induction of kanizsa contours requires awareness of the ... · 7/14/2016 · 86 a kanizsa...

1

InductionofKanizsacontoursrequiresawarenessoftheinducingcontext1

2

TheodoraBanica1&DietrichSamuelSchwarzkopf1,2*3

4

1.ExperimentalPsychology,UniversityCollegeLondon,26BedfordWay,London,UK5

2.InstituteofCognitiveNeuroscience,UniversityCollegeLondon,17QueenSquare,6

London,UK7

8

*Towhomallcorrespondenceshouldbedirected:[email protected]

10

Abstract11

12

Itremainsunknowntowhatextentthehumanvisualsysteminterpretsinformation13

about complex scenes without conscious analysis. Here we used visual masking14

techniquestoassesswhetherillusorycontours(Kanizsashapes)areperceivedwhen15

the inducing context creating this illusion does not reach awareness. In the first16

experiment we tested perception directly by having participants discriminate the17

orientationofan illusorycontour. In thesecondexperiment,weexploitedthe fact18

that the presence of an illusory contour enhances performance on a spatial19

localization task. Moreover, in the latter experiment we also used a different20

maskingmethodtoruleouttheeffectofstimulusduration.Ourresultssuggestthat21

participants do not perceive illusory contours when they are unaware of the22

inducingcontext.This isconsistentwiththeoriesofamultistage,recurrentprocess23

ofperceptualintegration.Ourfindingsthuschallengesomereports,includingthose24

from neurophysiological experiments in anaesthetized animals. Furthermore, we25

discusstheimportancetotestthepresenceofthephenomenalperceptdirectlywith26

appropriatemethods.27

28

Keywords29

30

Illusorycontours,Kanizsa,modalcompletion,awareness,masking31

32

.CC-BY-NC 4.0 International licenseavailable under anot certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint (which wasthis version posted July 14, 2016. ; https://doi.org/10.1101/050526doi: bioRxiv preprint

https://doi.org/10.1101/050526

http://creativecommons.org/licenses/by-nc/4.0/

2

33Introduction34

35

What role does conscious processing of the environment fulfill and how much36

processingoccurs in theabsenceofawareness? It is self-evident thatmuchof the37

internal bodily functions and the learned motor behaviors, such as walking or38

driving,operatemostlywithoutawareness.Butforprocessingthroughtheclassical39

senses,likevision,therehavebeenwidelydiscrepantfindingsonhowmuchstimulus40

processing can occur and how it affects decision-making when the subject is41

unaware of the stimulus. Moreover, the approach to be used when studying42

unconsciousstimulusprocessinghasalsobeensubjectofcontroversy[1].43

44

Several experiments suggest that the effect of contextual stimuliwithin a target,45

such as the percept of visual illusions or adaptation effects, persists even when46

participantsareunawareofthepresentedcontextualinformation[2–6].Theuseof47

continuous flash suppression (CFS), in which a dynamic, high-contrast stimulus is48

presentedtooneeyetosuppressthestimulusintheothereyefromawareness,has49

become a popular way to probe unconscious stimulus processing [7]. Using this50

procedure ithasbeenclaimed that theperceptionofphysical facial attributes [8],51

the complex analysis of naturalistic scenes [9], and even linguistic processing and52

arithmetic canbeperformedwithout awareness [10]. However, several of these53

findings have recently failed to be replicated and were challenged on theoretical54

grounds[11–13].55

56

Neuroimaging experiments showed that while both simple and more complex57

stimuli haveaneural signature in the visual cortexundermasking conditions [14–58

16], the encodingof unconscious stimuli appears tobequalitatively different.Not59

only is theoverall response tounconsciousvisual stimuliweaker [15]butcoupling60

betweendifferentstagesinthevisualprocessinghierarchyisalsoreduced[17]and61

the information content differs [16,18,19]. In particular, the response to these62

stimuli ismore variable [20], and also localized inmore posterior regions than to63

conscious stimuli [18]. . One reason for this variability could be that only simple64



https://doi.org/10.1101/050526


3

positionalorgeometric information isprocessed in theabsenceofawareness,but65

that more complex abstraction and perceptual integration requires consciousness66

Underthishypothesis,theneuralencodingofstimuliisnoisybecauselocalstimulus67

interactionsarepreservedbutabstractandthereforecoherentrepresentationsare68

disrupted. Specifically, we used shape stimuli that were either defined by the69

positionortheorientationofsimpleimageelements.Wedemonstratedthatwhen70

suchstimuliwererenderedinvisibleusingfastcounter-phaseflicker(at120Hz)they71

couldspeedupperformanceonashapediscriminationtaskofconsciousstimuli[21].72

Critically,thisprimingeffectwasonlypresentforstimulisharingthesamepositions,73

or – if oriented elements were used for priming – the positions along the path74

implied by the elements. We observed no priming by invisible primes if the test75

stimuluswas smaller than the prime stimulus. This suggests that invisible priming76

operatedlocally,possiblyinretinotopicspaceinearlyvisualcortex,butnoabstract77

integration of individual elements into the concept of a shape occurred without78

awareness.79

80

We further tested whether two brightness illusions manifest when the inducing81

context is rendered invisible by means of CFS [22]. We found that masking the82

context (a smooth gradient in luminance) had little impact on simultaneous83

brightness contrast of two targets that were unmasked. In a stark dissociation,84

participantscouldnotdiscriminatetheorientationofanillusorycontour,definedby85

a Kanizsa triangle, when the inducing context (the ‘Pacman’ shapes whose open86

segments define the corners of the triangle)weremasked selectively by CFS. This87

could indicate that thegenerationof the illusorycontourperceptoccursata later88

stageofvisualprocessingthansimultaneousbrightnesscontrast,eitherintermsof89

the visual hierarchy or in the latency of processing. These findings were also90

consistentwithpreviousreports that illusorycontoursarenotperceivedwhenthe91

inducersaresuppressedfromawarenessduringbinocularrivalry[23].92

93

These findings challenge some previous reports, using psychophysical tests in94

healthy volunteers [24] and even neurophysiological experiments in anesthetized95



https://doi.org/10.1101/050526


4

animals[25]thatsuggestedthatillusorycontourscouldbeformedintheabsenceof96

awareness.However,noneofthesepreviousstudiesspecificallytestedwhetherthe97

experimental participants actually perceived an illusory contour. Similarly,98

neuropsychological studies in neglect patients [26–29] suggested that illusory99

contourprocessingoccurswhenpartof the inducingcontext isplaced in theblind100

hemifield. However, this also does not conclusively support the assertion that101

contoursareformedintheabsenceofanycontextualawareness.102

103

However, two issuesburden the interpretationofpreviousexperiments that show104

no evidence of illusory contours when the inducers are masked. First, there is105

evidence that illusory contours are processed by binocular neurons in early visual106

cortex [30–33]. It does in fact seem unsurprising that the mechanisms inducing107

illusory contours at least partially overlap with those for segmenting surfaces in108

depth[34].Whenretinaldisparityimpliesthatinducersareatdifferentdepthsfrom109

thebackground,thevisualsystemnotonlyproducestheperceptofillusorycontours110

butthesurfaceboundedbyillusorycontoursisalsoperceivedinstereoscopicdepth111

[31]. Thuswhen binocular processing is disrupted or overwhelmed by a dichoptic112

