memory and learning in figure-ground perceptionmapeters/mp pubs/petersonskow03.pdf ·...

Memory and Learning in Figure-Ground Perception

Mary A. Peterson and Emily Skow GrantDepartment of Psychology

University of ArizonaTucson, Arizona, USA

In B. Ross & D. Irwin (Eds.) (2003). Cognitive Vision: Psychology of Learning and Motivation, 42, 1-34.

Table of ContentsI. IntroductionII. Phenomena taken as evidence for the assumption that figure and ground assignment precedes access to shape and

object memoriesIII. Review of research revealing object memory effects on figure assignmentIV. Questions raised by evidence challenging the figure-ground first assumptionV. Tests of the Parallel Interactive Model of Configural AnalysisVI. Learning: How much past experience is necessary before memory for the structure of an object can affect figure

assignment?

I. IntroductionIt has long been debated whether or not it a clear

dividing line can be drawn between perception andmemory; the debate continues to this day.Nevertheless, since the turn of the 20th century, it hasbeen assumed that certain visual processes occursufficiently “early” so as to be impenetrable bymemory and other higher level processes. Anexample of one such early visual process is figureand ground assignment. tructure of a shape or anobject.

Figure-ground assignment occurs when tworegions share a common border (as the black andwhite regions do in Figures 1A – 1C). One region --the figure -- is typically seen as shaped by the border.The other region -- the ground -- is seen as shapelessnear the border it shares with the figure; it typicallyappears to continue behind the figure as itsbackground.1 The Gestalt psychologists held thatfigure assignment imposes shape onto unorganizedvisual input; shaped entities simply do not exist priorto figure-ground assignment. On the Gestalt view, theshaped entities in the visual field (the figures)provided the substrate for matches to shape or objectmemories. Thus, it was impossible to access shape orobject memories until after shape had been assigned.Following figure-ground assignment memories wereaccessed only by the shaped entities (the figures), andnot by the shapeless entities (the grounds).Throughout this chapter, the assumption that figure-ground assignment precedes access to objectmemories will be called the “figure-ground firstassumption.”

Figure 1. Displays illustrating figure-ground segregation. Theblack regions of A and B are enclosed, symmetric, and smaller inarea than their surrounds. A. A deciduous tree. B. A novel object.C. Rubin’s vase/face display

The Gestalt figure-ground-first assumption aroseas a counter-argument to the Structuralist view ofvisual perception. The Structuralists held that pastexperience (memory) imposed shape ontounorganized, pointillistic, visual input. For instance,in the Structuralist framework, one perceives a tree inFigure 1A because one has seen trees before. Thispast experience with trees both groups the featuresand parts of the tree together and specifies that theblack region is the shaped entity at its border with thewhite region. The Gestalt psychologists questionedhow the proper memory could be chosen to organizea particular array if no organization had yet beenimposed on the visual input. They reasoned thatsome prior organization of the visual input wasnecessary to constrain the memory matching process.This prior organization had to be based on cues thatwere innate. Excluding memory from the process oforganizing the visual input into shaped and shapelessentities also allowed the Gestalt psychologists to

account for the perception of novel shapes, shapes forwhich memory matches were destined to fail.

How then, does figure assignment occur?According to the Gestalt psychologists, figureassignment is determined by any of a number of“configural” cues that can operate without accessingmemory. Examples of the configural cues are closure,symmetry, convexity, and area. Regions that possessthese attributes are more likely to be seen as figuresthan regions that are open, asymmetric, concave, andlarger in area, respectively. The black regions of bothFigures 1A and 1B possess all of the configural cues.The Gestalt psychologists would argue that in bothcases, the black regions are seen as shaped entities –figures – because they possesses these attributes. Inthe Gestalt framework, the fact that Figure 1A alsoportrays a well known object -- a tree – is irrelevantfor figure assignment.

The Gestalt school had a revolutionary impact onthe field of visual perception in the early 1900’s. As aconsequence, it has long been thought that access toshape/object memories can occur only after the visualfield has been organized into figures and grounds.That is, it has been assumed that figure-groundassignment is immune to influences from memory,even from memories that are intrinsically visual (e.g.,memory for shape or object structure). Of course, thefigure-ground-first assumption entails the belief thata line separating perception and memory can bedrawn somewhere between figure assignment andmemories of shape or object structure. Research inPeterson’s laboratory has shown that the Gestalt-based figure-ground-first assumption is incorrect,however. Some form of shape/object memory isaccessed before, and contributes to, figureassignment.

In this chapter, we begin by showing that theevidence that long served to support the figure-ground-first assumption is really quite weak (SectionII). Next, we review some of Peterson andcolleague’s early work revealing shape and objectmemory effects on figure assignment (Section III). Inthis early work, observers reported their subjectiveimpression of where the figure lay with respect to theborder of interest; in other words, figure-groundperception was assessed via direct report. In SectionIV, we address a number of questions that are re-opened by the findings of Peterson and hercolleagues, questions for which answers generatedwithin the figure-ground-first assumption are nolonger valid. We review some research conducted toanswer these questions and we introduce a newmodel of figure assignment (Peterson, 2000;Peterson, de Gelder, Rapcsak, Gerhardstein, andBachoud-Lévi, 2000). In this model, memory ofshape/object structure serves as one of an ensemble

of figure cues, along with the Gestalt configural cues.Hence, this model does not represent a return to theStructuralist tradition where past experience was theonly organizing factor, or even the dominantorganizing factor. In Section V, the heart of thispaper, we describe some recent experiments testingthe competitive model. In these experimentsprocesses involved in figure assignment are assessedindirectly via a priming paradigm. In Section VI, wereview an experiment showing that a single pastexperience with a novel border exerts a measurableinfluence on figure assignment the next time theborder is encountered. We close the chapter inSection VII with some remarks concerning learning,memory, and perception.

II. Phenomena taken as evidence for theFigure-Ground-First Assumption

Three lines of argument and evidence have longbeen taken to support the figure-ground-firstassumption, but we will show that the support theyprovide is weak at best. The first line of evidence isbased on demonstrations that the perception of novelshapes can be accounted for by the operation of theGestalt configural cues. For instance, fromdemonstrations showing that shape could be imposedon the visual input using only configural cues (e.g.,Figure 1B), the Gestalt psychologists concluded thatshape was always imposed on the visual input usingonly configural cues (i.e., the figure-ground firstassumption).

The figure-ground first assumption does notfollow as a logical conclusion from demonstratingthat configural cues can account for shape perceptionwhen past experience cannot (because the displaysare novel). Such demonstrations do not support theconclusion that past experience cannot affect figureassignment when familiar shapes and objects arepresent (Peterson, 1999). To reach this latterconclusion, one must conduct investigationsinvolving familiar shapes and show that largevariations in familiarity do not affect figureassignment when the configural cues are heldconstant. Neither the Gestalt psychologists nor theirdescendents conducted stringent tests using thisstrategy.2

A second line of support for the figure-groundfirst assumption arose from a neuropsychologicalinvestigation conducted by Warrington and Taylor(1973). They presented a visual agnosic patient who,although quite poor at object and shape identification(as visual agnosics are), nevertheless performedfigure-ground assignment correctly. Marr (1982)interpreted the patient’s pattern of impaired andspared performance within a serial hierarchical model

of vision, and took it as evidence that objectmemories are accessed only after figure assignmenthas been determined. Marr argued that the patient’slesion must be located higher than the brain regionresponsible for figure assignment, but lower than thebrain region where memories of objects are stored.

However, naming responses, such as thoserecorded by Warrington and Taylor (1973) can onlyindex whether or not conscious recognition andidentification has occurred. They do not necessarilyreveal whether or not some form of object memorywas accessed in the course of figure assignment(Peterson, et al., 2000). To address this latter issue, itis necessary to compare figure assignment for regionsthat are matched for Gestalt configural cues butmismatched in the degree to which they fit the shapesof known objects. Such tests might reveal that, forvisual agnosics as well as for normal perceivers,borders may be more likely to be seen as boundariesof regions (or portions of regions) portraying knownobjects rather than novel objects. (For furtherdiscussion and a relevant experiment, see SectionIV.)

A third phenomenon that has been taken asevidence for the figure-ground-first assumption is thewell-known coupling between figural status andconscious recognition, illustrated by the Rubin vase-faces display in Figure 1C. The vase can berecognized when the central black region appears tobe the figure at the border it shares with the adjacentwhite region, but not when it appears to be ground tothe surrounding white region. Likewise, the facescan be recognized when the surrounding white regionappears to be the figure at the vertical borders itshares with the black region, but not when the whiteregion appears to be ground at those borders. Thiscoupling between figural status and recognition ledmany vision scientists to accept the figure-ground-first assumption. A coupling cannot provideunequivocal evidence for a serial sequence, however.

