-sition of the missing dot) over the same time period Inexcess of 95 of the variance in accuracy improvementwas accounted for by the decrease in Array 1 error (Brock-mole et al 2002)

Why postulate that the two arrays were integrated intoa single representation in VSTM when intuitively thereare several strategies that an observer could use to solvethe missing dot task For example the first array mightbe used to visually mark (see Watson amp Humphreys2000) grid positions as locations that cannot constitutethe correct answer As a result these positions might beinhibited and attention directed to the positions thatArray 1 left empty enhancing the detection of the spaceleft unfilled by the second array (see Jiang amp Kumar2004 Jiang Kumar amp Vickery 2005) Another possi-bility is that two separate memory representations ofeach stimulus array may be created in memory If the twomemory traces could then be compared ldquoin the mindrsquoseyerdquo the combination or integration of two stimulus ar-rays need not be postulated We investigated and rejectedthese possible explanations in two recent studies de-scribed next

To address the visual-marking hypothesis a directmeasure of the locus of spatial attention during the tem-poral integration task was used (Brockmole et al 2003)During the interstimulus interval (ISI) separating the ar-rays two-alternative forced choice probes appeared inpositions either filled by or left empty by the first arrayProbes that occupied positions filled by the first arrayenjoyed a processing benefit in that they were respondedto more quickly than probes that were in positions leftempty by the first array Thus attention was selectivelyallocated to the positions filled by Array 1 during thetime between arrays no evidence for Array 1 inhibitionor Array 2 priming was found In a separate study (Brock-mole et al 2002) the two-representation hypothesis wasassessed by examining the time required to consolidate

the second array into VSTM Although 1000ndash1500 msecwere required to consolidate Array 1 into VSTM reac-tion time analyses demonstrated that the time needed toprocess the second array at long ISIs was about equal tothat needed during perceptual integration (when the ISIwas 0 msec) suggesting that Array 2 is not submitted tothe same consolidation processes as Array 1 Because at-tention is allocated to the first array throughout the delayseparating the arrays and because the second array doesnot undergo a slow consolidation process it was con-cluded that over a period of 1000ndash1500 msec observerscan generate a VSTM representation of one stimulus anddirectly integrate subsequent perceptual information withthat representation

In other experiments the flexibility or abstractnessof the VSTM representations involved in integration wasconsidered by examining whether integration can occurwhen the to-be-integrated stimuli do not match in termsof their spatial properties (Brockmole amp Wang 2003)Stimuli can be integrated even if they are presented indifferent size scales and orientations This result indi-cates that during integration attention is not dedicated toparticular areas of space independently of the stimulithat occupied them The memory representation can bemodified to account for changes in stimulus structureand ldquoprojectedrdquo to different locations on the display Ex-pressed another way the memory trace of the lead stim-ulus is not spatiotopically or retinotopically specificrather observers are able to retain their memory for thelead array abstractly enabling it to be transformed to ac-commodate spatial changes thereby providing spatialflexibility in the integration process

An important question remaining concerns the timecourse of integration Specifically why does optimal in-tegration require up to 1500 msec of separation betweenvisual stimuli We assume that this time reflects thespeed with which the lead stimulus is consolidated intoVSTM In this article we focus on uncovering the mech-anisms that underlie this consolidation process by ex-amining observersrsquo eye movements during integrationSpecifically we ask whether overt shifts of attention tothe to-be-remembered elements are required for integra-tion That is in order to be included in the VSTM repre-sentation must a stimulus elementmdashor in the case of abrief presentation its locationmdashbe overtly attended orfixated In essence the issue is one of whether or not theinitial stimulus is maintained by eye movements that re-hearse its spatial structure On the basis of temporalcharacteristics of eye movements and assuming an aver-age fixation duration of 200 msec and an average sac-cade duration of 25 msec it would take 1575 msec toindividually fixate seven elements a time course match-ing that in integration experiments requiring the mainte-nance of seven items If the ISI is less than this valuefewer elements can be overtly attended or rehearsed andintegration will suffer

The idea that eye movements are important to integra-tion stems from research in which the use of eye move-

EYE MOVEMENTS AND MEMORYndashPERCEPT INTEGRATION 497

ments during perception has been compared with themaintenance or recall of information from VSTM In histheoretical work on cell assemblies Hebb (1968) sug-gested that visual memory (ie imagery) may constitutea top-down reinstantiation of perceptual processes Thusjust as eye movements are necessary to sequentially viewdifferent parts of a visual stimulus so too may they benecessary to sequentially assemble image parts to con-struct complete VSTM representations of those samestimuli Specifically he argued that eye movements mayserve an ldquoorganizing functionrdquo in both visual perceptionand visual memory (see Neisser 1967 for similar argu-ments) The recent interest in eye movements and visualmemory has led to a variety of empirical papers in whichHebbrsquos proposal has been tested over the past few years

Brandt and Stark (1997) had observers study randompatterns of black-and-white squares arranged in a matrixand to later maintain the patterns in visual memory aftertheir disappearance Systematic scanpaths were foundduring both the study and the memory maintenance phasesof the experiment in that the eyes were directed to thedark or ldquofilledrdquo regions of the matrix For each stimu-lus the eye movements during the memory phase weremore closely related to the eye movements made duringthe study of that stimulus than to those made during thestudy of other stimuli Spivey and Geng (2001) notedthat while observers answered questions about objectsfrom a since-removed visual display the eyes were spon-taneously directed to the region of space once housingthe object Finally Laeng and Teodorescu (2002) ma-nipulated an observerrsquos ability to make eye movementsduring the study and recall of visual stimuli Subjectswho maintained a single fixation during the perceptionphase also maintained a single fixation during the recallphase Observers who were allowed to move their eyesfreely during the perceptual phase moved their eyes dur-ing the recall phase in systematic ways The percentageof time fixating any one location during perception wascorrelated with the percentage of time that region ofspace was fixated during the recall phase as was theorder in which locations were fixated during the percep-tual and recall phases The strength of these relationshipspredicted performance accuracy Together these studiessuggest that eye movements during the maintenance re-call and processing of previously viewed visual stimuliare not random but reflect the content of the visualizedscene in the same way that eye movements during sceneviewing are related to the scenersquos content (Buswell 1935Yarbus 1967) That is eye scanpaths engaged by visualmemory recapitulate those during perceptionmdashthe in-terrogation of VSTM is accompanied by an oculomotorinspection of space The functional role of these eyemovements may be to spatially link mental and retinalimages

The research above suggests that processes that callon memory to recreate the visual world operate similarlyto processes that call on direct perception of the worldAs such the success of memoryndashpercept integration may

be closely linked to eye movement behavior Thereforedetermining the relationship between eye movementsand the integration of visual memories and percepts isimportant for a complete understanding of integration asa mechanism that links temporally distinct but concep-tually related information in VSTM If f ixation on astimulus element is required for integration integrationcan take place only between stimuli that enjoy overt se-lection and processing This selection process would re-strict the possible aspects of a scene that would be linkedin VSTM and elucidate a circumstance necessary for in-formation to be linked in VSTM On the other hand in-dependence between eye movements and integrationperformance would suggest that covert attention alone issufficient for integration to occur A specific link be-tween covert attention shifts and the consolidation ofVSTM representations has not been established becausethe vast majority of experiments on memory consolida-tion have not monitored eye movements rendering it im-possible to dissociate covert and overt attention shiftsIndeed Neisser (1967) has pointed out that the role ofcovert attention in memory processes cannot be dis-missed because even during object perception shifts ofattention can occur without ocular motion A primarilycovert selection process would give greater cognitivecontrol over integration to the observer since he or shewould be able to integrate any attended information notjust that overtly scanned enabling greater flexibility inthe integration process

EXPERIMENT 1

There are two general means by which attention canbe deployed about a scene First attention can be as-signed to locations or objects covertlymdashthat is withoutthe aid of eye movements This kind of attention alloca-tion can be likened to attending to something ldquoout of thecorner of your eyerdquo and enables an observer to deploy at-tention to many locations or objects in parallel Alterna-tively attention can be assigned to objects or locationsovertlymdashthat is by executing eye movements to bringthe object of attention to the fovea Overtly guided at-tention allocation brings objects or locations into thescope of attention serially as eye movements seriallymove from one locus of fixation to another Overt atten-tion may also provide a rehearsal strategy Subjects maysystematically scan the grid focusing their fixations onlocations previously occupied by a dot from Array 1