maskorbinocularrivalry,theillusorycontourperceptisalsobroken.Anotherrecent113

studyfurthercomplicatedthissituationbyfindingthatKanizsashapesbrokethrough114

CFSfasterthancontrolstimuli[35].Leavingasideconceptualissueswiththetime-to-115

emergenceparadigm, it is imperative that thedependenceof illusory contourson116

awarenessmustbeconfirmedusingmaskingmethodsotherthanCFS.117

118

A second confoundwith theseprevious studies [22,23] is that even if participants119

perceivedan illusorycontour, thisperceptwas far less salient than real luminance120

contours, as it may have been obscured by the dominant masking stimulus. The121

additionofasimplecontrolconditioncouldremedythisproblem:Oneexperimental122

stimulusshouldbearealcontourdefinedbyasubtleluminancecontrastthatmimics123

the illusory contour percept as closely as possible. If participants can detect and124

discriminatethisstimulusbutareunabletodosofortheillusorycontourcondition,125

thisindicatesthattheillusoryperceptisindeeddisruptedspecifically,ratherthanthe126

moregeneralabilitytodetectsubtlestimuli.127



https://doi.org/10.1101/050526


5

128

In thepresentpaper,wecarriedout twoexperiments toaddress theseconfounds129

andanswerthequestionofwhetherKanizsacontoursareformedwheninducersare130

not consciously perceived. In experiment 1, we used a similar design as in our131

previousstudy[22].Here,participantswereaskedtodiscriminatetheorientationof132

a Kanizsa triangle. However, instead of CFS we employed a temporal masking133

methodtorendertheinducersinvisible.Moreover,weincludedacontrolcondition134

in which a real, luminance-defined contour was present. Because this masking135

methodreliesonverybriefstimulusdurations, inexperiment2wepresented long136

(500ms)stimulirenderedinvisiblebymeansoffastcounter-phaseflicker[3,21].This137

iscriticalbecausetheformationofillusorycontoursarisecomparablyslowly[36–38]138

andthusmaybedisruptedbyafasttemporalmaskingtechnique..Inadditiontothis,139

previous research demonstrated that the presence of illusory contours boosts140

participants’abilitytodiscriminatethepositionofatinytarget[23,39],providinga141

specifictestofwhethertheparticipant infactperceivesan illusorycontourornot.142

Wethereforemeasuredtheabilityofagroupofparticipants,whowerewelltrained143

atpsychophysicaltasks,todiscriminatethepositionofadottargetforKanizsaand144

controlstimulipresentedwithorwithoutmasking.145

146

MaterialsandMethods147

148

Both experiments were carried out at the UCL Department of Experimental149

Psychology.ProceduresadheredtotheDeclarationofHelsinki.Ethicalapproval for150

thisstudywasobtainedfromtheUCLResearchEthicsCommitteeandallparticipants151

gavewritten,informedconsent.152

153

Participantsinexperiments1wererecruitedamongtheUCLstudentpopulation.In154

experiment2werecruitedparticipantswhowerefamiliarwithpsychophysicaltasks.155

AllParticipantshadnormalorcorrected-to-normalvisualacuity.156

157

Allexperimentswereconductedinadark,sound-attenuatedroom.Fortheduration158

of the experimental sessions, participantswere asked to stabilize their head on a159



https://doi.org/10.1101/050526


6

chinrestlocatedatafixeddistanceof48cmfromthestimuluspresentationscreen160

wherestimuliwerepresentedtothembinocularly.161

162

Stimuli were generated by a computer and presented on a 22-inch Samsung163

SM2233RZLCDmonitorataresolutionof1680x1050pixels.Screenrefreshratein164

experiment1wassetto60Hz.Inexperiment2itwassetto120Hz.Theexperiment165

was controlled and behavioral responses were recorded using MATLAB (The166

Mathworks,Inc.)andPsychtoolbox3[40]usingastandardkeyboard.167

168

Experiment1169

170

The experiment comprised two tasks: the first, henceforth called ‘Kanizsa’ task,171

investigatedwhetherparticipantscanperceiveillusorycontourswithoutawareness172

oftheirinducers.Thesecondtask,the‘Visibility’task,assessedtheeffectivenessof173

themaskingtechniquedirectly.174

175

Studydesign176

177

Like our earlier experiment using continuous flash suppression, this experiment178

aimed to measure the perception of illusory contours in a direct manner. We179

implementeda2x3designwithvisibility(invisible,visible)andtypeofstimulus(real,180

illusory,control)aswithinsubjectfactors.181

182

Everyparticipant completed twodifferent tasks. Each task comprised25blocksof183

trials,whereoneblockconsistedof24 trials in theKanizsa taskand8 trials in the184

visibilitytask.Acrossonetask,eachconditionappeared100times.Thevisibilitytask185

onlycomprisedthevisibleandinvisiblecontrolconditions.186

187

Participantsmade behavioral responses in a forced-choice design by button press188

(leftor rightarrow)onastandardcomputerkeyboard.Eachtrial requiredeithera189

leftora right responseandeachresponsetypeappearedtwiceperblock foreach190



https://doi.org/10.1101/050526


7

condition. Conditions were selected pseudo-randomly for every trial but were191

counterbalancedovereachblock.192

193

Participants194

195

Seventeen (13 female; age range: 18-29, mean age: 23.8±2.5) normal, healthy196

participantstookpartintheexperiment.Anadditionaltwoparticipantsweretested197

buttheyfailedtodiscriminatetherealluminancecontourundermaskingconditions198

and were therefore excluded from further analysis (see more details below). All199

participantswereunawareoftheexperimentalhypothesis.200

201

Stimuli202

203

Stimuliwerecreatedbyplacingfourdiscs(inducerelements,diameter=2.2°) inthe204

configurationofasquare(width=4.3°).Theconfigurationofthesediscswascentered205

onfixation.Asbothstandardedgetypeandanumberofline-endinducingelements206

havecomparableefficacyintheclaritywithwhichillusorycontoursareperceivedby207

participants[41,42],herethediscsweredefinedbypartialconcentriccircles.Eachof208

the four lines forming the circles had a width of approximately 0.07° with a209

luminanceof0.6cd/m2.Thepositioningofthegapwithinthecirclesgaverisetothe210

percept of a Kanizsa triangle (Figure 1A). Thus, a number of line-end inducing211

elementsgaverisetoperceptionofillusorycontours[43].212

213

Wealso includedareal luminancecondition.Here, thestimulididnotcontainany214

discsbutinsteadtherewasatriangledefinedbyarealbutsubtleluminancecontrast215

at theexact locationwhere theKanizsa trianglewouldbeperceived in the illusory216

condition(Figure1B).Wereasonedthatifparticipantswereunabletodiscriminate217

theorientationofthissubtleluminanceedgewhentheinducersweremasked,this218

implied that theywould be unable to detect any illusory contour that could have219

formedwithoutawarenessoftheinducers.Therefore,weremovedtwoparticipants220

whosediscriminationperformanceforthisconditionwasatchancelevelsfromany221

furtheranalyses.222



https://doi.org/10.1101/050526


8

223

Control stimuli did not form any triangle and were created by altering the224

orientations of the inducers by a systematic rotation of 180º (Figure 1C). This225