Surprisingly, until the initial tests conducted inour laboratory were published in 1991 (Peterson &Gibson, 1991; Peterson, Harvey, & Weidenbacher,1991), there were very few direct tests of whether ornot past experience contributed to figure assignment.A few experiments had suggested that aspects of pastexperience might affect figure assignment (Schaffer& Murphy, 1943; Rubin, 1915/1958). These resultswere dismissed based on procedural criticisms,desultory attempts (and failures) to replicate, andalternative interpretations that did not fit the data anybetter than the original interpretation did (for review,see Peterson, 1995, 1999). The Gestalt argumentsagainst the Structuralist tradition continued to exert astrong hold on perception psychologists who, despiteevidence that memory and past experience affected

many other visual processes, continued to believethat figure-ground assignment lay far enough belowan implicit line dividing perception from memory tobe immune to influences from memory.

III. Review of Peterson’s researchrevealing object memory effects on figureassignment

Peterson and her colleagues directly testedwhether or not memories of well-known shapes wereaccessed in the course of figure assignment. Theybegan using the displays shown in Figure 2A and 2B,originally drawn by Julian Hochberg. The displayswere biased toward a center-as-figure interpretationby the Gestalt configural cues of smallness of relativearea, enclosure, and symmetry (or partial symmetry).The monocular depth cue of interposition alsofavored the interpretation that the black region lay infront of the white region in Figure 2B. In addition,the observers fixated the center region, whichincreases the likelihood that a region will be seen asfigure (Hochberg, 1971; Peterson & Gibson, 1994a).The vertical borders between the black and whiteregions sketched portions of known objects on thewhite side (standing women in Figure 2A, and faceprofiles in Figure 2B). Peterson, et al. (1991) showedthese displays to observers who viewed them for longdurations (30-40 seconds) and reported continuouslywhether the black or the white region appeared to befigure by pressing one of two keys.

Figure 2. The displays used by Peterson, Harvey, andWeidenbacher (1991). The displays are biased toward theinterpretation that the black center region is the figure. Portions ofknown objects are sketched along the white side of the verticalblack/white borders in both stimuli, portions of standing women in(A) and face profiles in (B).

Observers viewed all displays in both an uprightorientation, as shown in Figure 2, and in an invertedorientation (which can be seen by turning the bookupside down). Changing the orientation from uprightto inverted did not change the Gestalt configuralcues: the center black region is enclosed, symmetric,and smaller in area than the surrounding white regionboth when the display is upright and when it isinverted. Nor did it change the monocular depth cueof interposition in Figure 2B, or the fact that

observers fixated the black region on all trials.However, when the display is upright, the knownobject sketched on the white side of the black-whiteborder is portrayed in its typical orientation, whereaswhen the display is inverted, the known object isdisoriented from its typical upright.

Figure 3. Mean durations that the two regions of the displays inFigure 2 were maintained as figures in upright versus inverteddisplays. The black bars denote the center black regions; thestriped bars denote the surrounding white regions.

Access to shape and object memories isorientation-specific. For instance, it takes longer forobservers to identify objects and pictures of objectsthat are disoriented from their typical uprightorientation (Jolicœur, 1988; Tarr & Pinker, 1989).Perrett, Oram and Ashbridge (1998) have shown thatit takes longer for a population of cells coding anobject to reach some threshold if the object is shownin an atypical orientation. The orientation-specificityof object recognition led Peterson and her colleaguesto hypothesize that changing the orientation of thedisplays might reveal object memory effects onfigure assignment by modulating them. Specifically,if object memories affect figure assignment, theirinfluence should be larger for upright displays thanfor inverted displays. Therefore, Peterson et al.(1991) reasoned that object memory effects on figureassignment would be implicated if the figuresappeared to lie on the white side of the vertical black-white borders in Figures 2A and 2B more often whenthe displays were upright than when they wereinverted.

Their results, shown in Figure 3, supported thisprediction. Observers saw the white surrounds asfigures for longer durations in the upright orientationthan in the inverted orientation. Taken alone, thisfinding could simply indicate that regions portrayingfamiliar objects could be maintained as figures longeronce they had obtained figural status. Importantly,observers saw the black centers as figures for shorterdurations in the upright orientation than in theinverted orientation. In other words, reversals out ofthe black center as figure interpretation and into thewhite surround as figure interpretation were more

likely when the displays were upright than when theywere inverted. This finding suggested that objectmemories affected the likelihood that theorganization would reverse into the surround asfigure interpretation, as well as the likelihood that thesurround would be maintained as figure once it wasperceived as such.

Peterson et al. (1991) found that the order inwhich upright and inverted displays were presenteddid not matter. What mattered was that the parts ofthe well-known object were presented in their properspatial relationships, both with respect to the uprightand also with respect to each other. Peterson et al alsotested conditions in which the parts were re-arranged(scrambled) so that the object was no longerrecognizable. The effects of object memories onfigure assignment were diminished, as they were forinverted stimuli.

Importantly, Peterson et al. (1991) found thatknowledge could not overcome the effects ofchanging the orientation or rearranging the parts. Theorientation effects were obtained even if observersknew that the displays portrayed inverted women orinverted face profiles; the same was true for theeffects of scrambling the parts. This finding indicatedthat knowledge of any type could not produce theseeffects; access to memories of object structure via thevisual input was necessary (see also Gibson &Peterson, 1994).

The results obtained by Peterson et al. (1991)indicated that memories of object structure (at least)are accessed in the course of figure assignment andaffect its outcome. It was clear in the originalexperiments that semantic knowledge alone wasinsufficient for these effects, the proper structure ofthe object was necessary. Peterson and Gibson(1991, 1994b; Gibson & Peterson, 1994) showed thatthe Peterson et al. (1991) results extended to maskeddisplays exposed for brief durations (as short as 28ms).

The initial results showing that object memoriesaffected figure assignment were obtained usingdisplays that were biased against seeing the figurelying on the side of the border where a well-knownobject was sketched. Later, Peterson and Gibson(1994a; Gibson & Peterson, 1994) tested whetherobject memories affected figure assignment usingdisplays, such as those in Figure 4A, in which objectstructure was the only cue that reliably distinguishedbetween the regions on either side of a central border.They found orientation effects for these displays aswell: Observers were more likely to report seeing thefigure on the side of the border where the well-knownobject was sketched when the displays were uprightrather than inverted. Thus, object memory effects onfigure assignment were evident both with displays

02468

101214161820

UP INV

Simulus Orientation

Mea

n D

ura

tio

ns

See

n A

s F

igu

re (

sec)

that should have been unambiguous if only thetraditional Gestalt cues were taken to be relevant tofigure assignment (e.g., the displays in Figure 2) andwith displays that were ambiguous in that configuralcues were equated for the two adjacent regions (e.g.,displays like those in Figure 4A).

.Figure 4. A. Sample figure-ground stimuli in which two equal-area regions share a border; a known object was sketched on oneside of the central border. These stimuli portray a guitar, a lamp,and a standing woman, respectively. Although the known objectsare always shown in black on the left side of the border in thisfigure, in the displays used in the experiments, they were shownequally often in white and in black and on the left versus the rightof the border. B. ”Scrambled” versions of the stimuli in A. Tocreate the scrambled versions, the objects in A. were separated intoparts at the concave cusps, and those parts were reassembled sothat the new arrangement did not portray a known object.

The next question addressed by Peterson andGibson (1994b) was how the memory of objectstructure cue fared when it was placed in competitionwith a single other cue, such as the Gestalt configuralcue of symmetry. Consider displays in which asymmetric region shares a vertical border with anasymmetric region. The asymmetric region portraysa known object, whereas the symmetric region doesnot. When viewing inverted versions of such stimuli,where the object memory cue was absent ordiminished, observers were significantly more likelyto report seeing the symmetric region as figure.When viewing upright stimuli, there was a substantialand significant increase in reports that the figureappeared to lie on the side of the border where a well-known object was sketched compared to invertedstimuli. Importantly, the object memory cue did notdominate the symmetry cue in the upright orientation;instead, the two cues seemed to compete with eachother on a roughly equal footing. This finding ledPeterson and Gibson (1994b) to suggest that theobject memory cue is one of many cues that

determines figure assignment; it neither dominatesthe other relevant cues, nor is dominated by them.

In a different series of experiments, Peterson andGibson (1993) added binocular disparity to displayslike those in Figure 4A in which object memoryfavored seeing the figure on one side of a border, butGestalt configural cues did not reliably distinguishbetween the two sides. Binocular disparity indicatedthat the figure lay either on the same side or on theopposite side of the border as the known object.Peterson and Gibson expected that when both objectmemory and binocular disparity specified that thefigure lay on the same side of the border, the displayswould be unambiguous. The stimuli in which objectmemory and binocular disparity specified that thefigure lay on opposite sides of the border were theinteresting case. If the addition of binocular disparityrendered the displays unambiguous, then the figureshould always appear to lie on the side across theborder from the known object. Alternatively, if objectmemories always overpower binocular disparity, thefigure should always appear to lie on the knownobject side of the border.

Peterson and Gibson (1993) found that objectmemories did affect figure assignment in thesecritical displays, but they did not dominate thebinocular disparity cue. Instead, for the range ofdisparities Peterson and Gibson tested, the objectmemory cue appeared to compete with binoculardisparity on a roughly equal footing, as it had withsymmetry (see also Peterson, 2002). The figure wasseen to lie on the side of the border where the well-known object was sketched approximately half thetime, and on the opposite side, favored as figure bybinocular disparity, the rest of the time.