The purpose of Experiment 1 was to determine whichmode of attention deployment is used to commit the to-be-remembered grid locations to memory in preparationfor integration with a subsequent stimulus In order toassess whether attention is deployed covertly or overtlyduring the interval separating the to-be-integrated ar-rays eye movement behavior was recorded while sub-jects engaged in a temporal integration task If attentionis allocated to the to-be-remembered grid locationscovertly performance should not share a strong rela-


tionship with eye behavior Alternatively if attention isallocated overtly specific patterns of eye behavior shouldbe observed First as the ISI between arrays increasesso too should the number of eye movements made dur-ing the delay since the subjects would begin to seriallyscan the display Second the increasing number of eyemovements should strongly correlate with an increase inthe number of grid locations that are fixated That is ifthe subjects attempt to fixate as many Array 1 positionsas possible each eye movement should be directed to anew grid location with few intragrid position fixationsor intergrid position revisitations Third on the basis ofBrockmole et al (2003) the eye movements that aremade should also be more likely to bring the eyes to fix-ation on a grid location that was occupied by a dot fromthe first array Finally some researchers have argued thatfixation durations reflect the amount of processing thatis needed at that location (eg Henderson Weeks ampHollingworth 1999 Just amp Carpenter 1980 Rayner1998) Assuming that each eye movement is used to as-sign attention to a new location the fixation durationsassociated with each fixation should either remain con-stant because equal processing time should be neededat each fixation or potentially increase because eachfixation would add more information to the maintainedmemory trace thereby increasing processing demands

MethodSubjects

Eight members of the University of Illinois community partici-pated after providing informed consent All the subjects were naivewith respect to the experimental hypotheses and were paid $6 forparticipating

StimuliThe stimuli were very similar to those used in Brockmole et al

(2002) Two unique dot arrays were presented within an enclosed4 4 square grid The first array contained seven dots and the sec-ond array contained eight dots Together the arrays filled all butone square in the grid The grid was composed of interconnectedlines drawn over the background in such a way that the color withinthe grid spaces and the area surrounding the grid was the same Thesubjects viewed the stimuli at a viewing distance of 57 cm The totaldisplay subtended 41ordm horizontally and 31ordm vertically The squaregrid subtended 20ordm Each square within the grid subtended 5ordm Eachdot presented in the array subtended 45ordm The display backgroundwas light gray the grid lines were light blue and the dots were black

ApparatusThe stimuli were presented at a refresh rate of 60 Hz on a Hitachi

Superscan Elite 21 monitor A Gateway P5-150 microcomputercontrolled stimulus presentation and recorded responses Eye posi-tion was sampled at a rate of 250 Hz (every 4 msec) with an Eye-link eyetracker (Senso-Motoric Instruments Inc) and was recordedon an IBM microcomputer This system used video-based infraredoculography to measure eye and head positions A chinrest wasused to help stabilize the head and to ensure constant distance fromthe display

Design and ProcedureIn general the design and procedure were very similar to those

used in Brockmole et al (2002) On each trial two dot arrays (sevenand eight dots respectively) were presented sequentially within anenclosed square grid separated by a variable ISI On any given trialone position within the grid was never filled and the subjects wereinstructed that they were to identify the position of the missing dot

The trials were divided into blocks on the basis of ISI The orderof blocks was randomized and the subjects were informed of theISI duration prior to the start of each block A block of trials beganwith a calibration routine in which a fixation point was presentedserially in eight locations around the perimeter of the display aswell as one location in the center of the display The subjectsrsquo eyepositions were monitored during this procedure which served tocalibrate the output of the eyetracker against spatial position Fol-lowing successful calibration the subjects were informed of the ISIduration that would be used for the trials in that block The subjectsindicated their readiness to begin the trials with a keypress

The procedure is illustrated in Figure 1 At the start of each trialan empty 4 4 square grid was presented in the center of the dis-play When ready to see the first array the subjects pressed thespace bar Immediately the first array of seven dots was presentedwithin the grid for 33 msec The ISI between the offset of the firstdot array and the onset of the second array was 0 100 500 10001500 2000 2500 or 3000 msec during which the empty grid re-mained on the screen Following the delay the second array con-taining eight dots was presented for 33 msec After both arrays hadbeen presented the subjects used a number keypad to indicate therow and column coordinates corresponding to the grid position theythought was not filled on that trial The subjects were told to re-spond as accurately as possible and that they were under no speedstress

Each ISI (eight) occurred equally often With a repetition factorof 32 each subject completed a total of 256 trials divided intoblocks of 8 trials The order of blocks was randomized Prior to be-ginning the experimental trials the subjects completed 32 practicetrials which consisted of 8 trials each at ISIs of 0 100 750 and1500 msec During practice the subjects were given feedback con-cerning the accuracy of their responses However during the ex-perimental trials no feedback was given

Figure 1 Schematic illustration of the procedure in Experiment 1 The sub-jects were instructed to locate the one grid space that was not filled with a dot


The eyetracker monitored eye position during the presentation ofArray 1 the ISI and the presentation of Array 2 The subjects beganeach trial by looking at the center of the grid but were free to movetheir eyes without constraint during the trial Eye position wasscored in terms of the grid space that was fixated Saccades wereoperationally defined as a change in fixation that exceeded 02ordm ofvisual angle accompanied by a velocity exceeding 30ordmsec or an ac-celeration exceeding 8000ordmsec2 that was maintained for a mini-mum of 4 msec The saccade was considered to be terminated whenthese criteria were no longer met

Results

The results will be reported in two sections First gen-eral accuracy and error rates for the integration task wereassessed to establish that the present experiment repli-cated Brockmole et alrsquos (2002) result that as the ISI wasincreased beyond 100 msec accuracy improved Thisreplication would establish the appropriateness of inter-preting the eye movement record in relationship to priorresults Second various aspects of eye movement be-havior were examined with the goal of linking particularpatterns of fixation to performance Variables of interestincluded number of fixations location of fixations andfixation durations

Integration Task Accuracy and ErrorA response was classified as correct an Array 1 error

(erroneously selecting a position occupied by the firstarray) or an Array 2 error (erroneously selecting a posi-tion occupied by the second array) and were measured interms of the percentage of trials on which they occurredThe results are illustrated in Figure 2 A one-way re-peated measures analysis of variance with ISI as the in-dependent factor demonstrated that accuracy varied as afunction of ISI [F(749) 242 MSe 955] Consider-ing those ISIs greater than or equal to 100 msec singledegree of freedom polynomial tests indicated a reliablepositive slope linear trend [F(17) 206 MSe 233]and a quadratic trend [F(17) 144 MSe 114] in theaccuracy data These trends indicate that accuracy in-creased with ISI in a curvilinear fashion Howeverpost hoc contrasts showed that accuracy never trulyreached an asymptotic level attaining a maximum of61 in the data space measured This level of perfor-mance is comparable to the asymptotic level observed byBrockmole et al (2002) Array 1 errors occurred on av-erage on 40 of the trials and accounted for 82 of allthe errors The Pearson correlation between Array 1 er-rors and correct responses across ISIs was 97 reveal-ing almost perfect dependence between accuracy andArray 1 error Array 2 errors however as a percentage oftotal trials occurred on average on 9 of the trials andaccounted for 18 of all the errors The Pearson corre-lation between Array 2 errors and correct responses acrossISIs was 26 These overall patterns of results closelyparallel those observed by Brockmole et al (2002)

Eye Movement BehaviorThe locus of overt attention is indicated by the loca-

tion in space that is fixated Thus an analysis of patterns

in the distribution of fixations during the ISI separatingthe arrays can give insight into the manner in which overtattention is used during VSTM consolidation In thissection fixations are analyzed according to their num-ber location and duration as a function of ISI

Number If the locations of the dots in Array 1 aremaintained by overtly attending to each of the locationsthat once housed an Array 1 dot the number of fixationsobserved on a particular trial should be positively corre-lated with ISI since longer ISIs afford more time tomake more eye movements Indeed as Figure 3 illus-trates that is exactly the pattern that was observed ThePearson correlation between ISI and number of fixationswas 75 A linear regression analysis indicated that 15additional fixations were made for every 1000-msec in-crease in ISI [F(162) 807 MSe 190]

Location Three separate analyses were performed toconsider the location of fixations within the grid in orderto determine the locus of overt attention during integra-tion In these analyses the absolute number of grid po-sitions fixated the distribution of fixations to Array 1and Array 2 grid positions and fluctuations in the distri-bution of fixations on the 16 grid positions over timewere considered