conditionwassomewhatunnecessarybecausethedependentvariable,accuracyfor226

discriminating theorientationof the triangle,wasorthogonal to theconditionand227

for thesecontrol stimuli therewasnostrictly correctanswer in thisdiscrimination228

task. The “correct” responses in this condition were dummy coded so that the229

stimulus-responsemappingmatchedthatfortheequivalentillusorytrianglestimuli230

beforerotatingtheinducers.Weincludedthisconditionascatchtrials–participants231

should be guessing here because there was no triangle to discriminate. Thus it232

provided information on whether participants could indeed perceive the triangle233

contoursorwhethertheyadoptedastrategyofmatchingtheinducerorientationto234

feedback.235

236

Thebackgroundhadaluminanceof76.6cd/m2andrealtrianglesweredefinedbya237

subtly greater luminance of 86.9 cd/m2. The mask was only used in the invisible238

conditions and was designed as a square configuration of four black discs (0.6239

cd/m2), each containing four white concentric circles (230 cd/m2). It was wide240

enoughtocoverall fourPacmen in the illusoryandcontrolconditions (Figure1D).241

Themaskingtechniqueconsistedofasequenceofthreeframeswhichwasrepeated242

threeconsecutivetimes:Atfirstthemaskappearedonthescreen,followedbythe243

appearanceofthestimulusandafinalblankscreen.Inthevisibleconditions,ablank244

greyscreenreplacedthemaskframe.245

246



https://doi.org/10.1101/050526


9

247Figure1. Illustrationofthestimuli.A)(i)Kanizsatrianglepointingright(ii)Kanizsatriangle248

pointingleft.B)(i)Realtrianglepointingright(ii)Realtrianglepointingleft.C)Controlstimuli249

werecreatedbyasystematic180°rotationof the individualPacmen.D) Illustrationof the250

maskingprocedureintheinvisibleconditions:Amask-stimulus-blankscreenframesequence251

wasrepeatedthreetimes.Inthevisibleconditionsablankscreenreplacedthemask.252253

254

255



https://doi.org/10.1101/050526


10

Procedure256

257

Kanizsa task:At first, participants were instructed that they would see a triangle258

appearonthescreen.Theywereaskedtojudgewhetheritshypotenusewastilted259

clockwiseorcounter-clockwisefromverticalbypressingthecorrespondingresponse260

key.We explained this task to them as a decisionwhether the right angle of the261

trianglewaspointingtotheleftortotherightbuttheyweretoldexplicitlytojudge262

thecontoursofthetriangle,inparticularthelonghypotenuseextendingthroughthe263

centerofthestimulusdisplay.Becausewewishedtokeepthecontrastbetweenreal264

triangleandthebackgroundaslowaspossible,wetrainedparticipantsonthevisible265

andinvisiblerealtriangleconditionuntiltheywereabletodetectthemcorrectly.266

267

Participantswereinstructedtofixateasmallblackdot(0.2°wide)thatwaspresent268

inthecenterofthescreenthroughouttheexperiment.Oneachtrial,thefixationdot269

was displayed alone for 500ms. Thiswas followed by a sequence of three frames270

thatdefinedwhethertheconditionwasavisibleoraninvisibleone.Intheinvisible271

condition a 300msmask followed the fixation period. Subsequently, the stimulus272

appeared on the screen for one frame of approx. 16.7 milliseconds (ms),273

immediatelyfollowedbyablankscreenthatwasshownforoneframeaswell.This274

mask-stimulus-blanksequencewasrepeatedthreetimesbeforeasecondandfinal275

post-stimulus blank screen was presented until participants gave their response.276

Figure2Ashowsthegeneralparadigmfortheexperimentalprocedure.277

278

Visibility task:We further tested whether participants indeed did not consciously279

perceiveanycontextualinformationofthestimuliintheinvisiblecondition.Forthis280

purpose, the visibility taskassessed theeffectivenessof themasking techniqueby281

measuring whether participants could consciously discriminate the inducer282

elements.Participantswereaskedtojudgewhetherarightanglewaspresentedin283

the left or right bottom inducer. The timing of the trial sequence in this taskwas284

identical to the Kanizsa task, with one exception: only control stimuli were used285

(Figure2B).286

287



https://doi.org/10.1101/050526


11

288Figure2. Illustrationof thetrialsequence inexperiment1.A)Kanizsatask:Eachtrialwas289composedoffiveframes:fixationdot,mask,stimulus,blank,andpost-stimulusblank.After290

theparticipant’sresponse, thefixationpointprovidedfeedbackfor100ms(green:correct;291

red:incorrect).Thedurationofeachframeisshownonthetime-line.Notethatinthevisible292

conditions,ablankscreenreplacedthemask.B)Visibilitytask:Trialsproceededinthesame293

way as in the Kanizsa task except that only control stimuli were being shown and294

participants judged whether the right-angle gap was in the bottom left or right inducer295

(indicatedbythegreencircle,whichwasnotpresentintheactualstimuli).296

297

298

299



https://doi.org/10.1101/050526


12

Dataanalysis300

301

Performance in both tasks was defined in terms of mean proportion of accurate302

responsesineachcondition.Initially,weconductedbinomialtestsattheindividual303

level to quantify how many participants performed significantly above chance. A304

condition for any participant’s data to be considered in the group analysis of this305

experimentwasthattheirperformancetothe invisiblerealstimulusconditionwas306

significantlyabovechance(0.5).Thisisbecausethefeaturesnecessarytodistinguish307

the contours of the real triangle were not masked and if a participant could not308

perform the task for this particular stimulus, any test of theperceptionof illusory309

contourswouldberedundant.310

311

One-samplet-testswerecarriedoutatthegrouplevelforeachconditionindividually312

to assess whether the participants’ level of performance was significantly above313

chance (>0.5). In addition to traditional frequentist statistical tests, we also314

quantified the evidence for or against the hypothesis that participants could315

discriminate the stimuli by calculating a Bayes Factor using a default Cauchy prior316

with scaling factor 0.707 for the alternative hypothesis [44]. For conditions with317

performancenear chance levels, thisenabledus toalsoquantifyhowstrongly the318

evidencesupportedthenullhypothesisthatparticipantswereactuallyguessing.319

320

321

Experiment2322

323

Thisexperimentonlycomprisedonetaskthattestedtheperceptofillusorycontours324

by measuring participants’ threshold on a spatial localization task. Such a325

manipulationhasbeenusedsuccessfullyinpreviousexperiments[23,39].Whileitis326

an indirect test, thecontouraidsperformanceonthespatial localization task.This327

provides independent evidence about whether participants perceived any illusory328

contours and therefore helps to address confounds with measuring the percept329

directly as we did in experiment 1. Pilot experiments using a task in which we330

directlymeasuredtheillusorypercept(thesametaskasexperiment1butwiththe331



https://doi.org/10.1101/050526


13

long-lasting masking technique employed in experiment 2) could encourage332

participants to pay close attention to the masked context and thus reduce the333

effectiveness of the masking procedure. This would mean that participants use334

residualawarenessoftheinducerstoperformthetaskinsteadofactuallymakinga335

perceptualjudgmentoftheillusorycontour.336

337

Moreover, because the masking procedure in experiment 1 used very brief338

presentationsoftheinvisibleKanizsastimuli(16.7ms)whiletheseweremuchlonger339

(300ms)forthevisibleones,inexperiment2weusedadifferentmaskingprocedure:340

stimuli were defined by sinusoidal gratings which reversed contrast polarity at341