The results of these experiments, showing thatobject memories affect figure assignment in bothtwo-dimensional and three-dimensional displays,challenged the figure-ground first assumption. Theyalso raised anew number of questions, for whichanswers based on the figure-ground-first assumptionwere now inadequate. We address those questions inthe next section.

IV. Questions raised by evidencechallenging the figure-ground firstassumption.

A first set of questions is the following. How canobject memories be accessed before figure-groundorganization has been imposed on the visual field;that is, before shaped figures have been separatedfrom shapeless grounds? What serves as the substratefor access to object memories? Must we return to theStructuralist claim that past experience can be

accessed by completely unorganized pointillisticinput?

In response to these questions, Peterson andGibson (1993, 1994b) proposed that at least the initialstages of edge extraction precede access to objectmemories and that edges, rather than shaped entitiesor even whole regions, were the substrate for matchesto shape and object memories. They argued thatedge-based access to memories of object structurecould occur at the same time that the Gestaltconfigural cues are being assessed. This would allowmemories of object structure to serve as one morefigural cue (i.e., to add to the traditional ensemble ofGestalt configural cues).

Further, Peterson and Gibson (1993) argued thatnot all edges could support object memory effects onfigure assignment. One critical requirement is thatedges must be extracted early in processing; onlysuch edges can support quick access to objectmemories. Evidence that object memories must beaccessed quickly if they are to affect figureassignment comes from the orientation effects.Inverted stimuli do access memories of familiarobjects; they just take longer than upright stimuli todo so. The additional time required is sufficient torender object memory effects on figure assignmentless likely for inverted displays than for uprightdisplays. This is because figure assignment occursearly in the course of visual processing. Therefore,any factor that delays access to object memories canremove or diminish their effects on figureassignment. If edge extraction takes too long, edge-based access to object memories will not occurquickly enough to affect figure assignment.Consistent with this argument, Peterson and Gibson(1993) failed to observe effects of object memorieson figure assignment using random-dot stereograms,where edge extraction takes some time.

Peterson (1995, 2003; Peterson & Hector, 1996)proposed further that object memories could beaccessed by portions of edges, rather than by wholecontinuous edges or borders.3 Thus, like the Gestaltpsychologists, Peterson and her colleagues assumethat some organization is imposed on the visual inputbefore object memories are accessed; thus, they donot support a return to Structuralism. However,Peterson and her colleagues clearly assume that a lotless organization has been imposed before objectmemories are accessed than did the Gestaltpsychologists and their followers.

A second set of questions that was raised byPeterson and colleague’s challenge to the figure-ground-first assumption concerns the behavior ofvisual agnosic patients, such as the one tested byWarrington and Taylor (1973). If tested withdisplays designed to reveal object memory effects on

figure assignment, will visual agnosics behave likenormal observers or will they fail to show effects ofobject memories on figure assignment? If a visualagnosic cannot identify the objects portrayed infigure-ground displays, yet shows spared objectmemory effects on figure assignment, that wouldsuggest that impaired identification responses cannotbe taken to support a serial view of the relationshipbetween figure-ground assignment and access tomemories of object structure.

To address these questions, Peterson et al. (2000)tested a visual agnosic patient, A.D. They assessedA.D.’s object recognition/identification abilities via abattery of standard tests, including the BostonNaming Test, the impossible objects subtest of theBirmingham Object Recognition Battery (the BORB,Riddoch & Humphreys, 1993), and the Visual Objectand Spatial Perception Battery (VOSP, Warrington &James, 1991). These tests require either a namingresponse or a decision regarding whether a depictedobject is a familiar object or a novel (or impossible)object. The VOSP sub-test uses silhouettes of objects,which were particularly relevant to our displays.A.D. performed considerably below age-matchedcontrol observers on all of these tests. This type ofperformance is typical for visual agnosics, soperformance on these tests partially confirmed thatA.D. was a visual agnosic and did not simply havename-finding problems.

Other tests indicated that A.D.’s semanticknowledge regarding those objects she could notidentify was intact. She could define objects andgive a reasonable description of what they lookedlike. However, it seemed that this knowledgeregarding objects could not be accessed by visualinputs, at least as indexed by naming responses or byovert judgments regarding the familiarity/possibilityof objects. Again, this is a typical pattern ofperformance for visual agnosics.

Peterson et al. (2000) also assessed A.D.’s abilityto use the Gestalt configural cues of convexity andsymmetry to perceive figure-ground relationships innovel displays. A.D. performed well within normallimits on these tasks. Thus, A.D.’s performance onthese initial identification tests and figure-groundtests was similar to that shown by the patient reportedby Warrington and Taylor (1973).

Next, Peterson et al. (2000) performed the criticaltest of whether or not object memories could affectfigure assignment even in a visual agnosic. Theyasked A.D. to report which region was the figure(i.e., which region appeared to stand out as having adefinite shape at the central border) in 48 displayslike those shown in Figure 4. These displays wereconstructed from two equal-area regions separated bya central articulated border. Half of these displays

were “experimental” displays in that a portion of afamiliar object was sketched along one side of thecentral border separating black and white regions (thedisplays in Figure 4A). The critical side on which thefamiliar object was sketched was the left for half ofthe experimental displays and the right for the otherhalf; the critical region was black in half the displaysand white in the other half the displays. The rest ofthe displays were “control” displays in which thecentral border did not sketch a known object on eitherside. The control displays had critical regions thatwere formed by rearranging (scrambling) the parts ofthe familiar objects portrayed by the critical regionsin the experimental displays such that they were nolonger recognizable (the displays in Figure 4B).Thus, the critical sides of the control andexperimental stimuli were matched in part structure,but not in spatial structure. Therefore, they were notmatched in the degree to which they provided a goodfit to memories of object structure. None of theGestalt configural cues consistently favored seeingone of the two halves as figure in the experimentaldisplays compared to the control displays.

Peterson et al. (2000) reasoned that if objectmemories affect figure assignment even in theabsence of conscious recognition and identificationthen, like non-brain-damaged participants, A.D.should report seeing the figure lying on the criticalside of the central border more often in experimentaldisplays than in control displays. Their resultssupported this prediction: Like non-brain-damagedage-matched controls, A.D. reported seeing the figurelying on the critical side of the central bordersignificantly more often in experimental stimuli(75%) than in control stimuli (46%). As expected of avisual agnosic, A.D. was not able to identify theobjects portrayed by the critical regions of theexperimental displays, even though she clearly sawthem as figure. Her performance deviated from thatof the age matched controls in this respect.

Thus, conscious identification is not necessary forobject memories to affect figure assignment. The dataobtained from A.D. show that it is erroneous toconclude that figure assignment precedes access toobject memories based on a pattern of intact figureassignment and impaired identification. Instead,A.D.’s performance is consistent with the proposalthat quick, unconscious access to memories of objectstructure can occur and can contribute to figureassignment even when conscious recognition andidentification is impaired.

A third set of questions raised by the claim thatobject memories affect figure assignment is thefollowing: If object memories are accessed in thecourse of figure assignment, how can one account forthe fact that regions that would portray familiar

objects were they to be seen as figures appearshapeless when they are perceived to be grounds?Recall that Peterson et al. (1991; Peterson & Gibson,1993, 1994b) showed that, when other configural anddepth cues compete with object memories, the figuredoes not always appear to lie on the side of the borderwhere the known object is sketched. In such cases, ifobject memories matching the ground region wereaccessed in the course of figure assignment, whydon’t we recognize the familiar object sketched onthe ground side of the border? More specifically, whydo we generally not perceive both the vase and thefaces in the Rubin vase-faces display? Why do wetypically perceive only one of these shaped entities ata time? On the traditional figure-ground-first view,grounds were shapeless because they were notmatched to object memories.4 The Parallel InteractiveModel of Configural Analysis, introduced byPeterson and her colleagues (Peterson, 2000;Peterson et al., 2000), provides an explanation for theperceived shapelessness of grounds while assumingthat memories of object structure are accessed in thecourse of figure assignment.

The Parallel Interactive Model of ConfiguralAnalysis (PIMOCA) is illustrated in Figure 5.PIMOCA assumes that as soon as edges are detectedin the visual field, portions of those edges areassessed for configural cues on both sidessimultaneously. In PIMOCA, memories of objectstructure are considered to be a configural cuesbecause previous experiments in our laboratory haveshown that the parts of the familiar object must becorrectly configured in order for the object memorycue to be effective (Gibson & Peterson, 1994;Peterson, 2003; Peterson, Gerhardstein, Mennemeier,& Rapcsak, 1998; Peterson et al., 1991, 2000). Giventhat configuration matters, it seems appropriate toinclude object memories amongst the configural cues.

Figure 5. The Parallel Interactive Model of Figure Assignmentproposed by Peterson, et al. (2000). Shortly after edges aredetected (e.g., the curvilinear edge in the center of the figure),figural features such as Symmetry (Symm), Convexity (Conv),

Memory of Object Structure (MOS) and Closure are assessed forboth sides. Features on the same side of the edge cooperate (asindicated by double-headed arrows). Features on opposite sides ofthe edge compete (as indicated by the horizontal end-stopped linecrossing the edge).