First if the subjects attempted to fixate as many gridpositions previously occupied by a dot in Array 1 as pos-sible given the available ISI the number of grid loca-tions that were f ixated should also increase with in-creases in ISI More specifically if each eye movementcarried fixation from one grid position to another therate of this increase should exactly match the rate atwhich the number of fixations increased as a function ofISI The extent to which multiple fixations occur withina single grid positionmdashwhether caused by intraposition

Figure 2 Accuracy and error results from Experiment 1


refixation or interposition revisitationmdashis indicated bythe discrepancy between the rates at which the number offixations and the number of grid positions fixated in-crease with increases in ISI Figure 3 illustrates that al-though the number of grid positions fixated did increasewith increases in ISI the slope of this increasing func-tion was 05 items for every 1000-msec increase in ISI[F(162) 516 MSe 0314] This rate is one third thatof the increase in the number of fixations observed in-dicating that one out of every three fixations was on a unique grid position This disparity is striking For example assuming an average f ixation duration of300ndash500 msec (Rayner 1998) between 7 and 11 gridpositions could have been f ixated at the longest ISI(3000 msec) enough to fixate the location of each dotindividually However on average only 25 grid posi-tions were actually fixated

Second because the number of grid positions fixateddid increase with increases in ISI another analysis wasperformed to examine where those fixations occurredThe results of Brockmole et al (2003) showed that at-tention is deployed to Array 1 positions If the unique po-sitions that were fixated in the present experiment weremade to overtly guide attention systematically to thesepositions a plurality of fixations should rest on grid lo-cations that were occupied by a dot from the first arrayBecause Array 1 contained seven dots the probability offixating an Array 1 position randomly was 44 Figure 4illustrates the percentage of fixations on a grid positionthat was occupied by an Array 1 dot as a function of ISI

Although a reliable but nonsystematic effect of ISIwas observed [F(535) 411 MSe 432] 95 con-

fidence intervals constructed around each data point (il-lustrated by the error bars in Figure 4) revealed that thepercentage of fixations on Array 1 positions (the blackline in Figure 4) did not reliably differ from chance (thedashed line in Figure 4) at any ISI with the exception of1500 msec A subsequent analysis also demonstratedthat the percentage of fixations on Array 1 positions didnot vary as a function of fixation number In sum the

Figure 3 Number of total fixations versus the number of fixationsmade in unique grid locations in Experiment 1

Figure 4 Percentage of fixations on positions originally occu-pied by a dot from Array 1 in Experiment 1


percentage of fixations on Array 1 positions was at chanceand did not vary by ISI or fixation number

Third the location of fixations was analyzed accord-ing to the frequency with which each grid position wasfixated conditionalized on fixation number The subjectswere instructed to fixate the center of the grid at the startof the trial but were permitted to move their eyes freelyonce the trial began As a result the initial fixation in thegrid would be in one of the four center boxes If the sub-jects scanned the grid each subsequent fixation shouldbe more likely to occupy a position along the peripheryof the grid The percentage of total fixations across sub-jects that coincided with each grid position is illustratedin Figure 5 by fixation number Note that relatively fewfixations fell on peripheral grid locations even after sixfixations on the grid

In an effort to characterize the preferences that thesubjects displayed for fixating various grid positions thepercentage of fixations that occurred in the center fourgrid positions where the first fixation occurred was cal-culated for each fixation The solid line in Figure 6 illus-trates the discrete percentage of trials on which the centerfour grid positions were fixated as a function of fixationnumber The dashed line in Figure 6 illustrates the cu-mulative percentage of fixations that occurred in thecenter four grid positions This figure depicts a strongbias for the subjects to maintain fixation in the center re-gion of the grid Through the first six fixations 84 ofall the fixations occurred in the center four grid posi-tions Considering the sixth fixation alone this fixationfell in the center four grid positions on 60 of the trials

Duration Finally fixation durations were analyzedas a function of fixation number Fixation times havebeen considered to reflect the amount of processing thatmust be completed at a particular location in a visualscene on the basis of the assumption that a fixation doesnot end until processing has been completed (Hendersonet al 1999 Just amp Carpenter 1980 Rayner 1998 butsee Irwin 2004 for an alternative view) If each fixationa subject makes during VSTM consolidation and main-tenance is used equally to process the displaymdashin thiscase to allocate attention to particular components ofthe gridmdashfixation times should not vary as a function offixation number That is the information processing atFixation 1 should be at least equal to the processing atFixation 2 and so on Mean fixation durations are illus-trated in Figure 7 by fixation number (1ndash6) and ISI Anoverall main effect of ISI was observed since fixationdurations generally increased with increases in ISI Atlonger ISIs the subjects may have felt less ldquohurriedrdquo tomake their eye movements thus causing this generalslowdown More important a main effect of saccadenumber was observed In general f ixation durationswere shorter for later saccades This suggests that pro-gressively less processing occurred during later fixationsindicating less processing at each subsequent fixation

DiscussionThe fact that only one third of all the fixations oc-

curred in unique grid positions that on average only 25grid positions were f ixated even when the ISI was3000 msec that no bias for fixating Array 1 positions

05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34

Fixation 1 Fixation 2 Fixation 3


Figure 5 Percentage of all fixations within each grid space by fixation number (through 6) in Ex-periment 1 Darker shades indicate greater percentages of fixations


existed that the subjects displayed little tendency to fix-ate grid positions along the periphery of the grid andthat successive fixations in a trial exhibited shorter la-tencies strongly suggests that the subjects did not sys-tematically scan the grid during the ISI separating the ar-rays Instead it seems that quite the opposite was trueThe subjects tried to move their eyes as little as possibleThus it appears that the overt orienting of attention isnot required for VSTM consolidation and rehearsal andthat the minimum amount of temporal separation be-tween the to-be-integrated arrays that allows for optimalintegration performance is not determined by a mini-mum number of saccades required to drive overt atten-

tion to all of the elements of Array 1 Rather the ISI sep-arating the arrays appears to be used to assign covert at-tention to the grid locations maintained in memory

The fact that memoryndashpercept integration can occurwithout the deployment of overt attention does not how-ever necessarily indicate that the integration system iscompletely independent of eye movements Considertwo extreme positions that illustrate very different pos-sible interactions between integration and eye move-ments On the one hand it may indeed be that integrationis completely independent of eye movements Becausethe eyes do not need to fixate to-be-remembered infor-mation it is therefore possible that they may be free toscan aspects of the visual scene that are not ultimately in-cluded in the maintained memory representation Thiswould add some flexibility to the integration systemsince various aspects of a display could be attended towithout being integrated On the other hand integrationmay be completely dependent on eye movementsmdashmoreprecisely on a lack of eye movements Perhaps the eyesdo not move much during the VSTM consolidation pe-riod because eye movements would force attention to beshifted around the visual display perhaps overridingcovert attention shifts and thereby disrupting the inte-gration process Another possibility is that overt eyemovements to locations in the visual field draw attentionaway from memory In other words integration may bedependent on the eyesrsquo not moving since movementsmay cause interference Indeed eye movements havebeen shown to disrupt the retention of sequentially pre-sented visual stimuli as compared with stimulus-drivencovert shifts of attention of equal spatial magnitude (Pear-son amp Sahraie 2003) Tacit knowledge of this disruptionmight lead subjects to keep their eyes still when viewingthe sequential arrays presented in the temporal integra-tion paradigm Experiment 2 assessed integration when

Figure 6 The percentage of trials on which the first six fixa-tions were on the center four positions of the grid (dotted line) aswell as the cumulative number of fixations on those positionsacross fixation number (solid line) in Experiment 1

Figure 7 Durations of the first six fixations as a function of interstimulus intervalin Experiment 1


subjects were required to move their eyes to visually ac-quire the information necessary for integration a ma-nipulation that distinguished between these possibilities

EXPERIMENT 2

Experiment 2 tested memoryndashpercept integration incases in which the to-be-remembered stimulus and theto-be-integrated stimulus were not presented in the samespatial location rather the first and the second arrayswere separated by 20ordm of visual angle Although eyemovements played little or no role during integrationwhen the first and the second arrays were presented inthe same grid in Experiment 1 (where no eye movementswere required to view both arrays) in Experiment 2 eyemovements were critical to integration insofar as a sac-cade was required to shift fixation from one grid to theother That is Experiment 2 required a lengthy (in termsof spatial displacement) shift of overt attention This re-quirement may alter the manner in which eye movementsinteract with VSTM consolidation and maintenance Thecritical question in Experiment 2 then is the followingDoes the need to make eye movements to obtain the rel-evant information for integration influence the role eyemovements play in the generation and maintenance ofthe memory representation of Array 1