120Hz.Thismethodcaneffectivelyrenderstimuliinvisibleforprolongedperiodsso342

thatparticipantsonlyperceiveagreyscreen.Previousresearchsuggeststhatstimuli343

masked in thiswayareprocessed inearlyvisual cortex [3].Moreover,weshowed344

thatstimulirenderedinvisiblebythismethodcouldinducelocalprimingeffectsona345

shapediscriminationtask[21].346

347

Participants348

349

Five normal, healthy participants (3 female; age range: 24-37; mean age: 30.2)350

completedthisexperiment,includingoneoftheauthors(DSS).Allparticipantswere351

experienced with psychophysical experiments. While the results from a larger352

subject basemight generalizemore to abroadpopulation,we reasoned thatonly353

precise measurements of participants’ visual performance would afford354

interpretabledata.Unlikeinthefirstexperiment,inexperiment2wemeasuredhow355

positiondiscriminationthresholdschangedbetweenexperimentalconditions.Naïve,356

untrained participants from the general population would be more likely to357

contributenoisydataandthismightobscurepotentialsubtleperceptionof illusory358

contoursinthemaskedcondition.359

360

Stimuli361

362



https://doi.org/10.1101/050526


14

WegeneratedaKanizsashapebypresentingfourGaborpatches(sinusoidalgratings363

with wavelength 0.33° visual angle convolved with a Gaussian with standard364

deviation0.8°at30%contrast)inthelocationsofthefourcornersofasquarewitha365

side-lengthof8.2°.Weturnedthesepatches intoPacmenbysettingaright-angled366

regionofeachpatchtozerocontrast(uniformbackgroundgrey).Thisstimulusthus367

describedanillusorysquare(Figure3A).Inthecontrolconditions,thePacmenwere368

rotated by 180° so that the corners faced outward breaking the illusory percept369

(Figure3B).Intherealluminancecontrol,wedidnotpresentanyGaborpatchesbut370

insteadasquareregionwithasubtleluminancecontrast(54cd/m2)relativetothe371

background (Figure 3C). The orientation and phase of each Gabor patch was372

randomizedineachtrial.Finally,wecreatedamaskstimulusbyoverlaying24Gabor373

patches in the four locations. These patches were presented at 5% contrast and374

covered the full range of orientations in equal steps of 15° but their phaseswere375

randomized.This resulted inapatchypatternwithoutanyobviousorientationcue376

(Figure3D).Themaskpatternwasalsogeneratedanewineverytrial.377

378

Thesestimuliwerepresentedinthecenterofthescreen.Thetargetstimuluswasa379

tiny dark grey dot (diameter: ~0.1°) which could appear somewhere along the380

horizontalmeridian,eitherneartherightortheleftverticalboundaryofthesquare.381

382

The square stimuli were either invisible or visible. In the invisible condition, the383

gratings reversed contrast polarity at 120Hz. In the visible conditions and the real384

luminance control, every even-numbered framewas uniform grey (except for the385

fixationdot).386

387



https://doi.org/10.1101/050526


15

388Figure3.Stimuli(A-D)andtrialsequence(E)inexperiment2.Duringthestimulusinterval389

eitheran illusoryKanizsasquare(A),acontrolstimuluswithrotated inducers(B),orareal390luminance square without any inducers (C) was presented. In visible trials every odd-391

numberedvideoframeat120Hzcontainedthestimuluswhileeveryeven-numberedframe392

contained only a blank screen with the blue fixation dot. In invisible trials, both frames393

contained the Gabor patches for A and B but their contrast polarity alternated between394

frames. For real luminance stimuli the even-numbered frames always contained a blank395

screen. D) A mask stimulus preceded and followed each stimulus interval. E) Each trial396

started with 500ms of fixation, followed by a 100ms mask, the 500ms stimulus interval,397

another100msmask,and finallyablankscreenwithouta fixationdot that remaineduntil398

participantsgavetheirresponse.Theirtaskwastolocatethesmalltargetdotinthestimulus399

displayanddecidewhether itappearedto the leftor therightof theverticalboundaryof400

thesquare(here,thecorrectresponseisright).401

402

403

Procedure404

405

Figure3EdepictsthesequenceofaninvisibleKanizsatrialinexperiment2.Eachtrial406

startedwith500msofablankgreyscreen thatonlycontainedablack fixationdot407



https://doi.org/10.1101/050526


16

(diameter0.26°).Thiswasfollowedbya100mspresentationofthemaskandthen408

500ms of the stimulus, after which another 100ms mask interval was presented.409

Then the screen turned grey and the fixation dot was removed, indicating that410

participantscouldgivetheirbehavioralresponse.411

412

Duringthestimulusintervalthefixationdotwasblueinsteadofblacktodenotethat413

thiswas the task-relevant interval. As described above, the stimuli in this interval414

either reversed in contrast polarity at 120Hz (invisible Kanizsa and control415

conditions)or the frameswere interleavedwithblank frames (visibleKanizsa, real416

luminance stimuli and control). During the stimulus presentation the small dark417

targetdotalsoappearedatsomelocationalongthehorizontalmeridian.Itsposition418

was randomized to be either near the left or the right vertical boundary of the419

squareregion.Thedotcouldeitherappearinsideoroutsidethesquareregion.420

421

Thedependentvariableinthisexperimentwasthedistancebetweenthetargetdot422

and the boundary of the square region and it was controlled by a 2-down, 1-up423

staircase procedure that converged on the threshold distance in each of the five424

experimental conditions at which performance was approximately 70.7% correct.425

That is, after every consecutive two correct trials the distancewould decrease by426

onepixel (~0.03°)whileafterevery incorrect trial it increasedbyonepixel.All five427

staircasesstartedatadistanceof15pixels(0.49°).Theminimumandmaximumthat428

theycouldreachwereonepixeland25pixels(0.82°),respectively.429

430

BeforetheactualexperimentweshowedparticipantsstaticexamplesoftheKanizsa431

andthecontrolstimuli.Participantswereinstructedtojudgewhetherthetargetdot432

was left or right of the (imaginary) vertical boundary of the square region.While433

showingthemstillimagesofthestimuli,wespecificallyexplainedtothemthatthis434

boundarywasdefinedby theexactcenterof theGaborpatches (i.e. thecornerof435

thePacman’smouth)andthatthiswasidenticalinboththeKanizsaandthecontrol436

conditions.Wefurtherinformedthemthattherewouldbeathirdconditioninwhich437

they should only see a subtly lighter grey square against the background but no438