According to PIMOCA, configural cues presenton the same side of an edge cooperate with eachother, whereas configural cues present on oppositesides of an edge compete with each other. When thecues are unbalanced, the cues on the more weaklycued side are inhibited by the cues on the morestrongly cued side.5 The inhibition of configural cueson the more weakly cued side of a border accountsfor the perceived local shapelessness of the regionlying across the border from a more strongly cuedside. Peterson (2002; Peterson et al., 2000) arguedthat in two-dimensional displays, such as those usedin our experiments, one perceives shape byperceiving properties such as symmetry, convexity,area, enclosure, familiar object structure, etc. If thosecues are inhibited on the relatively weakly cued sideof an edge, shape simply cannot be seen in that localvicinity (provided that configural cues are the onlycues present). The cross-border inhibition proposedin PIMOCA accounts for the fact that regionsadjacent to strongly cued figures are perceived to belocally shapeless, both under conditions where aportion of a known object is sketched on the moreweakly cued side of the border, and under conditionswhere the more weakly cued side is convex orsymmetric.

On the more strongly cued side, continuedcooperation among cues leads ultimately to theperception of shape, and interactions between theconfigural cues and semantic and functionalknowledge lead ultimately to recognition, if the shapeis familiar (barring brain damage).

In Figure 5 boxes of the same size portray all ofthe configural cues. By representing the cues in thisfashion, we do not mean to imply that all of theconfigural cues are equally potent. We know that isnot the case. Kanizsa and Gerbino (1975) showedthat convexity is more potent than symmetry, forexample. Likewise, the configural cues all appear tolie on one plane in Figure 5. By presenting all of theconfigural cues in this way, we do not mean to implythat they are all computed at the same level ofprocessing. Indeed, there is some suggestion thatthese cues may be assessed at different levels. Forinstance, cells that respond differentially to convexand concave shapes have been found in V3(Pasupathy and Connor, 1999). And, based on workby Tanaka (1996) Peterson (2003) has hypothesizedthat the relevant object memories may be found in thehuman analogue of V4. The figure is designed toimply that the configural cues (including the memory

of object structure cue) are accessed in parallel, andthat configural cues accessed on the same side of aborder cooperate with each other, whereas thoseaccessed on opposite sides compete with each other.

In PIMOCA, figure and ground assignment is alocal outcome of a cross-border competition. It is nota stage of processing through which visual inputsmust pass before object memories can be accessed(Peterson, 2002). Nor must figure and groundnecessarily be assigned consistently to the same sideacross a continuous border; figures can be assigned todifferent sides along different extents of a continuousborder (Hochberg, 1962; Peterson, 1995, 2003;Peterson & Hector, 1996). There is evidence thatfigure and ground assignment is affected by theglobal context in which a border is found (Kim &Peterson, 2001, 2002; Peterson & Kim, 2001b). Weare currently working on integrating context effectsinto the model.

PIMOCA is one of a class of competitive modelsof figure assignment (see also Keinker, Sejnowski,Hinton, & Schumacher, 1986; Sejnowski & Hinton,1987; Vecera & O’Reilly, 1998). PIMOCA is uniquein

• Assuming that memories of partial objectstructure are accessed via edges rather than viaregions or shapes. (Sejnowski and colleagues did notconsider a role for object memories, and Vecera andO’Reilly proposed a holistic, region-wide match toobject memories.)

• Assuming that memories of partial objectstructure are accessed in parallel with assessments ofthe Gestalt configural cues.

• Treating figure-ground segregation assimply an outcome of the cross-competitive processrather than as a stage of processing.

• Accounting for the perceived shapelessnessof grounds via cross border competition.

V. Tests of The Parallel InteractiveModel of Configural Analysis

Peterson and Kim (2001a) tested PIMOCA’spredictions regarding the inhibition of cues on therelatively weakly cued side of a border. To do so,they isolated the memory of object structure cue onthe white side of a black/white border where themajority of cues favored assigning figural status tothe opposite, black side. Those cues included theconfigural cues of symmetry, convexity, enclosure,and smallness of relative area, along with other cuessuch as fixation and expectation. Peterson and Kim’s(2001a) stimuli were black silhouettes like thoseshown in Figure 6. Because more configural cuesfavored assigning figural status to the black side ofthe border and because previous evidence indicated

that the memory of object structure cue did notdominate the other configural cues, Peterson and Kim(2001a) expected that the figure would appear to lieon the black side of the border. They predicted thatthe object structure memory accessed on the whiteside of the border would be inhibited.

All of the black silhouettes were novel shapes.Silhouettes like those in Figures 6A – 6C were shownon 75% of the trials; these were control silhouettes.Silhouettes like Figure 6D, where a portion of afamiliar object was sketched on the white side of theborder, were shown on 25% of the trials; these wereexperimental silhouettes (see below). The silhouetteswere exposed briefly, for 50 msec.

Figure 6. Sample black silhouettes used as primes by Peterson andKim (2001a). All silhouettes were novel. The figure was seen onthe black side of the black/white border because a larger number ofcues favored assigning the figure to that side than to the other side(e.g., symmetry, enclosure, smallness of area). A - C. Controlprimes; the borders of control primes did not sketch a knownobject on either the inside or the outside of the silhouette. D.Experimental prime. In all experimental primes, the verticalborders sketched a portion of a known object along the outside (thewhite, ground, side). A portion of an anchor is sketched on thewhite side of the black silhouette in (D). Hence, for experimentalprimes the memory of object structure cue is present on the whiteside of the black/white border.

Observers saw the bounded black regions as theshaped entities; they saw the white regions asshapeless grounds, even for the experimentalsilhouettes. The stimuli were designed to be seen thisway, because (1) a larger number of configural cuesfavored seeing the figure on the black side of theborder rather than on the white side, (2) theexperimental stimuli were embedded amongst manycontrol stimuli in which there was no familiar objectsketched on either side of the black/white border, and(3) the silhouettes appeared on the point where theparticipants were fixating.

Observers made no response to the silhouettes;they were asked to simply look at them. Their task

was to judge quickly whether a line drawing shownafter each silhouette portrayed a familiar object or anovel object. The silhouettes served as primes beforethe line drawings. The critical trials were thoseinvolving familiar line drawings. As shown in Figure7, half of the line drawings of familiar objects werepreceded by experimental silhouette primes in whicha portion of the same basic level object was sketchedon the white (ground) side of the black silhouette.These were the experimental trials. The other half ofthe line drawings of familiar objects were precededby control silhouette primes with no familiar objectsketched on the ground side (control trials). Controlsilhouettes preceded all line drawings of novelobjects. The experimental and control silhouetteswere matched for size, area, convexity, andcurvilinearity so that observers could not distinguishbetween them.6 A different unique silhouette wasshown on each trial.

Figure 7. Examples of prime and line drawing matches forfamiliar line drawings. Line drawings shown on experimental trialswere preceded by silhouette primes in which an object from thesame basic level category was sketched on the ground side. Asample experimental trial is shown in the right panel. Linedrawings shown on control trials were preceded by silhouetteprimes that sketched novel shapes on both sides of the black-whiteborder. A sample control trial is shown in the left panel.

The dependent measure was participants’ latencyto correctly categorize the line drawings as familiaror novel objects. We were primarily interested inparticipants’ responses to the familiar line drawings.If the inhibition proposed in PIMOCA occurs, thenobject memories accessed for the white side of theexperimental silhouette primes should be inhibited.This is because, according to PIMOCA, when thecues for seeing the figure lying on one side of theborder are stronger than the cues for seeing the figurelying on the other side (as they are in the silhouetteprimes), configural cues (including memories ofobject structure) accessed on the more weakly cued

A. B.

C. D.

Tim

e

Control Experimental

side are inhibited. Peterson and Kim (2001a)hypothesized that evidence of this inhibition wouldbe revealed if RTs to correctly categorize familiarline drawings were longer following experimentalprimes rather than control primes. This predictionsupposes that the line drawing following theexperimental prime must access some of the samememories of object structure as the object sketchedon the more weakly cued side of the border of theprime because it is drawn from same basic-levelcategory. If those memories are inhibited because ofthe cross-contour competition occurring during theperception of the prime, then responses to the linedrawings shown on experimental trials should beslowed, provided that the inhibition lasts long enoughto be probed by the line drawing. (No familiar objectwas sketched along any portion of the border of thecontrol primes; hence, no inhibition of specific objectmemories was expected.)

It is important to point out that although thefamiliar line drawings shown on experimental trialsportrayed an object from the same basic levelcategory as the portion of a known object sketched onthe ground side of the silhouette, the contours of theline drawing were not the same as the contours of thesilhouette. We made the contours different becausewe wanted to be sure that any RT differences weobserved reflected access to previously establishedmemories of known objects in the course of figureassignment and not simply memory for the specificshape of the border of the silhouette. (See SectionVI.) Indeed, it could be argued that the participantshad not seen the particular borders of the silhouettesbefore, although they had certainly seen similarborders bounding objects from the same basic levelcategory (or, at least portrayals of such objects). Wedesigned these experiments to assess whether or notpreviously established memories of known objectswere accessed in the course of figure assignment andwere inhibited if they were accessed on the moreweakly cued side of a border.