Several outcomes are possible First the overt shift ofattention may not alter the eye movement patterns (orlack thereof ) in Experiment 1 if covert attention can re-assemble the first array after the eyes land on the secondarray Second if covert attention cannot reassemble theinitial stimulus in the new spatial location evidence foran overt selection process may result Third the overtshift of attention from one grid to the other may interferewith integration generally if eye movements are detri-mental to the maintenance of VSTM representations inthis task In previous integration studies in which the to-be-integrated arrays did not spatially align the mis-alignment was created by either scaling the arrays or ro-tating them Both arrays however were anchored to thesame point in space and were presented so that bothcould be viewed in a single fixation (Brockmole amp Wang2003) Indeed research on transsaccadic memory in the1980s indicated that very little visual information is car-ried across a saccade and that integration of pattern in-formation is not possible when a saccade intervenes be-tween stimulus presentations (Bridgeman amp Mayer 1983Henderson 1997 Irwin Brown amp Sun 1988 IrwinYantis amp Jonides 1983 McConkie 1991 McConkie ampRayner 1976 OrsquoRegan amp Levy-Schoen 1983 Rayneramp Pollatsek 1983) However these studies assessed in-tegration abilities only at very short ISIs (a few hundredof milliseconds) and on the basis of our current knowl-edge of memoryndashpercept integration much more timemay be needed As such a secondary question addressedin Experiment 2 is whether or not integration is evenpossible if the arrays are presented in separate fixations

MethodSubjects

Twenty members of the University of Illinois community partic-ipated after providing informed consent All the subjects were naivewith respect to the experimental hypotheses and were paid $8 forparticipating

StimuliThe stimuli were the same as those in Experiment 1 save that the

first array was presented approximately 9ordm of visual angle to the leftof the center of the display and the second array was presented 9ordmof visual angle to the right of the center of the display No portionof the left and right grids overlapped in space the rightmost bound-ary of the left grid and the leftmost boundary of the right grid wereseparated by 5ordm the equivalent of one grid space

ApparatusThe same apparatus as that in Experiment 1 was used

Design and ProcedureEach trial consisted of two sequentially presented dot arrays

within enclosed square grids separated by a variable ISI During theISI separating the arrays the grid changed location The first andthe second arrays together filled all but one position in the gridUnder the constraint that no members of the two arrays could oc-cupy the same position in the grid the dot patterns were randomlygenerated on each trial Thus the location of the missing dot wasequally likely to occur in each grid position The subjects were in-structed to identify the position of the missing dot

There are two time intervals that may be critical to memoryndashperceptintegration First the time that elapses between the offset of the firstarray and the displacement of the grid may be important for ex-tracting information from the initial array required to generate anaccurate memory representation If the grid changes location tooquickly so that the first array cannot be adequately encoded littleinformation may be retained from which to generate a memorytrace Second the time that elapses between the displacement of thegrid and the onset of the second array may be important for pro-cessing the information committed to memory in the new spatiallocation Each of these time components which will be referred toas early components and late components respectively was inde-pendently manipulated in Experiment 2 to determine how muchtime was necessary before and after grid displacement to produceoptimal performance In addition this manipulation allowed the re-lationship between eye movements and memory consolidation bothprior to and following the displacement of the grid to be examinedThat is depending on the duration of the early and late componentsthe VSTM representation had to be formed before the displacementof the grid (eg long early component short late component) orfollowing the displacement of the grid (short early component longlate component)

Two conditions were created that contrasted the contributions ofthe time available for processing prior to grid displacement and thetime afforded for processing following grid displacement (see Fig-ure 8) In the early component condition performance was exam-ined as a function of the time available prior to grid displacementThis was accomplished by varying the time that elapsed betweenthe offset of the first array and the presentation of the second spa-tially displaced grid This will be referred to as the predisplacementdelay In order to examine performance as a function of the predis-placement delay this condition severely limited the amount of timeavailable for postdisplacement processing by holding it constant at100 msec That is after the second grid had been presented it wasfilled with a dot array 100 msec later This time allowed the subjectsto recover from any saccadic suppression which can last up to100 msec postsaccade (Volkmann Schick amp Riggs 1968) associ-


ated with moving the eyes from the left to the right side of the dis-play and to acquire a new fixation Unless postdisplacement pro-cesses require less than 100 msec this time course does not how-ever allow such processes to be completed prior to the onset of thesecond array As a result performance in this condition shouldlargely reflect the time course of predisplacement processing andshould then reveal how well the arrays can be integrated given onlythose processes at work

In the late component condition on the other hand performancewas examined as a function of the time available for processingafter grid displacement This was accomplished by varying the timethat elapsed between the displacement of the grid and the presenta-tion of the second array This will be referred to as the postdis-placement delay As opposed to the early component conditionhowever it was essential to allow early processing to be optimallycompleted prior to grid displacement If predisplacement process-ing is not complete either it will be continued postdisplacement orit will not be completed at all which may reduce general perfor-mance Thus if predisplacement processing is complete trends inperformance as a function of postdisplacement time should largely re-flect the contribution of postdisplacement processing on memoryndashpercept integration

The early and late component conditions constituted a between-subjects factor The early component condition was completed priorto the late component condition In both conditions prior to theonset of each trial an empty 4 4 square grid was presented on theleft side of the display When ready to begin the trial the subjectspressed the space bar Immediately the first array of seven dots waspresented within the grid for 33 msec The events following the off-set of the dot array depended on the condition (early component vslate component) of the trial

For the subjects in the early component condition followingArray 1 offset the grid remained on the left side of the display (thepredisplacement delay) for 0 33 67 100 300 500 700 1500 or3500 msec whereupon it was erased Following grid erasure thedisplay remained blank for 300 msec The (empty) second grid thenappeared on the right side of the display for 100 msec The seconddot array was then presented for 33 msec Thus the ISI separatingthe arrays was 400 433 467 500 700 900 1100 1900 or3900 msec For the subjects in the late component condition fol-lowing Array 1 offset the grid remained present on the left side ofthe display for 150 msec whereupon it was erased (the determina-tion of this time will be discussed in detail in the Results section)This was followed by a 300-msec interval during which the displaywas blank The second grid then appeared on the right side of thedisplay followed by a variable delay of 100 200 300 500 750

1000 1500 2000 or 3500 msec (the postdisplacement delay)Finally the second dot array was presented for 33 msec Thus theISI separating the arrays was 550 650 750 950 1200 14501950 2450 or 3950 msec In both conditions the 300-msec blankinterval was inserted between the offset of the first grid and theonset of the second grid in order to avoid any potential backwardlateral or object substitution masking effects which are known todissipate within 300 msec (Averbach amp Coriell 1961 Breitmeyer1980 1984 Breitmeyer amp Ganz 1976 Enns amp Di Lollo 1997Kahneman 1968 Lefton 1973 Matin 1975 Turvey 1973)

The trials were divided into blocks on the basis of the pre- orpostdisplacement delay The subjects were informed of the delayprior to the start of each block and were told to move their eyes awayfrom the first array as soon as it disappeared Thus the disappear-ance of the first grid provided a cue for the subjects to shift theirgaze across the display The subjects completed 252 trials dividedinto blocks of 7 Each of the nine delays occurred equally often Foreach subject the blocks of trials were presented in a different ran-dom order Prior to beginning the experimental trials the subjectscompleted two practice sessions to familiarize themselves with theexperimental procedure In the first session the subjects completed32 trials in which both the first and the second grids were presentedin the same spatial location This session consisted of 8 trials eachat ISIs of 0 100 750 and 1500 msec In the second session thesubjects completed 28 trials identical to the experimental trials inthe condition in which they participated For the subjects in theearly component condition this session consisted of 7 trials at pre-displacement delays of 33 300 700 and 1000 msec each For thesubjects in the late component condition this session consisted of7 trials at postdisplacement delays of 33 300 700 and 1000 mseceach During the practice sessions the subjects were given feed-back concerning the accuracy of their responses However duringthe experimental trials no feedback was given

Results

The results will be reported in several sections Firstthe behavioral results of the early component conditionwill be discussed then the behavioral results of the latecomponent condition will be discussed The focus ofthese analyses was on trends in accuracy error and eyemovement behavior as a function of the delay associatedwith each condition That is performance in the earlycomponent condition was assessed in terms of the pre-displacement delay and performance in the late compo-

Figure 8 Schematic illustration of trial events for the early and late component conditions in Experiment 2


nent condition was assessed in terms of the postdis-placement delay The accuracy and error data for eachcondition are illustrated in Figures 9 and 10 Second eyemovement behavior in both the early and the late com-ponent conditions was assessed as it was in Experiment 1with a focus on fixation number location and durationThese data are illustrated in Figures 11 and 12