Gaborpatches.Inordertobecomeacquaintedwiththetask,theythenperformed1-439



https://doi.org/10.1101/050526


17

2 blocks of the experimentwith only the visible Kanizsa, the control and the real440

luminance condition. Finally, the actual experimentwould commence. Participants441

wereinformedthatthetaskwaslargelyalwaysthesameasthefamiliarizationrun,442

although during this experiment there would be many more trials in which they443

eitheronlyseealightgreysquarebutnoGaborpatchesorthattheymightevenonly444

seeagreyblankscreen.Exceptfortheauthor,allparticipantswereunawareofthe445

experimentalhypothesis.Indebriefingnoneoftheparticipantsreportedseeingany446

Gaborpatchesduringtheinvisiblecondition.447

448

Participantsperformedtwoconsecutiverunscomprising500trialseach.Theseruns449

were further subdivided into 25 blocks. In every block each of the five conditions450

appeared4timesinapseudo-randomly interleavedorder.Blockswereinitiatedby451

button-press.Duringblockbreaksamessageonthescreenremindedtheparticipant452

ofthetaskinstructionsandofthenumberofblockstheyhadalreadycompleted.453

454

Dataanalysis455

456

Wedeterminedthethresholddistanceineachofthefiveexperimentalconditionsby457

calculating the mean distance across the final 15 reversals in the staircase458

procedure. Then we conducted a group level analysis in which we compared the459

average thresholds for conditions across participants. For thiswe used a two-way460

repeated-measures ANOVA with the factors visibility (visible vs invisible) and461

stimulustype(illusoryvscontrolstimulus).Astherewasonlyoneconditionwitha462

real luminance contour, the thresholds for this conditionwerenot included in the463

ANOVA. However, we used paired t-tests to compare results for the illusory and464

controlstimulidirectlytotherealluminancecondition.465

466

467

Results468

469

Experiment1470

471



https://doi.org/10.1101/050526


18

Inthisexperiment,17participantsjudgedtheorientation(leftvsright)ofatriangle472

thatwaseitherdefinedbyanillusorycontourorasubtlerealluminanceedge.Ina473

thirdcontrolconditionthePacmenstimuliinducingillusorycontourswererotatedso474

thatnotriangleshapecouldbeperceived.475

476

KanizsaTask477

478

Accuracyacrossparticipantsfordiscriminatingtheorientationofthetriangleforthe479

six conditions is displayed in Figure 4. We tested whether discrimination480

performanceineachconditionwasabovechanceatthegrouplevel.481

482

Inthevisiblecondition,themeanproportionofcorrectresponsesforthegroupwas483

closetoceilingandclearlybetterthanchancebothforthereal(M=0.97,t(16)=83.5,484

p<0.001,BF10>9.8*1018)andtheillusorycondition(M=0.99,t(16)=120.8,p<0.0001,485

BF10>2.4*1021).Inthecontrolcondition,however,thegroupperformedsignificantly486

below chance level (M=0.30, t(16)=-3.87, p=0.001, BF10=29.3). There was487

considerablevariabilityinperformanceforthisconditionrangingfrom0to0.74.488

489

In the invisible condition, only performance on the real triangle condition was490

significantly above chance level (M=0.71, t(16)=6.4, p<0.001, BF10=2608.7).491

Performance on both the illusory triangle condition (M=0.51, t(16)=1, p=0.332,492

BF10=0.385)andthecontrolcondition(M=0.50,t(16)=0.2,p=0.859,BF10=0.253)were493

at chance level. Importantly, performance on the real triangle condition was also494

significantly greater than for either the illusory contour (t(16)=6.1, p<0.0001,495

BF10=1475.4) or the control condition (t(16)=5.6,p<0.0001,BF10=688). In contrast,496

performance for the illusory contour did not differ from the control condition497

(t(16)=0.7,p=0.525,BF10=0.3).498

499

VisibilityTask500

501

In a supplementary control taskwe askedparticipants tomake a decisiondirectly502

basedontheinducersbyreportingwhichoneofthetwobottominducerscontained503



https://doi.org/10.1101/050526


19

the rightangle.Thisassessedany residualperceptionof the inducer shapesunder504

masking conditions. In the visible condition, the mean proportion of correct505

responsesacrossparticipantswasagainclosetoceilingandfarabovechance level506

(M=0.96, t(16)=22.2, p<0.001, BF10>2.9*1010). In contrast, participants’ mean507

performance in the invisibleconditiondidnotsignificantlydiffer fromchance level508

(M=0.48, t(16)=-1.3, p=0.227, BF10=0.489), suggesting that participants could509

generallynotperceivethe inducersundermaskingconditions.Note,however,that510

therewasonlyanecdotalsupportforthenullhypothesis(BF10>1/3).511

512

The first experiment thus suggested that if participants were unaware of the513

inducersbecausetheyhadbeenmasked,theywereunabletojudgetheorientation514

oftheKanizsatriangle.Thissupportstheinterpretationthatillusorycontoursarenot515

formedundertheseconditions.516

517

518Figure4.Resultsofexperiment1.Accuracyfordiscriminatingtheorientationofatriangle519

stimulus in visible or invisible (masked) trials. Each dot represents the performance of an520

individualparticipantineachoftheconditions.Thelargesymbolsanderrorbarsdenotethe521

mean ±1 standard error for each condition. Black: real luminance contour. Red: illusory522

(Kanizsa)contour.Blue:controlstimuli.523

524

Experiment2525

526



https://doi.org/10.1101/050526


20

Inexperiment2wechangedtheapproachinanumberofways.First,insteadofthe527

brief stimulus presentations used in the first experiment we used counter-phase528

flicker to render stimuli invisible for prolonged periods. Moreover, we used an529

indirect measure of illusory contour processing: we took advantage of previous530

reports that the presence of an illusory contour stimulus enhanced participants’531

thresholds at discriminating the position of a small dot [23,39]. Here we tested532

whetherthisalsooccurredwhentheinducersgeneratingtheillusorycontourwere533

invisible. Because this task was more challenging than those in previous534

experiments, and to rule out that our previous results might have been due to535

insufficient practice or familiarity with psychophysical experiments, in this536

experimentweonlytestedasmallgroupofwell-trainedpsychophysicsparticipants.537

538

Figure 5 plots position discrimination thresholds in the different conditions for all539

individualparticipantsandthegroupaverages.Overall,thresholdsmeasuredwhilea540

Kanizsa stimulus was presentedwere significantly lower than thosemeasured for541

control stimuli (F(1,4)=34.6, p=0.004). There was also a non-significant trend of542

lowerthresholdsduringvisiblethaninvisibletrials(F(1,4)=7.5,p=0.052).Importantly,543

therewasasignificantinteractionbetweenvisibilityandstimulustype(F(1,4)=15.9,544

p=0.016). In the visible condition thresholds measured while an illusory Kanizsa545

contour was present were significantly lower (M=0.08°, t(4)=-5.27, p=0.006,546

BF10=9.95) compared to the control condition without a contour (M=0.21°). In547

contrast, intheinvisibleconditionthedifferenceinthresholdsfor illusorycontours548

(M=0.20°)wasnotsignificantlydifferentfromthatforthecontrolstimulus(M=0.21°,549

t(4)=-0.16,p=0.882,BF10=0.4).550

551

Thresholdsmeasuredfortherealluminancecontourwereofasimilarmagnitudeas552

those measured for the visible illusory contour (M=0.10°, t(4)=-0.92, p=0.409,553

BF10=0.55).Incontrast,thresholdsfortherealluminancecontourweresignificantly554

lower than for invisible Kanizsa stimuli (t(4)=3.84,p=0.019,BF10=4.53). Thresholds555

forthereal luminancecontourwerealsosignificantly lowerthanforeithercontrol556

stimulus(visible:t(4)=4.4,p=0.012,BF10=6.19;invisible:t(4)=5.3,p=0.006,BF10=9.9).557