The novel line drawings were included just so theparticipants had to categorize the line drawingtargets. Although the borders of the controlsilhouettes shown before the novel line drawingssketched novel objects on both the figure side and theground side, no attempt was made to match theshapes of the novel silhouettes to the shapes of thenovel objects. Hence, responses to the novel linedrawings will not be discussed further, except to saythat observers took longer to correctly categorize thenovel line drawings than the familiar line drawings.

Peterson and Kim (2001a) reported twoexperiments. In Experiment 1, the silhouette primeswere exposed for 50 ms and the line drawings weredisplayed following an inter-stimulus interval of 33

ms. In Experiment 2, the silhouette primes wereexposed for 50 ms and the line drawings were shownimmediately afterwards. In both experiments, the linedrawings remained on until a response was made. Ascan be seen in Figure 8A, the results supported thepredictions generated from PIMOCA. In bothexperiments, observers took significantly longer tocorrectly categorize the familiar line drawings onexperimental trials than on control trials.7

Peterson and Kim’s (2001a) results provideindirect evidence that object memories are accessedin the course of figure assignment. Until theseexperiments were conducted, the evidence supportingthe proposal that object memories were accessed inthe course of figure assignment was based onparticipants’ direct reports regarding theirphenomenological experience. Some investigatorshad wondered whether our observers were indeedreporting the first figure-ground organization theyperceived, as we had assumed. Driver and Baylis(1995) suggested that our observers might have beenresponding to some implicit demand to try to findfamiliar objects in the figure-ground displays. If so,they might have reversed the first figure-groundorganization of the displays in search of familiarobjects, and may have reported them when theyfound them. In the latter case, our direct reportevidence could not be taken as inconsistent with thefigure-ground first assumption. That Peterson andKim (2001a) obtained evidence for the inhibition ofobject memories matched by the more weakly cuedside of a border even though the experimental taskdid not direct participants to make figure reportsregarding the silhouettes provides convergingevidence that, contrary to the figure-ground-firstassumption, memories of object structure areaccessed in the course of figure assignment.

Before the Peterson and Kim (2001a) resultscould be taken to reflect the cross-border inhibitionas predicted by PIMOCA, a few questions remainedto be addressed. One question stems from the factthat a known object was sketched on the white side ofthe silhouette primes shown on experimental trials,but not those shown on control trials. As aconsequence, more cross-border competitionoccurred for experimental than control silhouettes.This increased competition may have led to longerresolution times for experimental silhouettes than forcontrol silhouettes. The differences in RTs mayreflect differences in the time required to resolve thefigural status of the silhouette primes rather thandifferences in the state of the object memorymatching the line drawing itself.

Peterson, Skow Grant, and Kim (submitted; SkowGrant, Peterson, & Kim, 2002) tested this alternativeresolution time hypothesis against the inhibition

hypothesis by altering Peterson and Kim’s (2001a)design such that known objects were sketched on thewhite sides of both experimental and controlsilhouettes shown before familiar line drawings.Whereas the known objects sketched on the whiteside of the silhouettes shown on experimental trialswere from the same basic level category as theirpaired line drawings, the known objects sketched onthe white side of the silhouettes shown on controltrials were from a different category (e.g., livingversus non-living) than their paired line drawings.Thus, in this experiment, the competition for figureassignment was equated for all silhouettes precedingline drawings of familiar objects. The time requiredto resolve the figure assignment in the silhouettesshould be equated as well. (As in Peterson andKim’s experiment, no familiar objects were sketchedalong the borders of the silhouettes shown before linedrawings of novel objects.)

Figure 8. Latency differences between accurate responses to linedrawings of familiar objects shown on control and experimentaltrials obtained by Peterson and Kim (2001a) (A) and Peterson et al(submitted) (B). In (B) the results obtained replicating Petersonand Kim’s experiment under masking conditions (i.e., no knownobject was sketched on the ground side of the control primes) areshown on the left. The results obtained using the new controlcondition are shown the right. Negative difference scores indicatethat RTs were longer for experimental trials than for control trials.

Peterson et al. (submitted) reasoned that if theslower responses to experimental line drawings thanto control line drawings reported by Peterson andKim (2001a) reflected longer resolution times for thesilhouettes shown on experimental versus controltrials, then that pattern of results should not beobtained in their experiment. Indeed, there should beno differences in the latencies to respond to familiar

line drawings shown on experimental versus controltrials. However, if the slowed responses to Petersonand Kim’s (2001a) experimental line drawingsreflected the inhibitory component of PIMOCA (andif inhibition is specific to the category of the knownobject sketched in ground), then responses to linedrawings shown on experimental trials should beslower than responses to line drawings shown oncontrol trials.

In the experiment designed to examine theresolution time hypothesis, silhouettes weredisplayed for 35 ms and were followed by a 70 msmask (to ensure that participants could not use thesilhouette to predict the line drawing type). Becausethey added a mask to the sequence of stimulipreceding the line drawings, Peterson et al(submitted) tested two groups of observers. Onegroup was tested with a control condition like thatused by Peterson and Kim (2001a) (i.e., for theseparticipants the contours of the control silhouettes didnot sketch a familiar object on the ground side). Asecond group was tested with the new controlcondition (in which the contours of the controlprimes sketched an object from a different categorythan the line drawing shown afterwards). Includingboth of these conditions allowed Peterson et al. tocompare the magnitude of the difference scoresobtained with the different types of control primesunder similar presentation conditions.

The results were consistent with the inhibitionhypothesis rather than with the resolution timehypothesis. As can be seen in Figure 8B, RTs onexperimental trials were longer than RTs on controltrials for both groups of observers, even though thecompetition for the borders of experimental andcontrol silhouettes was equated for the observers inthe new control condition, whereas it was not equatedfor observers in the Peterson and Kim (2001a)control condition. The differences between theresults obtained using the two different controlconditions were not statistically significant. It appearsthat any differences in the competition occurring forexperimental versus control primes is not evident inresponses to the line drawings used in theseexperiments.

An alternative interpretation arising from anattentional framework remained to be consideredbefore these results could be taken as supporting thePIMOCA model, however. Suppose that the longerRTs obtained on experimental trials compared tocontrol trials reflect the fact that participants ignoredthe silhouette primes. After all, the silhouette primeswere irrelevant to the participants’ task, whichconcerned the line drawings. Milliken, Joordens,Merikle, and Seiffert (1998) showed that whenobservers ignored primes shown immediately before

-25

-15

-5

5

15

Co

ntr

ol -

Exp

. Lat

ency

Dif

f. (

ms)

-25

-15

-5

5

15

Co

ntr

ol -

Exp

. Lat

ency

Dif

f. (

ms)

B..

P&KControl

NewControl

A. Exp. 1 Exp. 2

target stimuli, they responded more slowly tomatched than to mismatched target stimuli. On thisalternative attention hypothesis, the withdrawal ofattention from the silhouette primes accounts for theRT differences, rather than the fact that the side ofthe border where the known object was sketched wasseen as the ground. In other words, the slowedresponses to the line drawings may not have reflectedthe fact that the memory of object structure cue wasaccessed on the more weakly cued side of a border.They may simply have reflected the fact that theprime was ignored.

To test the attention hypothesis, we altered thesilhouette primes so that the regions seen as groundsin the silhouettes used by Peterson and Kim (2001a)and Peterson, et al. (submitted) would now be seen asfigures. We report this experiment here. If theattention hypothesis is correct, the RTs should beslower on experimental trials than on control trialseven when memories of object structure matching theexperimental line drawings were accessed by regionsdetermined to be figures rather than grounds in theprime. Alternatively if the previous results reflect theground status of the side of the contour where theknown object was sketched, they will not bereplicated here. Indeed, a prediction generated fromnumerous priming experiments conducted by others(e.g., Dell’Acqua & Grainger, 1999) is that RTs willbe faster when line drawings are preceded by figuresportraying an object from the same category.

We created new figure silhouette primes from thesilhouette primes used in the previous experiments(henceforth called “figure” primes and “ground”primes, respectively). Sample figure primes areshown in Figure 9 along with the ground primes fromwhich they were generated. Figure primes werematched to ground primes on a number ofdimensions that could influence the results. Forinstance, the contour sketching the known object waspresented in approximately the same location in thefigure primes as it had been in the ground primes.This was important in case differences between thelocations of the known objects sketched in the primeversus the line drawing affected the magnitude of thepriming. In addition, we took care to portray the sameportion of the known object in the figure prime aswas portrayed in its associated ground prime. Thiswas important because if the new “figure” primesportrayed the entire object whereas the old “ground”primes portrayed only a portion of the known objectin the ground, then any differences in the resultsmight reflect those differences in the amount of theobject portrayed rather than the change from groundto figure status of the prime. In order to portray aportion of a known object effectively as a figureprime without introducing any spurious edges that

could interfere with recognition (Gerbino &Salamaso, 1987), we added gray boxes to the figureprimes positioned in such a way that they wouldappear to be occluding the rest of known object.