Data TrimPrior to data analysis the trials were required to meet

several selection criteria based on eye movement behav-ior First the eyes were required to be fixated on the leftgrid for a total duration at least equal to the duration ofthe predisplacement delay for that trial Second the eyeswere required to have acquired fixation on the right gridprior to the onset of the second array These timing re-quirements ensured that the subjects executed a saccadeacross the display from the left grid to the right grid dur-ing the 300-msec blank interval separating grid presen-tations Therefore the subjects viewed each grid for aperiod at least equal to that of the pre- or postdisplace-ment delay for the trial and the subjects were looking atboth the first and the second arrays when they were pre-sented In essence this trim ensured that the subjects in-deed moved their eyes away from the first grid as soonas it disappeared In the early component condition thistrim excluded 27 of the trials The majority of the tri-als excluded were those for which the delay was 100 msecor less where approximately 40 of the trials were ex-cluded The exclusion rate fell to approximately 20 ofthe trials when the delay was greater than 100 msec Inthe late component condition 25 of the trials were ex-cluded The trim did not differentially exclude trials ateach delay This trim did result however in empty cells

for 1 subject and that subjectrsquos data are not included inthe analyses

Integration PerformanceEarly component condition A one-way repeated

measures analysis of variance with predisplacement delayas the independent factor demonstrated that accuracyvaried as a function of this delay [F(872) 420 MSe 732] Single degree of freedom polynomial tests indi-cated a reliable positive slope linear trend [F(19) 105 MSe 1147] in the accuracy data Planned com-parisons showed that accuracy improved from 30 to41 as the predisplacement delay increased through100 msec (500-msec ISI) Surprisingly accuracy thendecreased through 500 msec falling to 34 before re-covering to 42 by 1500 msec Statistically perfor-mance at 100 msec was equal to performance when thepredisplacement delay was 1500 msec or greater Thetemporary decrease in accuracy when the predisplace-ment delay was approximately 500 msec was examinedin depth in a separate experiment and was found to betheoretically uninteresting since it was the result of non-systematic variation or Type II statistical error1 There-fore for purposes of the following analyses it is simplyobserved that the earliest point at which performancereached a maximum level coincided with a predisplace-ment delay of approximately 100 msec

In addition the source of errors was assessed with anemphasis on the strength of relationship between differ-ent types of error and accuracy As a percentage of totaltrials Array 1 errors occurred on average on 49 of thetrials and accounted for 74 of all the errors The Pear-son correlation between Array 1 errors and correct re-sponses across ISIs was 84 revealing a very high level

Figure 9 Accuracy and error results from the early component con-dition in Experiment 2


of dependence between accuracy and Array 1 error Thismeans that the improvement in accuracy was heavily dueto the formation of a more complete memory represen-tation of the first array Array 2 errors however as a per-centage of total trials occurred on average on 17 ofthe trials and accounted for 26 of all the errors ThePearson correlation between Array 2 errors and correctresponses across ISIs was 32 revealing relative inde-pendence between accuracy and Array 2 error

Late component condition The aim of the late com-ponent condition was to measure performance as a func-tion of time available for postdisplacement processinggiven that the predisplacement delay was long enoughfor all predisplacement processing to be completed Tokeep overall ISI durations reasonable ideally the timebetween the offset of the first array and the removal ofthe grid in the late component condition should be theshortest duration possible that still allows all predis-placement processes to be completed On the basis of theresults of the early component condition this duration is100 msec (the time after which performance in the earlycomponent condition reached maximum) However be-cause different subjects completed the late componentcondition and because there was certainly individualvariance around this 100-msec estimate the predisplace-ment time in the late component condition was increasedto 150 msec to be more certain that enough time wasprovided for predisplacement processing

A one-way repeated measures analysis of variancewith postdisplacement delay as the independent factordemonstrated that accuracy varied as a function of thisdelay [F(864) 252 MSe 2080] Single degree offreedom polynomial tests indicated a reliable positiveslope linear trend [F(18) 426 MSe 5750] in theaccuracy data The quadratic trend was not reliable How-

ever planned comparisons showed that accuracy im-proved from 40 to 60 as the postdisplacement delayincreased through 750 msec (1200-msec ISI) but did notdiffer thereafter averaging 57 A linear spline regres-sion was performed on these data to interpolate the pointat which accuracy reached a ceiling level between the se-lected delay values (see Brockmole et al 2002) Accuracyimproved until the postdisplacement delay was 650 msec(1100-msec ISI)

An error analysis showed that as a percentage of tri-als Array 1 errors occurred on average on 37 of thetrials and accounted for 76 of all the errors The Pear-son correlation between Array 1 errors and correct re-sponses across ISIs was 80 revealing a high level ofdependence between accuracy and Array 1 error Thismeans that the improvement in accuracy was heavily dueto the formation of a more complete representation of thefirst array Array 2 errors however as a percentage oftrials occurred on average on 12 of the trials and ac-counted for 24 of all the errors The Pearson correla-tion between Array 2 errors and correct responses acrossISIs was 34 revealing relative independence betweenaccuracy and Array 2 error

Early component versus late component Perfor-mance in the early component condition was comparedwith that in the late component condition by collapsingacross pre- and postdisplacement delays that exceededthe point at which accuracy no longer varied and per-forming a two-sample t test Asymptotic accuracy in thelate component condition was reliably higher than thatin the early component condition [t(18) 225] Themagnitude of this advantage was on average 20 Dur-ing this same time frame Array 1 errors were reliablylower in the late component condition [t(18) 287] aswere Array 2 errors [t(18) 355] The magnitude of

Figure 10 Accuracy and error results from the late component con-dition in Experiment 2


these differences was on average 15 and 5 respec-tively indicating that the increase in accuracy betweenthe early and the late component conditions was due to abetter representation of the first array

Eye Movement BehaviorThe same analyses of eye movement behavior (fixa-

tion number location and duration) were conducted asthose in Experiment 1 for the pre- and postdisplacementdelays (early and late component conditions respectively)

Number For both the early and the late componentconditions the number of fixations observed on a par-ticular trial increased with ISI Linear regression analy-ses indicated that in the early component condition 13additional fixations were made for every 1000-msec in-crease in ISI [F(188) 377 MSe 0444] In the latecomponent condition 12 additional fixations were madefor every 1000-msec increase in ISI [F(179) 123MSe 0857]

Location The absolute number of grid positions fix-ated increased with ISI in both the early and the latecomponent conditions In the early component condi-

tion the slope of this increasing function was 04 itemsfor every 1000-msec increase in ISI [F(188) 141MSe 0135] In the late component condition the slopewas also 04 items for every 1000-msec increase in ISI[F(179) 479 MSe 0224] These rates are aboutone third those of the increase in the number of fixationsobserved indicating that approximately one out of everythree fixations was on a unique grid position In both theearly and the late component conditions the rate at whichfixations corresponded to locations previously occupiedby a dot from Array 1 was greater than chance Acrosspredisplacement delay Array 1 locations were fixated atan average rate of 55 Across postdisplacement delayArray 1 locations were fixated at an average rate of 60Furthermore these rates varied linearly as a function ofthe corresponding pre- or postdisplacement delay In thecase of the early component condition the probability offixating an Array 1 location increased from 47 whenthe predisplacement delay was 0 msec to 76 when itwas 3500 msec Nearly identical rates of increase wereobserved in the late component condition in which theprobability of fixating an Array 1 location increased

Figure 11 Fixation number location and duration in the early component condition in Experi-ment 2


from 46 when the postdisplacement delay was 100 msecto 75 when it was 3500 msec However in both con-ditions the subjects demonstrated a preference for lo-calizing the majority of their fixations in the center fourgrid positions but to a lesser extent than in Experiment 1Through the first six fixations in the early componentcondition 84 of all the fixations occurred in the cen-ter four grid positions in the late component conditionthis value was 74

Duration In both the early and the late componentconditions an overall effect of delay was observed sincefixation durations generally increased with ISI howeveronly in the early component condition was an effect ofsaccade number observed where fixation durations wereshorter for later saccades In the late component condi-tion fixation number had no systematic impact on fixa-tion duration