558



https://doi.org/10.1101/050526


21

559Figure 5. Results of experiment 2. Spatial discrimination thresholds per condition (real560

luminancevsillusoryvscontrol)onadotlocalizationtaskwhileparticipantswerepresented561

with visible (blue squares), invisible (orange diamonds), or real luminance stimuli (black562

circles). A-E) Plots for the five individual participants. F) Thresholds averaged across563

participants±1standarderrorofthemean.564

565

566

Discussion567

568

In two psychophysical experiments we tested whether the visual system forms569

illusory (Kanizsa) contours in the absence of awareness of the inducing context.570

Takentogether,withpreviousexperimentsusingdichopticstimulation[22,23]allthe571

findings suggest that illusory contours are not perceived when the inducers are572

masked.573

574

Inourfirstexperiment,wedirectlymeasuredillusorycontourperceptionbyasking575

participantstodiscriminatetheorientationofaKanizsatriangle.Thisprocedurewas576

akin to our earlier experiments [22] that showed no evidence either that illusory577

contours are formed when inducers are suppressed from awareness. In those578



https://doi.org/10.1101/050526


22

experimentsweemployedCFSinwhichadynamic,high-contrastmaskpresentedto579

one eye suppresses awareness of a stimulus viewed by the fellow eye. Similarly,580

previous studies showed that illusory contours are not formedwhen the inducers581

are suppressed from awareness during binocular rivalry [23]. However, because582

previousreports indicatethat illusorycontoursareprocessedbybinocularneurons583

[30–33]theuseofdichopticstimulimaysimplydisrupttheirprocessing.584

585

Thereforehereweusednormalbinocularviewingconditionsandadifferentmethod586

to render the inducers invisible. Inaddition,we includeda real luminancecontour587

conditionasabaselinecheck.Thishelpedtoruleoutanothertrivialexplanationfor588

ourearlierfindings:Thepresenceofthebrightmaskscouldpossiblyhaveobscured589

thedetectionofthesubtleillusoryboundary.Ourrealluminancecontourwasnever590

masked because the masks only overlapped with the corners of the triangle;591

therefore,ifparticipantswereabletodiscriminatethesubtlerealluminancecontour592

theyshouldalsobeabletodosoifanillusorycontourwereindeedformed.Weonly593

included participants for whom discrimination of this real luminance contour was594

significantly above chance levels. Nonetheless, participants were unable to595

discriminatetheorientationoftheillusorycontourwheninducersweremasked.The596

phenomenologicalexperienceof thereal luminanceandthe illusorycontour isnot597

perfectly identical (and itobviously isnot for thevisible conditions).However, the598

real luminance conditions were very faint while illusory contours tend to be599

subjectivelyquitesalient.Thusitseemsunlikelythatparticipantsshouldbeableto600

detectonlytherealluminancecontourbutnottheillusorycontour.601

602

Experiment1alsocontainedcontrolconditionsinwhichtheindividualinducershad603

been rotated by 180° and thus no illusory triangle should be formed. Thesewere604

essentiallycatchtrialsbecauseparticipantsshouldhavebeenguessing,astherewas605

noactualtriangletodiscriminate.Thiswasclearlythecaseforcontrolstimuliwhen606

inducers were invisible. However, for visible control stimuli discrimination607

performancevariedwidelyfrom0-0.74proportioncorrectresponsesandwaseven608

significantly below chance level. The “correct” responses in this condition were609

dummy coded so that the stimulus-response mapping matched that for the610



https://doi.org/10.1101/050526


23

equivalent illusory triangle stimuli before rotating the inducers. Therefore one611

inducerinthetoprowhadaright-angledcutoutthatpointedinwards(Figure1)and612

this would be consistentwith the presence of a triangle oriented in the opposite613

direction (even though no actual triangle should be perceived). Some participants614

may have adopted this stimulus responsemappingwhile others did not. Itwould615

thereforehavebeenbetter tohaveall inducerspointoutwards in thiscontroland616

randomize their locations.However, our actual results serendipitously support the617

fact that masking was effective in this experiment: The fact that participants618

performedatchancelevelwhenthecontrolstimuliweremaskedrulesoutthatthey619

hadanyresidualawarenessoftheinducers.TheresultsofourexplicitVisibilitytask620

afterthemainexperimentalsocorroboratedthisconclusion.621

622

Theresultsofexperiment1contradictpreviousclaimsthata“ghostlytriangle”could623

beperceivedwhen the inducersofaKanizsa triangleweremasked [24].However,624

follow-upexperiments failed to replicate theseexperimentsbut instead suggested625

that conscious processing of the inducers precedes awareness of the illusory626

contoursandthattheseearlierresultswereinfactduetoresidualawarenessofthe627

inducers[45].Anothercriticalissuewiththeseexperimentsisthattheperceptionof628

illusorycontoursisbasedonaYesorNojudgmentofwhetheratriangleshapewas629

present. Such a task could theoretically be performed based on any residual630

awarenessof thecorners in the inducers.Our task requiredanexplicitorientation631

discriminationofthe illusorycontour. Ifparticipantshadbeenabletoperformthis632

task in spiteofbeingunable toperceive the inducers, thiswouldhavebeenmore633

conclusive evidence that illusory contours are indeed formed when inducers are634

masked.635

636

Naturally,ourdesignthatsplitthemaintaskfromatestofawarenessdidnotallow637

ustomeasuretheawarenessoneachactual trialof themainexperiment–onlya638

dual-taskdesign inwhichvisibility isprobeddirectly inthemainexperimentwould639

permitthis.However,adual-taskdesigncannotclassifytheawarenessofindividual640

trials perfectly as the participant’s judgment of their own awareness is subject to641

variability. Dual-tasks also entail a division of attentional resources across the642



https://doi.org/10.1101/050526


24

different task components. It could be argued that the Visibility task was more643

difficult than theKanizsa taskbecause the former required the judgmentofonea644

small,peripheralfeatureofthestimulus.Thismaycomplicatetheinterpretationof645

chanceperformanceonthistask.646

647

Experiment1alsousedverybriefstimuluspresentationsandpowerful,high-contrast648

maskstorenderstimuliinvisible.Ratherthantheabsenceofawareness,thereason649

why invisible contours were not formed could be the brevity of the stimuli or650

because themask fundamentally disrupted stimulus processing. This is certainly a651

possibility because previous research indicated that the processing of illusory652

contoursoccursrelativelyslowly[36–38].Therefore, inoursecondexperiment,we653

used counter-phase flicker tomask the inducers instead of themaskingmethods654

employed in the earlier experiments. This allowed us to present stimuli for655

prolonged periods.While this is also a temporalmasking procedure, the stimulus656

energyisconstantduringtheentirepresentationbecauseonlythepolaritychanges657

betweenframes.Furthermore,weexploitedthefactthatthepresenceofanillusory658

contourenhancesperformanceonadotlocalizationtaskbecauseitprovidesavisual659

aid fordetermining itsspatial location [23,39].Weconfirmedthisadvantagewhen660

inducerswerenotmaskedandperformancewascomparabletowhenwepresented661

realluminancecontoursonly.However,wheninducerswererenderedinvisiblethis662

advantageforillusorycontoursdisappeared.Thistaskmayactuallybeanevenmore663

appropriatetestofthe inductionof illusorycontoursthantestingdiscriminationof664

the contour itself as in experiment 1. For that experiment, the interpretation is665