Figure 9. Sample figures primes are shown on the bottom and the“ground primes” from which they were generated are shown abovethem. A known object depicted is a face profile on the left, and ananchor on the right. The figure primes were created to match theground primes on several dimensions including: location andportion of known object visible. The gray boxes were intended toportray surfaces that might be occluding the rest of the knownobject.

In these experiments, half of the figure primesportrayed portions of known objects; the other halfportrayed portions of novel objects. We did not maskthe primes in these experiments, so we expected thatobservers might see the differences between thefigure primes portraying known objects versus novelobjects. Therefore, we designed this experiment sothat line drawings of both familiar and novel objectswere preceded equally often by primes portrayingfamiliar and novel figures.

As in the previous experiments, our predictionsconcern responses to the familiar line drawings. Thefamiliar line drawings were divided into experimentaland control sets based on whether they were precededby figure primes portraying known objects from thesame basic level category as the line drawing orfigure primes portraying novel objects, respectively.8

If the delayed responding found on experimentalversus control trials in the previous experiments wasa consequence of inhibition induced by ignoring theprimes, then we should obtain the same pattern ofresults using figure primes rather than ground primes.This is because, as in the Peterson and Kim (2001a)experiments, on experimental trials, the objectportrayed in the figure prime matches the basic levelcategory of the object portrayed in the line drawingwhereas there is no match on control trials. However,if the previous results reflect the inhibition of thememory of object structure cue accessed on the sideof the prime seen as the ground, we should not

observe longer RTs on experimental trials than oncontrol trials in experiments using figure primes.Instead responses might now be faster onexperimental trials compared to control trials.

We conducted two experiments using figureprimes, using slightly different exposure durations. Inthe first experiment, the figure prime was displayedfor 50 ms and was followed by an ISI of 33 ms. Inthe second experiment, the figure prime wasdisplayed for 35 ms, followed by an ISI of 35 ms.9 Inboth experiments, the line drawing was shown afterthe ISI; it remained on the screen for 646 ms in thefirst experiment and for 660 ms in the secondexperiment.

As can be seen in Figure 10, our results provideno support for the attention hypothesis. An ANOVAconducted on the RTs for correct responses tofamiliar line drawings showed that, in contrast to theresults reported by Peterson and Kim (2001a),responses on experimental trials were faster thanresponses on control trials, F(1, 27) = 4.36, p< .05 forthe first experiment, and F (1, 33) = 4.00, p = .054 forthe second experiment. Thus, on experimental trials,when the figure prime portrayed an object from thesame basic level category, responses to the target linedrawings were faster than on control trials where theprime was a novel figure. The silhouette primes wereequally irrelevant in these experiments as they werein the Peterson and Kim (2001a) experiments; yethere RTs were faster on experimental trials than oncontrol trials. Therefore, it does not appear to be thecase that the irrelevance of silhouette primes isresponsible for the slower RTs recorded onexperimental trials versus control trials by Petersonand Kim (2001a) and by Peterson, et al. (submitted).The critical difference between the presentexperiments and the previous experiments appears tobe that the matching known objects were sketched onthe figure side of the border of the prime in thepresent experiments and on the ground side of theborder of the prime in the previous experiments.

Figure 10. Latency differences between accurate responses to linedrawings of familiar objects shown on control and experimentaltrials we obtained in experiments using figure primes. Positivedifference scores indicate that the RTs were shorter onexperimental trials than on control trials.

Based upon the results of the experiments wehave summarized here, including the new experimentutilizing figure primes, we are confident that theslower RTs obtained on experimental trials byPeterson and Kim (2001a) and by Peterson et al.(submitted) reflect the cross-border competition andinhibition proposed in PIMOCA. Thus, theseexperiments provide support for PIMOCA; especiallyfor the proposals that configural cues (includingmemory of object structure) lying on opposite sidesof a border compete and that cues on the relativelyweakly cued side of the border are inhibited.Therefore, it is conceivable that cross-borderinhibition accounts for the apparent shapelessness ofground regions in the vicinity of more strongly cuedfigures.

VI. Learning: How much past experience isnecessary before memory for the structure of anobject can affect figure assignment?

In our initial work investigating whether or notobject memories affected figure assignment, we usedstimuli in which well known objects were sketchedalong one side of a border (e.g., objects such asstanding women, table lamps, guitars, etc.). On thebasis of those experiments, we knew that memoriesof objects could be accessed in the course of figureassignment, but we did not know how much pastexperience was required with an object beforememory of its structure could affect figureassignment.

We avoided the learning question in part becauseother research using initially novel displays hadfailed to find any influence from past experience onfigure assignment following a single past exposure tothe novel object (e.g., Rock & Kremen, 1957). Inthose previous experiments, investigators had testedfor effects of past experience on figure assignmentsome time after the experience was induced.Therefore, the results confounded questionsconcerning how long memories of novel objects lastwith questions concerning whether or not pastexperience affects figure assignment. In addition,Rock and Kremen measured direct reports aboutfigure and ground relations; they did not record RTs,which might have permitted them to assess whetheror not memories of newly learned objects competefor figural status with other cues, even if they do notdominate them.

Recently, Peterson and Lampignano (2003) foundthat a single prior exposure to a novel shape wassufficient to observe its influence on figureassignment the next time a portion of the border ofthe shape was encountered. They obtained these

-25

-15

-5

5

15

Co

ntr

ol -

Exp

. Lat

ency

Dif

f. (

ms)

Exp. 1 Exp. 2

results using a paradigm initially used by Treismanand DeSchepper (1996; Experiment 6). Treisman andDeSchepper had adapted a negative primingparadigm for use with novel displays. Using thisparadigm, they obtained some results that they tookto be evidence that, even though the ground of anovel figure-ground display was phenomenologicallyshapeless, its shape was nevertheless stored in visualmemory along with an “ignore” tag. Peterson andLampignano thought that Treisman andDeSchepper’s results could be better interpretedwithin PIMOCA than within a negative primingframework. In particular, Peterson and Lampignanothought that Treisman and DeSchepper’s resultsmight show that a single prior experience with anovel figure was sufficient to establish a memory thataffected figure assignment the next time the borderwas encountered. We describe Treisman andDeScehpper’s paradigm in some detail so thatPeterson and Lampignano’s variant of it, and thealternative conclusion they reached, can beunderstood.

Treisman and DeSchepper (1996, Experiment 6)showed observers paired prime-probe trials. On thefirst, "prime," trial, an ambiguous figure-grounddisplay was shown on a gray field above a fixationcross. (See Figure 11A.) The ambiguous display hada central articulated border shared by a black regionand a white region. Observers were instructed tomatch the (standard) black region in the figure-ground display shown above fixation to a blackcomparison shape shown below fixation. Theyassumed that, in order to perform the shape-matchingtask, observers perceived the black region as theshaped figure and the white region as the shapelessground in the prime figure-ground display.

On the next, "probe," trial, two separated shapes,one black and one white appeared above fixation, anda second white comparison shape appeared belowfixation (See Figure 11B.) The left-right arrangementof the black and white shapes above fixation was thesame as that of the black and white regions in theprime figure-ground display. On the probe trials,however, the two shapes above fixation did not shareany borders. The observers' task on probe trials wasto determine whether or not the standard white shapeshown above fixation was the same as thecomparison white shape shown below fixation. (Theblack shape shown on the probe trials was a distractorwith a novel border.) On experimental probe trials,the standard white shape was the white regionisolated from the prime figure-ground displays (theregion that was perceived as a shapeless ground onthe prime trial).10 On control probe trials, thestandard (and comparison) white shapes had novelborders that had not been seen previously.

Figure 11. A. The prime display used by Treisman andDeSchepper (1996, Experiment 6). B. Sample probe displays. Halfof the probe trials were experimental trials; the other half werecontrol trials. On experimental probe trials, the white “standard”probe shape shown above fixation was the same as the boundedwhite ground region of the prime figure-ground display. On controlprobe trials, the standard white shapes were novel shapes. On halfof the probe trials of both types, the white shapes shown above andbelow fixation were the same; on the other half of the trials, theywere different. In the experiment, a given distractor was seen onlyonce, and a given white shape was seen on only one probe trial.The shapes are repeated here for illustrative purposes only. C. Theprime display used by Peterson and Lampignano (2003).

In Treisman and DeSchepper’s experiment (1996;Experiment 6), observers took longer to respond onexperimental probe trials than on control probe trials.These results led them to conclude that, before figureand ground are determined, equivalent memories areestablished for the whole shapes of the figure and theground, regardless of the fact that these regions areperceived quite differently (i.e., the figure isperceived to be shaped by the central border whereasthe ground is perceived to be shapeless in the vicinityof that border). Treisman and DeSchepper explainedthe fact that they obtained longer latency responseson experimental compared to control probe trials asreflecting an “ignore” tag attached to the memory ofthe shape of the ground.