DiscussionThe data from Experiment 2 suggest that memoryndash

percept integration is possible when the to-be-remembered

stimulus is not presented in the same spatial location asthe subsequent to-be-integrated stimulus As in the otherexperiments timing was found to be critical The ISIseparating the arrays although an essential factor in al-lowing memoryndashpercept integration to occur howeverwas not the only temporal determinant of performanceThe point during the ISI at which the grid changed loca-tion was also a major factor in determining performanceOptimal performance was achieved when the time priorto grid displacement was approximately 150 msec andwhen the time after displacement was approximately650 msec If the pre- or the postdisplacement time wasless than these values integration was suboptimal As afinal note considering the additional 300-msec blank in-terval these values produced an ISI of 1100 msec whichis very similar to that observed by Brockmole et al(2002) in their original studies in which the grids did notchange position

The eye movement results generally paralleled thoseobtained in Experiment 1 but several important differ-ences emerged In both experiments ISI had a similar ef-

Figure 12 Fixation number location and duration in the late component condition in Experi-ment 2

Fix 3Fix 2Fix 4Fix 5Fix 6Fix 1


fect on the number of fixations made and approximatelyone third of those fixations occurred in new grid posi-tions This indicates that the requirement that visual in-formation be acquired in separate fixations did not sig-nificantly alter the total number of fixations made or thenumber of grid positions visited by the eyes HoweverExperiment 2 revealed a systematicity in the distributionof fixations in the grid that was not evident in Experi-ment 1 In both the early and the late component condi-tions the subjects were more likely to fixate distal re-gions of the grid as well as positions previously occupiedby a dot from Array 1 especially at longer ISIs Notabledifferences between the early and the late componentconditions were evident however Fixation times de-creased as the predisplacement delay increased but wereunaffected by the postdisplacement delay suggestingthat when consolidation was confined largely to a spatialposition distant from the site of the initial stimulus eachfixation was used equally to process the displaymdashin thiscase to allocate attention to particular components ofthe grid That is the information processing at Fixation 1was equal to the processing at Fixation 2 and so on Thedecreasing fixation durations in Experiment 1 and in theearly component condition in Experiment 2 combinedwith the absence of fixation duration effects in the latecomponent condition in Experiment 2 suggest that eyemovements are instrumental in forming and maintainingthe structure of the to-be-remembered array only whenthe memory trace must be reconstructed in a unique po-sition in space When spatial memory does not have to bemoved or if it can be constructed prior to such move-ment eye movements play little to no role in its con-struction Rather eye movements are used to reconstructVSTM in a new space

Why might the patterns of fixations have differed be-tween Experiments 1 and 2 One possibility is that inExperiment 1 eye movements were not necessary to ac-quire information and so they were not strategicallyused to do so This would result not only in a reluctanceto move the eyes but also in randomness in the fixationsthat were made In Experiment 2 however the eye move-ment system had to be engaged strategically and once ithad its effects extended beyond moving the eyes betweenstimuli and into moving the eyes within each stimulusThese subsequent eye movements were not made to ac-quire new perceptual information since no informationabout the arrays was visually available but more likelyreflect a visual rehearsal process The eyes retraced thepattern formed by the lead array in order to reinstantiatethe initial stimulus array in a new spatially appropriatememory representation

So what kind of relationship is shared between the in-tegration and the eye movement systems Earlier in Ex-periment 1 two possibilities were laid out Integrationcould be either independent of eye movements or de-pendent on them in the sense that eye movements inter-fered with integration Experiment 2 clearly indicatesthat the latter was not the case Integration performance

was quite high when eye movements were required forintegration to be successful Instead the evidence pointsto at least partial independence between integration andeye movements However the magnitude of this inde-pendence is not clear When to-be-integrated informa-tion was presented in distinct spatial locations integra-tion relied on eye movements in order for the observer toperceive all the information Whether eye movementsare necessary to integrate information that is presentedin different spatial locations is not known however Forexample if Array 1 and Array 2 were to be presented indifferent locations both of which could be perceivedwithout moving the eyes could integration carry on nor-mally This question lies outside the scope of the pres-ent research but it is one that is important for develop-ing a complete understanding of the role that eyemovements may play in integration

Finally the result that integration performance was as-sociated with systematic patterns in gaze control is in-consistent with previous demonstrations that eye move-ments disrupt memory span for spatial locations Forexample Pearson and Sahraie (2003) found that sub-jectsrsquo memory span for spatial locations was lower ifthey executed saccades around a visual display as com-pared with cases in which they shifted covert attention tothe same locations Their paradigm however forced ob-servers to move their eyes to preestablished locationsaround the periphery of the display As such eye move-ments could not be used strategically to aid in the mem-ory task The present experimental paradigm howeverplaced no inherent restrictions on an observerrsquos gaze Asa result eye movements could have been used strategi-cally in order to rehearse the structure of the first arrayThis suggests that eye movements may help and hindermemory performance depending on whether gaze isunder the control of the observer

GENERAL DISCUSSION

The visual world contains more information than canbe perceived and understood in a single glance and thevisual system is confronted by the physical limitationsof the eye as well as by the cognitive limitations of at-tention and memory In order to overcome these limita-tions during scene processing information about theworld is obtained over time Observers move their eyesaround a scene to accommodate the resolution problemsassociated with peripheral vision and shift attentionaround the scene to sequentially process various objectsAs a result what has been seen needs to be analyzed inconjunction with what is being seen

Because visual perception has temporal extent tem-porally discontinuous input must be linked in memoryRecent research has suggested that this may be accom-plished by integrating the content of active VSTM withsubsequently perceived information (Brockmole et al2003 Brockmole amp Wang 2003 Brockmole et al 2002)The purpose of the present article was to explore the re-


lationship between VSTM consolidation and eye move-ments in order to discover how attention selects the in-formation that is to be integrated Specifically we ad-dressed the question of whether stimuli had to be overtlyattended to be included in the VSTM representation orwhether covert attention was sufficient

Previously Brockmole et al (2003) showed that dur-ing the delay separating the presentation of the arraysattention is deployed to those grid locations that werepreviously occupied by a dot from the first array But be-cause eye movements were not monitored whether thisdeployment was overt or covert could not be determinedExperiment 1 indicated that this deployment does notrely on overt shifts of attention That is integration seem-ingly proceeded independently of the eyes since the sub-jects made few eye movements f ixating only two orthree grid positions rarely along the periphery of thematrix Experiment 2 however demonstrated that the re-lationship between integration and eye movement be-havior differed in dynamic displays in which the first andthe second arrays were presented in different spatial lo-cations In this case in which eye movements were di-rectly related to task performance observers did viewthe grid systematically Although the number of eyemovements did not differ from that in the case of the sta-tic display the eye movements that were made were morestrategic since positions previously occupied by a dotfrom the first array were the most common loci of fixation

These results suggest that the differing use of eyemovements in each experiment may be related to differ-ent encoding strategies that were engaged on the basis ofthe different task demands in the static and the dynamiccases In the static display case VSTM was maintainedin the same spatial location as that in which it was gen-erated This apparently could be accomplished with covertdeployments of attention In the dynamic case howeverVSTM was generated in a location that did not overlapwith one of the to-be-integrated percepts In order toldquomoverdquo the memory trace overt shifts of attention wererequired

We hypothesize that these strategy differences mayalso explain why given previous evidence for a link be-tween eye movements and memory consolidation andmaintenance a tighter relationship between eye move-ments and integration was not found Although the re-sults of Experiment 2 lend support to the hypothesis thatVSTM consolidation and maintenance employ eye move-ments integration was found to be quite independent ofeye movements in Experiment 1 The present resultshighlight Neisserrsquos (1967) prediction that the relation-ship between eye movements and the reinstantiation of avisual display in memory should not be absolute At leastin some instances the consolidation and maintenance ofVSTM can rely on covert shifts of attention When covertshifts are insufficient for bringing a stimulus into thefocus of attention during perception observers employovert shifts by moving their eyes The same may be trueof memory

As a general conclusion overt shifts of attention arenot necessary for integration although they appear toplay a greater role in integration when temporally andspatially separated stimuli must be integrated This meansthat nonfoveated aspects of a display can be included inthe memory representation used in integration Indeedeven in Experiment 2 when the eyes moved systemati-cally to the to-be-remembered array very few positionsin the grid were actually foveated In order to producethe observed levels of performance then many regionsthat were not foveated must have been remembered Interms of memory rehearsal eye movements may be help-ful but this seems limited to cases in which they are re-quired in order to view all the relevant information Thusthe ISI required for integration (1000ndash1500 msec) isnot determined by scan time alone