unproblematicbecausewefoundnoevidenceofdiscriminationwheninducerswere666

masked.However,ifwehadfoundabovechanceperformanceformaskedstimuli,it667

wouldhavebeenimpossibletoconcludethatthiswasnotduetodiscriminationof668

theinducers.Pilotexperimentsforexperiment2suggestedthatwiththelong-lasting669

counter-phaseflickermasking,participantsmightinoccasionaltrialshavehadsome670

residual awareness of the corners in the inducers. Thus they might still have671

performed above chance at a shape/orientation discrimination on the purported672

illusorycontoureventhoughtheydidnotinfactperceiveanyillusorycontour.This673

problemalsoplaguesmanypreviousexperimentsthatdirectlytestedthepresence674



https://doi.org/10.1101/050526


25

ofanillusoryshape[36,37].Therefore,onlyataskthatexploitsthepresenceofthe675

illusory contour to modulate performance on an orthogonal task, like the dot676

localization taskweused, canprovideconclusiveevidence thatan illusorycontour677

wasinfactperceived.Analternativepossibilitycouldbeataskthatreliesonafine678

discriminationofafeatureoftheillusorycontoursuchasitscurvature[38]buteven679

such discriminations may be confounded (with?) discrimination of the inducers680

[46,47].681

682

InalltestsofperformanceagainstchancelevelsweusedBayesianhypothesistests683

that can quantify the strength of evidence for the null hypothesis indicating that684

participantswereguessing[44].Noneofthesetestsrevealedstrongevidenceforthe685

null hypothesis as typical Bayes Factors fell between 0.3-0.5. To establish more686

compellingsupportforthenullhypothesisinthosecasesmuchlargersampleswould687

be required. Crucially, the Bayes Factor indicates by how much the observed688

evidenceshouldupdateone’spriorbeliefinthenulloralternativehypothesis.Even689

if theevidence is relativelyweak,anyBayesFactorbelow1 isevidence in favorof690

thenullandnotforthealternativehypothesis.Unlessonestartswithapriorbelief691

thatpeopleareclearlyabletodiscriminatemaskedstimuli,eventhesemodestBayes692

Factorssuggestthatparticipantswereprobablyguessing.693

694

Several previous studies used stimulusmanipulations that seek to disentangle the695

factors associated with Kanizsa-type illusory contours [39,48,49]. Rounding the696

cornersofthePacmaninducersresultsinanotablereductionintheillusorycontour697

percept and abolishes the concordant improvement on a spatial localization task.698

However, stimuli like this nonetheless activated higher extrastriate cortex to a699

similar degree as Kanizsa stimuli [39]. This is consistentwith the theory that later700

stages of visual processing, presumablymediated by higher visual areas, segment701

surfacesandassignboundariestoobjects.Thesesegmentationsarethenfedbackto702

earlyvisualcortextogeneratesignalsthatareinterpretedasillusorycontours[48].703

Practical support for this idea comes from transcranial magnetic stimulation704

experiments that disrupted neural processing either in object-sensitive lateral705

occipital(LO)cortexorinearlyvisualareasV1andV2[50].Critically,thedisruption706



https://doi.org/10.1101/050526


26

of LOcortexonlyabolished the illusory contourperceptearlyafter stimulusonset707

while disruption of early visual areas only did so at a later stage – presumably708

affectingfeed-backsignalsratherthantheearlyfeed-forwardresponse.709

710

Additional stimulusprocessing that is unrelated to the actual formationof illusory711

contourscouldalsoexplainpreviousreportsthatKanizsastimuliarefastertobreak712

through CFS masking than control stimuli [35]. The collinearity of edges and the713

thereby inferred surfacemay be processed evenwhile the stimulus is suppressed714

from awareness – this may in turn produce an attentional signal that causes the715

stimulus to break suppression. Recent experiments that tested a range of visual716

controlstimuliunderCFSsuggestthatlow-levelpropertiesofthestimulusdetermine717

thetimeittakestobreaksuppression[51].Suchstimulus-dependenteffectsarealso718

plausible because attentional processing can occur without awareness of the719

stimulus [52–54]. However, this does not prove that any percept of an illusory720

contourwasactuallyformed.Thisprocessmayalsoexplainwhycrowdinginterferes721

with discrimination of the inducer orientation but not with illusory contour722

formation[55].723

724

Whilewe did notmanipulate the presence of illusory contours in thisway in our725

experiments, we nonetheless controlled this factor by including real luminance726

contour conditions. In experiment 2 the presence of a contour should afford an727

improvement on an orthogonal spatial localization task. Such an improvement in728

localizationthresholdsonlyoccurredfortherealluminancecontourorwhenKanizsa729

inducerswerevisible.Improvementslikethisarenotobservedforstimulithatmatch730

theglobalcharacteristicsbutwhichdonotproduceillusorycontours[39].Therefore731

our results from this experiment strongly support the conclusion that illusory732

contoursweresimplynotformedwheninducersweremasked.733

734

Ourresultsalsoagreewithpreviousfindingsthatonlylocalprocessingoccursinthe735

absenceof awarenessbut thatmore complexanalysisof scenegeometry requires736

conscious processing [21]. This would also explain why the nature of stimulus737

representationsinhighervisualareasdiffersdependingonawareness.738



https://doi.org/10.1101/050526


27

739

However, our findings do not accordwith a number of other studies that suggest740

that illusory contours are processed unconsciously. Experiments on a patientwith741

extinctionduetoaparietallesionsuggestthatperceptionofaKanizsashapeoccurs742

evenwhensomeinducersareplacedintopartsofthevisualfieldwherethepatient’s743

conscious perception is impaired [26–29]. These findings indicate that visual744

processingoperatesatthesurface-orobject-basedlevel.Segmentingandgrouping745

the local features of an image into a coherent, global shape may only require746

awareness of some component features but it then spreads to the whole object.747

However,thisisnotsufficientevidencetodemonstratethattheseprocesseswould748

occurwhentheparticipantisunawareofallcomponents.Moreimportantly,italso749

does not demonstrate that the percept of illusory contours was actually formed750

under these conditions but only that some processing of the features producing751

illusorycontoursundernormalviewingconditionsstilloccurred.752

753

Another study used functional magnetic resonance imaging (fMRI) and an754

inattentional blindness paradigm to study illusory contours processing [49].755

Participantswerepresentedwith a sequenceof images, someofwhich contained756

Kanizsashapeswhilstotherswerevarious typesofcontrol stimuli.Simultaneously,757

participantswereengagedinademandingattentionaltaskatfixation.Asub-group758

of participants subsequently reportednot to have seen any Kanizsa stimuli. These759

participants nonetheless showed stronger fMRI responses to Kanizsa than control760

stimuli.Multivariateclassificationmethodsfurtherdemonstratedthattheactivation761

patterns produced by unseen Kanizsa stimuli were more reliable than those762

produced by control stimuli. The authors suggested that the neural signature of763

illusory contours differs from that of other, carefully matched stimuli and thus764

arguedthatillusorycontoursareprocessedevenwithoutawarenessofthestimuli.765

766

This finding is interestingbecause it tests the consequencesof awarenesswithout767

any experimentalmanipulation of the stimuli. The distinction of what is or is not768

processed without awareness only depends on the contents of the participant’s769

consciousness.However,thisalsomakesitdifficulttointerprettheseresults.First,it770



https://doi.org/10.1101/050526


28

isunclearwhetherparticipantsapparentlyoblivious to thepresenceof the stimuli771

really did not perceive any illusory shapes. Due to the design of the experiment,772

awarenesscouldonlybeassessedafterthemainfMRIexperimentratherthanona773

trial-by-trialbasis(butseeaboveourdiscussionwhysuchtrial-by-trialjudgmentsof774

awarenessarecomplicated).Onlyparticipantswhoreportedhavingseentheactual775

Kanizsastimulusduringtheexperimentwereclassifiedashavinghadawarenessof776

the stimuli. However, many of the candidate stimuli were similar Kanizsa shapes.777