Peterson and Lampignano (2003) pointed out that,in reaching this conclusion, Treisman andDeSchepper (1996) neglected to consider a criticalaspect of their displays. As can be seen in Figure 12,when the shape of the region seen as ground wasextracted from the prime display and repeated on theprobe display, the shape of the region seen as thefigure in the prime was necessarily sketched alongthe outside of its articulated border. Therefore, anyslowing on experimental trials compared to controltrials may just as well have reflected competitionfrom a memory of a portion of the figure seen on theprime trial as an ignore tag attached to the shape ofthe ground. Peterson and Lampignano favored theformer interpretation because it is consistent with acompetitive model of how figure assignment occurs,such as PIMOCA, and because of its excitingimplications that one previous experience with anobject was sufficient to establish a memory that couldexert an influence on figure assignment. They did notfavor the latter interpretation (Treisman andDeSchepper’s interpretation), both because it did nottake the process of figure-ground segregation intoaccount, and because it implied an inconceivablylarge capacity for shape memory. In order todistinguish these two interpretations for the originalresults Peterson and Lampignano (2003) changedTreisman and DeSchepper’s (1996) design in twoways.

Figure 12. A sample prime on the left and an experimental probeon the right. The probe has been altered to highlight the fact that aportion of the shape of the black prime figure is sketched along thegray side of the border of the white standard probe shown onexperimental trials.

First, they decreased the similarity between whatTreisman and DeSchepper would consider the globalshape of the ground on the prime trial and the whitestandard shape shown on the probe trial. They didthis by removing the white region from the primefigure-ground display and by presenting the blackregion alone on the larger gray ground in the standardprime display (see Figure 11C). Their probe displayswere the same as those used by Treisman andDeSchepper (1996, Experiment 6). In the probedisplay, the standard was a closed white shape.

Except for the repetition of the articulated border ofthe prime, the shape of the standard probe was quitedifferent from the ground in the prime display.

Priming effects are larger when the shapes ofprime and probe stimuli are the same rather thandifferent. Therefore, Peterson and Lampignanoreasoned that this manipulation would diminish thelatency differences between experimental and controltrials if those differences reflect memory for theshape of the ground, as Treisman and DeSchepperclaimed. Alternatively, this manipulation should notdiminish the latency differences if those differencesreflect cross-border competition for figural status. Allthat is necessary for competition is the repetition ofthe border of the prime figure on the probe trial. Thecompetition hypothesis predicts that the memory ofthe structure of the figure seen on the prime trial willcompete with the cues favoring seeing the figure onthe inside of the probe shapes. This competitionmight increase the time required to resolve the figuralstatus of the experimental probes, and consequently,could be responsible for the longer RTs observed onexperimental probe trials compared to control probetrials. Note that the competition hypothesis does notrequire that memories of the structure of the figureseen on the prime trial dominate the perception of theprobe stimuli. More cues favor the interpretation thatthe figure lies on the inside (white side) than theoutside of the border of the probe display.

Second, Peterson and Lampignano (2003)attempted to obtain evidence for the competitionhypothesis by examining the consequence of adding asecond cue that favored assigning the repeatedarticulated border to the outside rather than to theinside of the standard white probe. This second cue --partial closure – was manipulated by positioning thedistractor near to or far from the white probe shape,as shown in Figure 13. Partial closure is a variant ofthe Gestalt configural cues of closure. Gillam (1975)had shown that partial closure served as a groupingcue; Peterson and Lampignano reasoned that it mightalso serve as a figural cue.

Peterson and Lampignano (2003) found robustslowing on experimental compared to control trials,despite the decreased similarity between the probeshape and the prime ground. They also found that thedistance to the distractor mattered more in theexperimental condition than in the control condition,suggesting that the addition of another cue, partialclosure, favoring assigning the border to the sameside as the shape memory cue increased thecompetition for the border. A second experimentshowed that the mere repetition of the border of thefigure seen on prime trials was sufficient for theseeffects; the presence of distractors was not necessary.

Figure 13. The near- and far-distractor conditions. The distancesshown in this figure are approximations of those used in theexperiment.

Thus, it seems that the Treisman and DeSchepper(1996, Experiment 6) results are better interpretedwithin a competitive model of figure assignment thanwithin a negative priming paradigm. Peterson andLampignano’s results show that a single pastexposure is sufficient to establish a memory thatenters into the competitive figure assignment processthe next time the border is encountered. Futureresearch will have to test how long this memory lastsand whether its longevity is affected by exposure toother, similar, novel shapes. In the paradigm used byPeterson and Lampignano, the interval betweenpresentation of the novel stimulus and test was on theorder of 1700 ms. Longer intervals (with and withoutthe introduction of new stimuli) must be tested inorder to determine how long these new memorieslast.

Peterson and Lampignano (2003) interpretedtheir results as evidence of cross-border competitionrather than cross-border inhibition for a number ofreasons. First, the SOAs they used were much longerthan those at which consequences of cross-borderinhibition have been observed. The longest SOA overwhich Peterson and Kim (2001a) and Peterson et al.(submitted) observed inhibitory effects was 105 ms;they failed to find evidence for inhibition using SOAsof 200ms, 350ms, 500ms, and 650ms. And,Treisman and DeSchepper (1996, Experiment 6)obtained longer latencies on experimental probesthan on control probes even when the experimentalprobes were shown three trials after their associatedprimes. The cross-border inhibition predicted byPIMOCA is expected to be short-lived, and therefore,unlikely to be observed over long SOAs. In contrast,new object memories may last (and can potentiallyinfluence figure assignment) for an unlimitedduration of time. Second, the articulated bordershown in the prime display was repeated on the probedisplay in Peterson and Lampignano’s experimentsand in Treisman and DeSchepper’s experiment,whereas it was not repeated in the experiments

conducted by Peterson and Kim (2001a; Peterson etal., submitted; Skow Grant et al., 2002). Thus,Peterson and Lampignano assayed memory for aparticular novel border that had been seen only oncebefore the probe trials (i.e., on the prime trial),whereas Peterson and Kim (2001a; Peterson et al.,submitted; Skow Grant et al., 2002) assayed theconsequences of accessing pre-existing memories ofportions of similar basic-level objects. Themechanisms mediating short-lived inhibition andmemory for past experience with a previously seenborder may be different. They certainly seem tofollow a different time course. Future experimentswill investigate the relationship between inhibitionand competition in more depth.

VII. Concluding Remarks

The body of research reviewed here shows thatpast experience affects figure assignment. Onereason many scientists sought to exclude pastexperience from inclusion among factors that mightaffect early perceptual processes was the belief that,were past experience to have an effect, it wouldnecessarily dominate other cues. The cue competitionexperiments show that object memories do not exerta dominating influence; instead they constitute justone more among many configural cues used by thevisual system.

The results showing that past experience doesaffect figure assignment raised a number ofquestions that had been answered under the oldfigure-ground-first assumption. Peterson and hercolleagues offered new answers to these questions inthe form of the Parallel Interactive Model ofConfigural Analysis (PIMOCA). They providedsome empirical support for predictions arising fromthe model. But the model must be tested furtherbefore its full value can be known.

The surprising results reviewed in the last sectionshowing that a single past experience with a border issufficient to establish a memory that is accessed thenext time the border is encountered suggest thatmemories of object structure are remarkably plastic.These results were observed in RT measures; theywould not have been evident in direct reportsregarding what was seen as figure because the pastexperience cue did not win the cross-bordercompetition. Thus, these experiments attest to theimportance of using measures that can reveal thecourse of figure assignment rather than simply itsoutcome.

The research reviewed here opens up manyavenues for future research using computational,

physiological, and behavioral techniques. Oneimportant question is where in the stream of visual-cognitive processes these memories lie, as well aswhere the configural cues are assessed. The answer tothese questions will be valuable, not for finding theplace to draw a line dividing visual perception andmemory, but rather for understanding both the natureof object memory and the nature of the interactionsthat determine figure assignment.

References

Dell’Acqua, R. & Grainger, J. Unconscious semanticpriming from pictures. Cognition 73, B1-B15.

Driver, J., & Baylis, G. C. (1995). One-sided edgeassignment in vision: 2. Part decomposition,shape description, and attention to objects.Current Directions in Psychological Science, 4,201-206.

Gerbino, W. & Salmaso, D. (1987). The effect ofamodal completion on visual matching. ActaPsychologica, 65, 25-46.

Gibson, B. S., & Peterson, M. A. (1994). Doesorientation-independent object recognitionprecede orientation-dependent recognition?Evidence from a cueing paradigm. Journal ofExperimental Psychology: Human Perception andPerformance, 20, 299-316.

Gillam, B. (1975). The evidence for “closure” inperception. Perception & Psychophysics, 17, 521-524.

Hochberg, J. (1962). The psychophysics of pictorialperception. Audio-Visual Communication Review,10, 22-54.

Hochberg, J. (1971). Perception I: Color and shape.In J. W. Kling & L. A. Riggs (Eds.), Woodworthand Schlossberg's Experimental Psychology, (3rded., pp. 395-474). New York: Hold, Rinehart, &Winston.

Jolicoeur, P. (1988). Mental rotation andidentification of disoriented objects. CanadinaJournal of Psychology, 42, 461-478.

Kanizsa, G. & Gerbino, W. (1976). Convexity andsymmetry in figure-ground organization. In M.Henle, (Ed.), Vision and Artifact. NY: SpringerPublishing Co.