Another conclusion that can be drawn from the pres-ent experiments is that engaging the eye movement sys-tem does not hurt integration if sufficient time is allowedfor encoding and VSTM generation In fact peak accu-racy in Experiment 1 (no eye movements because of spa-tial overlap) was 61 and in Experiment 2 (eye move-ments because of spatial displacement) it was 57 Theresults differ strikingly from those in prior studies inwhich the integration of visual information across sac-cadic eye movements has been investigated In the pasta transsaccadic memory store that maintains and accruesvisual information across successive fixations was pro-posed as a mechanism for relating information in onefixation to that obtained later (Breitmeyer Kropfl ampJulesz 1982 McConkie amp Rayner 1976) In generalhowever little evidence for information accrual over sac-cades has been found (Bridgeman amp Mayer 1983 Hen-derson 1997 Irwin et al 1988 Irwin et al 1983 Mc-Conkie 1991 OrsquoRegan amp Levy-Schoen 1983 Rayneramp Pollatsek 1983) although some limited visual infor-mation such as pattern structure and spatial relation-ships is remembered from fixation to fixation (egCarlson-Radvansky 1999 Carlson-Radvansky amp Irwin1995 Irwin 1991 1992 Pollatsek amp Rayner 1992) Alikely reason for the different results from prior work andthe present experiments concerns the role of time Pastresearchers investigated integration from one fixation tothe next within the time frame of one fixation (a fewhundred milliseconds) in the present study the observerswere afforded much more time (thousands of millisec-onds) Because of this integration did not occur fromone fixation to the next but rather over the course ofmultiple fixations that could be used to help reconstructan initial stimulus after its disappearance from viewThis presents an interesting contrast to past work on in-tegration and eye movements and should be examinedmore closely in future research

The finding that spatially displaced visual stimuli canbe integrated suggests that the memory representation ofthe first array is not tightly bound to a particular regionof space It is not spatial locations that are being re-membered but information about the stimulus itself in-


dependently of where it was presented What is the na-ture of the information that is retained during memoryndashpercept integration One could conceive of memoryndashpercept integration as a process that combines object rep-resentations That is each dot (or a bound unit or ldquochunkrdquoof a few dots) in each array is an independent entity main-tained in VSTM as a unique object One could also con-ceive of it as a process that integrates pattern informa-tion Instead of treating the dot arrays as collections ofindividual objects they are treated as global patterns akinto a spatial layout It is this layout information not objectidentity information that is integrated Currently we aredesigning experiments to dissociate these possibilities

Whether memoryndashpercept integration unites object orlayout information the properties of memoryndashperceptintegration observed in the present experiments suggestthat this integration process may serve a functional rolein scene processing Memoryndashpercept integration is ca-pable of combining information that is perceived at dis-crete points in time as well as information that misalignsin space In real-world perception conceptually relatedpieces of information are rarely viewed under conditionsof constant temporal retinal and spatial overlap Theability of the system to account for variability along thesedimensions simultaneously is critical if memoryndashperceptintegration is to play any role in the development of scenerepresentations We are now beginning to directly exam-ine memoryndashpercept integration as a means for con-structing object and scene representations

REFERENCES

Averbach E amp Coriell H S (1961) Short-term memory in visionBell Systems Technical Journal 40 309-328

Brandt S A amp Stark L W (1997) Spontaneous eye movementsduring visual imagery reflect the content of the visual scene Journalof Cognitive Neuroscience 9 27-38

Breitmeyer B G (1980) Unmasking visual masking A look at theldquowhyrdquo behind the veil of the ldquohowrdquo Psychological Review 87 52-69

Breitmeyer B G (1984) Visual masking An integrative approachNew York Oxford University Press

Breitmeyer B G amp Ganz L (1976) Implications of sustained andtransient channels for theories of visual pattern masking saccadic sup-pression and information processing Psychological Review 83 1-36

Breitmeyer B G Kropfl W amp Julesz B (1982) The existenceand role of retinotopic and spatiotopic forms of visual persistenceActa Psychologica 52 175-196

Bridgeman B amp Mayer M (1983) Failure to integrate visual infor-mation from successive fixations Bulletin of the Psychonomic Soci-ety 21 285-286

Brockmole J R Irwin D E amp Wang R F (2003) The locus ofspatial attention during the temporal integration of visual memoriesand visual percepts Psychonomic Bulletin amp Review 10 510-515

Brockmole J R amp Wang R F (2003) Integrating visual images andvisual percepts across time and space Visual Cognition 10 853-874

Brockmole J R Wang R F amp Irwin D E (2002) Temporal inte-gration between visual images and visual percepts Journal of Exper-imental Psychology Human Perception amp Performance 28 315-334

Buswell G T (1935) How people look at pictures A study of the psy-chology of perception in art Chicago University of Chicago Press

Carlson-Radvansky L A (1999) Memory for relational informa-tion across eye movements Perception amp Psychophysics 61 919-934

Carlson-Radvansky L A amp Irwin D E (1995) Memory for struc-

tural information across eye movements Journal of ExperimentalPsychology Learning Memory amp Cognition 21 1441-1458

Coltheart M (1980) Iconic memory and visible persistence Per-ception amp Psychophysics 27 183-228

Di Lollo V (1980) Temporal integration in visual memory Journalof Experimental Psychology General 109 75-97

Enns J T amp Di Lollo V (1997) Object substitution A new form ofmasking in unattended visual locations Psychological Science 8135-139

Eriksen B A amp Eriksen C W (1974) Effects of noise letters uponthe identification of a target letter in a nonsearch task Perception ampPsychophysics 16 143-149

Eriksen C W amp Collins J F (1967) Some temporal characteristicsof visual pattern perception Journal of Experimental Psychology 74476-484

Eriksen C W amp St James J D (1986) Visual attention within andaround the field of focal attention A zoom lens model Perception ampPsychophysics 40 225-240

Hebb D O (1968) Concerning imagery Psychological Review 75466-477

Henderson J M (1997) Transsaccadic memory and integration dur-ing real-world object perception Psychological Science 8 51-55

Henderson J M Weeks P A amp Hollingworth A (1999) The ef-fects of semantic consistency on eye movements during complexscene viewing Journal of Experimental Psychology Human Per-ception amp Performance 25 210-228

Irwin D E (1991) Information integration across saccadic eye move-ments Cognitive Psychology 23 420-456

Irwin D E (1992) Memory for position and identity across eye move-ments Journal of Experimental Psychology Learning Memory ampCognition 18 307-317

Irwin D E (2004) Fixation location and fixation duration as indicesof cognitive processing In J M Henderson amp F Ferreira (Eds) Theintegration of language vision and action Eye movements and thevisual world (pp 105-134) New York Psychology Press

Irwin D E amp Andrews R V (1996) Integration and accumulationof information across saccadic eye movements In T Inui amp J L Mc-Clelland (Eds) Attention and performance XVI Information inte-gration in perception and communication (pp 125-155) CambridgeMA MIT Press

Irwin D E Brown J S amp Sun J S (1988) Visual masking and vi-sual integration across saccadic eye movements Journal of Experi-mental Psychology General 117 276-287

Irwin D E Yantis S amp Jonides J (1983) Evidence against visualintegration across saccadic eye movements Perception amp Psycho-physics 34 49-57

Jiang Y amp Kumar A (2004) Visual short-term memory for two se-quential arrays One integrated representation or two separate repre-sentations Psychonomic Bulletin amp Review 11 495-500

Jiang Y Kumar A amp Vickery T J (2005) Integrating visual arraysin visual short-term memory Experimental Psychology 52 39-46

Just M A amp Carpenter P A (1980) A theory of reading From eyefixations to comprehension Psychological Review 87 329-354

Kahneman D (1968) Method findings and theory in studies of vi-sual masking Psychological Bulletin 70 404-425

Laeng B amp Teodorescu D (2002) Eye scanpaths during visual im-agery reenact those of perception of the same visual scene CognitiveScience 26 207-231

Lefton L A (1973) Metacontrast A review Perception amp Psycho-physics 13 161-171

Luck S J amp Vogel E K (1997) The capacity of visual workingmemory for features and conjunctions Nature 390 279-281

Matin E (1975) The two-transient (masking) paradigm Psychologi-cal Review 82 451-461

McConkie G W (1991) Perceiving a stable visual world In J vanRensbergen M Deuijver amp G drsquoYdewalle (Eds) Proceedings ofthe Sixth European Conference on Eye Movements (pp 5-7) LeuvenLaboratory of Experimental Psychology

McConkie G W amp Rayner K (1976) Identifying the span of the ef-fective stimulus in reading Literature review and theories of reading


In H Singer amp R B Ruddle (Eds) Theoretical models and pro-cesses of reading (pp 137-162) Newark DE International ReadingAssociation

Neisser U (1967) Cognitive psychology New York Appleton-Century-Crofts

OrsquoRegan J K amp Levy-Schoen A (1983) Integrating visual infor-mation from successive fixations Does trans-saccadic fusion existVision Research 23 765-768

Pearson D G amp Sahraie A (2003) Oculomotor control and themaintenance of spatially and temporally distributed events in visuo-spatial working memory Quarterly Journal of Experimental Psy-chology 56A 1089-1111

Phillips W A (1974) On the distinction between sensory storage andshort-term visual memory Perception amp Psychophysics 16 283-290

Pollatsek A amp Rayner K (1992) What is integrated across fixa-tions In K Rayner (Ed) Eye movements and visual cognitionScene perception and reading (pp 166-191) New York Springer-Verlag

Rayner K (1998) Eye movements in reading and information pro-cessing 20 years of research Psychological Bulletin 124 372-422

Rayner K amp Pollatsek A (1983) Is visual information integratedacross saccades Perception amp Psychophysics 34 39-48

Shulman G L amp Wilson J (1987) Spatial frequency and selectiveattention to spatial location Perception 16 103-111

Spivey M J amp Geng J J (2001) Oculomotor mechanisms activatedby imagery and memory Eye movements to absent objects Psycho-logical Research 65 235-241

Turvey M T (1973) On peripheral and central processes in vision In-ferences from an information-processing analysis of masking withpatterned stimuli Psychological Review 80 1-52

Volkmann F C Schick A M amp Riggs L A (1968) Time courseof visual inhibition during voluntary saccades Journal of the Opti-cal Society of America 58 562-569

Watson D G amp Humphreys G W (2000) Visual marking Evi-dence for inhibition using a probe-dot detection paradigm Percep-tion amp Psychophysics 62 471-481

Yarbus A (1967) Eye movements and vision New York Plenum

NOTE

1 An additional experiment conceptually replicated Experiment 2but only tested predisplacement delays of 150 600 and 1500 msec toincrease statistical power at each ISI No reliable effect of predisplace-ment delay on accuracy was observed [F(110) 1] We conclude thatthe temporary decrement in accuracy in the early component conditionin Experiment 2 at predisplacement delays between 500 and 700 msecalthough reliable is not indicative of any differences in the image rep-resentations available at those time points but rather is likely due tounsystematic variation or Type II error

(Manuscript received January 26 2004revision accepted for publication July 15 2004)


integration of the arrays and they appeared to be pre-sented simultaneously If the delay was extended to100ndash300 msec however accuracy was poor since thesecond array masked the first In the past this findinghas been taken as evidence that visual integration is im-possible once sensory memory has decayed (see Colt-heart 1980 Phillips 1974) However given that in theabsence of sensory memory successful performance re-quires both the formation and the retention of a longer-lasting representation integration might fail in theseshort delay situations because of a failure to consolidatethe initial representation into a stable and durable for-mat Indeed Brockmole et al (2002) demonstrated thatwhen the delays are longer than 300 msec performancebegins to improve and reaches an asymptotic level be-tween 1000 and 1500 msec The improvement in accu-racy over time was attributed to the generation of a men-tal representation of the lead array because the increasein accuracy was accompanied by a reduction in errors as-sociated with the lead array (ie erroneously selecting aposition previously occupied by the first array as the po-sition of the missing dot) over the same time period Inexcess of 95 of the variance in accuracy improvementwas accounted for by the decrease in Array 1 error (Brock-mole et al 2002)

Why postulate that the two arrays were integrated intoa single representation in VSTM when intuitively thereare several strategies that an observer could use to solvethe missing dot task For example the first array mightbe used to visually mark (see Watson amp Humphreys2000) grid positions as locations that cannot constitutethe correct answer As a result these positions might beinhibited and attention directed to the positions thatArray 1 left empty enhancing the detection of the spaceleft unfilled by the second array (see Jiang amp Kumar2004 Jiang Kumar amp Vickery 2005) Another possi-bility is that two separate memory representations ofeach stimulus array may be created in memory If the twomemory traces could then be compared ldquoin the mindrsquoseyerdquo the combination or integration of two stimulus ar-rays need not be postulated We investigated and rejectedthese possible explanations in two recent studies de-scribed next

To address the visual-marking hypothesis a directmeasure of the locus of spatial attention during the tem-poral integration task was used (Brockmole et al 2003)During the interstimulus interval (ISI) separating the ar-rays two-alternative forced choice probes appeared inpositions either filled by or left empty by the first arrayProbes that occupied positions filled by the first arrayenjoyed a processing benefit in that they were respondedto more quickly than probes that were in positions leftempty by the first array Thus attention was selectivelyallocated to the positions filled by Array 1 during thetime between arrays no evidence for Array 1 inhibitionor Array 2 priming was found In a separate study (Brock-mole et al 2002) the two-representation hypothesis wasassessed by examining the time required to consolidate

the second array into VSTM Although 1000ndash1500 msecwere required to consolidate Array 1 into VSTM reac-tion time analyses demonstrated that the time needed toprocess the second array at long ISIs was about equal tothat needed during perceptual integration (when the ISIwas 0 msec) suggesting that Array 2 is not submitted tothe same consolidation processes as Array 1 Because at-tention is allocated to the first array throughout the delayseparating the arrays and because the second array doesnot undergo a slow consolidation process it was con-cluded that over a period of 1000ndash1500 msec observerscan generate a VSTM representation of one stimulus anddirectly integrate subsequent perceptual information withthat representation

In other experiments the flexibility or abstractnessof the VSTM representations involved in integration wasconsidered by examining whether integration can occurwhen the to-be-integrated stimuli do not match in termsof their spatial properties (Brockmole amp Wang 2003)Stimuli can be integrated even if they are presented indifferent size scales and orientations This result indi-cates that during integration attention is not dedicated toparticular areas of space independently of the stimulithat occupied them The memory representation can bemodified to account for changes in stimulus structureand ldquoprojectedrdquo to different locations on the display Ex-pressed another way the memory trace of the lead stim-ulus is not spatiotopically or retinotopically specificrather observers are able to retain their memory for thelead array abstractly enabling it to be transformed to ac-commodate spatial changes thereby providing spatialflexibility in the integration process

An important question remaining concerns the timecourse of integration Specifically why does optimal in-tegration require up to 1500 msec of separation betweenvisual stimuli We assume that this time reflects thespeed with which the lead stimulus is consolidated intoVSTM In this article we focus on uncovering the mech-anisms that underlie this consolidation process by ex-amining observersrsquo eye movements during integrationSpecifically we ask whether overt shifts of attention tothe to-be-remembered elements are required for integra-tion That is in order to be included in the VSTM repre-sentation must a stimulus elementmdashor in the case of abrief presentation its locationmdashbe overtly attended orfixated In essence the issue is one of whether or not theinitial stimulus is maintained by eye movements that re-hearse its spatial structure On the basis of temporalcharacteristics of eye movements and assuming an aver-age fixation duration of 200 msec and an average sac-cade duration of 25 msec it would take 1575 msec toindividually fixate seven elements a time course match-ing that in integration experiments requiring the mainte-nance of seven items If the ISI is less than this valuefewer elements can be overtly attended or rehearsed andintegration will suffer

The idea that eye movements are important to integra-tion stems from research in which the use of eye move-






EXPERIMENT 1





MethodSubjects














Results

























05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34












EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE








EXPERIMENT 1





MethodSubjects














Results

























05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34












EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE





MethodSubjects














Results

























05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34












EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE





Results

























05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34












EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE

















05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34












EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE










05

10

11

162

56

05

14

514

236

05 05

15

20 16

16

89

215

110

24 89

185

422

30 24

89

16

22

22

11 67

94

218

28 39

406

150

17

34

16

11

22

51

12

68

41

206

103

22 27

150

274

59

36

37

12

24

72

24

50 69 14

179 255 50

72 100 134 45

24 19 24 21

38 57 84 38

11

11

179 198 57

126 99 61

19 46 34












EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE











EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE





EXPERIMENT 2



MethodSubjects
















Results











































GENERAL DISCUSSION














REFERENCES























































NOTE










Results











































GENERAL DISCUSSION














REFERENCES























































NOTE











































GENERAL DISCUSSION














REFERENCES























































NOTE


































GENERAL DISCUSSION














REFERENCES























































NOTE


























GENERAL DISCUSSION














REFERENCES























































NOTE


















GENERAL DISCUSSION














REFERENCES























































NOTE









GENERAL DISCUSSION














REFERENCES























































NOTE














REFERENCES























































NOTE






REFERENCES























































NOTE


















NOTE



Documents

Eye movements and the integration of visual memory and visual