Kienker, P. K., Sejnowski, T. J., Hinton, G. E., &Schumacher, L. E. (1986). Separating figurefrom ground in a parallel network. Perception,15, 197-216.

Kim, J. H. & Peterson, M. A. (2001, May).Contextual modulation of the strength of theGestalt configural cues. Poster session at the firstannual meeting of the Vision Sciences Society,Sarasota, FL.

Kim, J. H., & Peterson, M. A. (2002, May).Contextual modulation of the strength of Gestaltconfigural cues. Poster session at the secondannual meeting of the Vision Sciences Society,Sarasota, FL.

Marr, D. (1982). Vision. NY: W. H. Freeman andCompany.

Milliken, B., Joordens, S., Merikle, P. M., & Seiffert,A. E. (1998). Selective attention: A reevaluationof the implications of negative priming.Psychological Review, 105, 203-229.

Pasupathy, A., & Connor, C. E. (1999). Responses tocontour features in macaque area V4. Journal ofNeurophysiology, 82, 2490-2502.

Perrett, D., Oram, M. W., & Ashbridge, E. (1998).Evidence accumulation in cell populationsresponsive to faces: An account of generalizationof recognition without mental transformations.Cognition, 67, 111-145.

Peterson, M. A. (1994). Shape recognition can anddoes occur before figure-ground organization.Current Directions in Psychological Science, 3,105-111.

Peterson, M. A. (1995). The relationship betweendepth segregation and objectrecognition: Oldassumptions, new findings, and a new approach toobject recognition. Unpublished manuscript.

Peterson, M. A. (1999). On the role of meaning inorganization. Intellectica, 28, 37-51.

Peterson, M. A. (2000). Object perception. In E. B.Goldstein (Ed.), Blackwell Handbook ofPerception, Chapter 6, pp. 168-203.

Peterson, M. A. (2002). On figure, grounds, andvarieties of amodal surface completion. InR.Kimchi, M. Behrmann, & C. Olson (Eds.),Perceptual Organization in Vision: Behavioraland Neural Perspectives. Hillsdale, NJ: LEA.

Peterson, M. A. (2003). Overlapping partialconfigurations in object memory: An alternativesolution to classic problems in perception andrecognition. In M. A. Peterson & G. Rhodes (Eds.),Analytic and Holistic Processes in the Perceptionof Faces, Objects, and Scenes. NY: OxfordUniversity Press.

Peterson, M. A., de Gelder, B., Rapcsak, S. Z.,Gerhardstein, P. C., and Bachoud-Lévi A. C.(2000). Object memory effects on figureassignment: Conscious object recognition is notnecessary or sufficient. Vision Research, 40,1549-1567.

Peterson, M. A., Gerhardstein, P. C., Mennemeier,M., & Rapcsak, S. Z. (1998). Object-centeredattentional biases and object recognitioncontributions to scene segmentation in righthemisphere- and left hemisphere-damagedpatients. Psychobiology, 26, 557-570.

Peterson, M. A., and Gibson, B. S. (1991). Directingspatial attention within an object: Alteringthefunctional equivalence of shape description.Journal of Experimental Psychology: HumanPerception and Performance, 17, 170-182.

Peterson, M. A., and Gibson, B. S. (1993). Shaperecognition contributions to figure-groundorganization in three-dimensional displays.Cognitive Psychology, 25, 383-429.

Peterson, M. A., and Gibson, B. S. (1994a). Objectrecognit ion contr ibut ions to f igure-groundorganization: Operations on outlines andsubject ive contours . Perception &Psychophysics, 56, 551-564.

Peterson, M. A., and Gibson, B. S. (1994b). Mustshape r ecogn i t i on fo l l ow f i gu re -groundorganization? An assumption in peril.Psychological Science, 5, 253-259.

Peterson, M. A., Harvey, E. H., & Weidenbacher, H.L. (1991). Shape recognition inputs tofigure-ground organization: Which route counts?Journal of Experimental Psychology: HumanPerception and Performance, 17, 1075-1089.

Peterson, M. A., and Hector, J. E. (1996, November).Evidence for the Piecemeal Nature ofPre-depthObject Recognition Processes. Paper presented atthe Annual Meeting of the Psychonomic Society,Chicago, IL.

Peterson, M. A., & Kim, J. H. (2001a). On what isbound in figures and grounds. VisualCognition. Special Issue: “Neural binding ofSpace and Time, 8, 329-348.

Peterson, M. A., & Kim, J. H. (2001b). ContextModulates the Gestalt Configural Cue ofConvexity. Paper presented at the AnnualMeeting of the Psychonomic Society, Orlando,FL.

Peterson, M. A., & Lampignano, D. W. (2003).Implicit memory for novel figure ground displaysincludes a history of border competition. Journalof Experimental Psychology: Human Perceptionand Performance, 29, 808-822.

Peterson, M. A., Skow Grant, E., & Kim, J. H.( submi t ted) . Inhibi t ion of objectmemoriesaccessed on the side of borderperceived to be ground. Manuscript submitted forpublication.

Riddoch, M. J., & Humphreys, G. (1993).Birmingham Object Recognition Battery. (test6).Washington, D. C.: Psychology Press.

Rock, I., & Kremen, I. (1957). A re-examination ofRubin's figural after-effect. J o u r n a lofPsychology, 53, 23-30.

Rubin, E. (1958). Figure and ground. In D.Beardslee & M. Wertheimer (Ed. And Trans.),Readings in perception. (pp. 35-101). Princeton,

NJ: Van Nostrand. (Original work published in1915.)

Schafer, R., & Murphy, G. (1943). The role ofautism in a visual figure-ground relationship.Journal of Experimental Psychology, 32, 335-343.

Sejnowski, T. J., & Hinton, G. E. (1987). Separatingfigure from ground in a Boltzman machine. In M.Arbib & A. Hanson (Eds.), Vision, brain, andcooperative computation. Cambridge, MA: MITPress.

Skow Grant, E., Kim, J. H., & Peterson, M. A. (2002,May). Tests of a competitive interactive model offigure assignment. Poster session at the SecondAnnual Vision Sciences Society Meeting,Sarasota, FL.

Tanaka, K. (1996). Inferotemporal cortex and objectvision. Annual Review of Neuroscience, 19, 109-139.

Tarr, M. J., & Pinker, S. (1989). Mental rotation andorientation-dependence in shaperecognition. Cognitive Psychology, 21, 233-282.

Treisman, A., & DeSchepper, B. (1996). Objecttokens, attention, and visual memory. In T. Inuiand J. McClelland (Eds.), Attention andPerformance, XVI: Information Integration inPerception and Communication. (pp. 15-46).Cambridge, MA: MIT Press.

Vecera, S. P., & O'Reilly, R. C. (1998). Figure-ground organization and object recognitionprocesses: An interactive account. Journal ofExperimental Psychology: Human Perception andPerformance, 24, 441-462.

Warrington, E. K., & James, M. (1991). The VisualObject and Space Perception Battery. Suffolk,England: Thames Valley Test Company.

Warrington, E. K., & Taylor, A.M. (1973). Thecontribution of the right parietal lobe to objectrecognition. Cortex, 9, 152-164.

Footnotes 1 There are some situations in which both regions canbe perceived as figures, and there are other situationsin which a contour itself can be perceived as thefigure. However, the most likely outcome is a figure-ground outcome.

2 Some tests of this assumption were attempted butthey were neither straighforward nor stringent. TheGestalt point of view was the Zeitgeist; consequently,evidence consistent with the Gestalt view was soughtand obtained. See Peterson (1995, 1999).

3 Hence, there is no need to distinguish betweencontours that are intrinsic versus extrinsic to theobject before object memories are accessed (seePeterson, 2003). This is important because the figure-ground first assumption has been used to separatesuch contours.

4 The traditional view can account for why familiarshapes can’t be seen in grounds. However, it neverwent far enough to account for why even novelgrounds appear locally shapeless.

5 In any competitive system, cues will inevitably beunbalanced. In PIMOCA, any slight advantage forthe cues on one side of the edge will be amplified bythe cooperative and competitive interactions.

6 Even if observes had been able to distinguishbetween experimental and control silhouettes, theycould not have predicted the response to thesubsequent line drawing, since control silhouettesappeared before half the line drawings of familiarobjects.

7 Only observers who responded quickly showedthese effects. Results obtained from observers whose

RTs on control trials exceeded a threshold set by theexperimenters were excluded from the analysis. (Fordetails, see Peterson & Kim, 2001a.)

8 For this experiment, the familiar objects weredivided into three sets so that none of the control linedrawings portrayed an object from the same basiclevel category as any of the figure primes. In anygiven experiment, two sets were shown as linedrawings, one as control and one as experimental.The experimental line drawings were preceded bysilhouette primes portraying an object from the samebasic level category. The third set of objects wasshown as figure silhouettes before novel linedrawings. These three sets were balanced acrossthese conditions.

9 The difference in the length of time for the inter-stimulus interval (ISI) was due to computer monitorreplacement between the experiments.

10 On same experimental probe trials, the whiteregion isolated from the prime figure-ground displaywas shown both above and below fixation.

memory and learning in figure-ground perceptionmapeters/mp pubs/petersonskow03.pdf ·...

Documents