Thereforeitispossiblethatparticipantshadsomeawarenessofthestimuli,evenif778

animpreciseone.779

780

Second,while thecontrol stimuli in themain fMRIexperimentwerevery carefully781

matchedtoruleouttheinfluenceofglobalcharacteristicsthiscanbydefinitiononly782

beanapproximation: if conditionswereperfectlymatched, the stimuluswouldbe783

identical and thus an illusory contour would be perceived. It is possible that the784

conditions resulting in an illusory contourpercept are alsoparticularly effective in785

producing discriminable activation patterns in visual cortex. For instance, the786

contrast energy along the mouths of the Pacmen inducers differed considerably787

between theKanizsaand theglobal control stimuli. Thus, surface segmentationor788

collinearinteractionsmayhavealsodifferedbetweentheseconditions.789

790

Alltheseissuesagainhighlightanimportantpoint:Theonlywayonecantrulyinfer791

that illusory contours are formed is by using a measure that is specific to the792

presence of an illusory contour. This is indeed what we did in our previous CFS793

experiment [22], our present experiment 1, or what was done in several other794

previous experiments [23,34,50,55], by asking participants to directly report a795

featureof the contour.Anotherusefulmanipulation is the spatial localization task796

used by other studies [23,39] as well as our experiment 2. This is because the797

performanceenhancementonlyoccursinthepresenceofanactualcontourhelping798

theparticipanttolocalizethetarget.Manypreviousstudiessuggestingthatillusory799

contoursareprocessedwithoutawarenessoftheinducersareconfrontedwiththis800

problem. In the appendix we included another experiment in which we tested801

whether Kanizsa shapes masked from awareness could nevertheless provide an802



https://doi.org/10.1101/050526


29

attentionalcue forasubsequentvisualsearchtask.Even ifa robustprimingeffect803

were found in such an experiment, this design simply cannot rule out alternative804

explanations. Itonly tests theconsequencesof the stimuli thatmay in factnotbe805

specifictoillusorycontours.806

807

In the same vein, neurophysiological and neuroimaging experiments have shown808

that illusory contours are encoded by neurons even in the early visual cortex809

[25,33,56–67].Theoverwhelmingmajorityoftheseexperimentswereconductedon810

awake participants. However, some neurophysiological studies reported neuronal811

tuning to illusory contours even in anesthetized animals [25,58,67]. Such findings812

seem to superficially contradict our conclusion that awareness of the inducing813

stimuli is necessary for the formation of illusory contours. However, these814

experiments are in fact a perfect illustration of the importance to distinguish815

between perceptual experience and correlated processing. Because animals are816

anesthetizedintheseexperimentsandthusbydefinitionunawareofthestimuli,itis817

impossible to determine whether illusory contours were formed. The neural818

correlates of these stimuli could be related to the contextual processing of the819

inducingstimuli,suchasthediscontinuitiesdetectedby“end-stoppedcells”orthe820

detection of collinearity in the image. Such processing may indeed occur in the821

absenceofawarenessasissupportedbyourfindingofcollinearpriming[21]andthe822

inductionofcontextualvisualillusionsundermaskingconditions[2,3]orinattention823

[6]. Furthermore, it is quite likely that such stimulus processing is a necessary824

prerequisite for the formation of illusory contours. However, they do not825

conclusivelyprovethatillusorycontoursareformedunderanesthesia.826

827

Arelatedissueisthatallofthesestudiesuseillusorycontoursinducedbyabutting828

lines (offset gratings) rather than Kanizsa shapes. While the two share a similar829

phenomenology, contours inducedbyabutting linesarearguably simpler andmay830

bebasedmostlyonlocalprocessingwhileKanizsashapesarelikelytoinvolvemore831

complexinferencesofsurfacedepthandboundaryownership.Ourexperimentsdid832

notexplicitly test illusory contoursgeneratedwithabutting linesand thereforedo833



https://doi.org/10.1101/050526


30

not speak to the question whether such simpler illusory contours are in fact834

perceivedwhentheinducersaremasked.835

836

We conclude that there is little evidence that Kanizsa-type illusory contours are837

processedwhenparticipantsarenotawareoftheinducingcontext.Thisappearsto838

be the case for a range of different methods to render the inducers invisible.839

However,alloftheseexperimentsemployedphysicalstimulusmanipulations,while840

it is likelythatanexperimentaldesignthatallowstheuseoftheparticipant’sown841

report to determine awareness on a trial-by-trial basis whilst also testing directly842

whether an illusory contour was in fact formed can answer this question843

conclusively.844

845

846

Acknowledgements847

848

This researchwas supportedby an ERC StartingGrant (310829) toDSS.We thank849

PieterMoors,JacobJolij,andDavidShanksaswellastwoanonymousreviewersfor850

commentsonapreviousmanuscriptandJonasLarssonforinspiringthisstudyinthe851

firstplace.852

853

854

855

DataandMaterials856

857

Allstimuluscodeanddataforthisstudyareavailablefordownloadat:858

https://dx.doi.org/10.6084/m9.figshare.2009952.859

860

861

References862

863

1. NewellBR,ShanksDR.Unconsciousinfluencesondecisionmaking:acritical864review.BehavBrainSci.2014;37:1–19.doi:10.1017/S0140525X12003214865



https://doi.org/10.1101/050526


31

2. CliffordCWG,HarrisJA.Contextualmodulationoutsideofawareness.Curr866Biol.2005;15:574–8.doi:10.1016/j.cub.2005.01.055867

3. FalconbridgeM,WareA,MacLeodDIA.Imperceptiblyrapidcontrast868modulationsprocessedincortex:Evidencefrompsychophysics.JVis.8692010;10.doi:10.1167/10.8.21870

4. PearsonJ,CliffordCWG.Suppressedpatternsaltervisionduringbinocular871rivalry.CurrBiol.2005;15:2142–8.doi:10.1016/j.cub.2005.10.066872

5. VulE,MacLeodDIA.Contingentaftereffectsdistinguishconsciousand873preconsciouscolorprocessing.NatNeurosci.2006;9:873–874.874doi:10.1038/nn1723875

6. MooreCM,EgethH.Perceptionwithoutattention:evidenceofgrouping876underconditionsofinattention.JExpPsycholHumPerceptPerform.8771997;23:339–352.878

7. TsuchiyaN,KochC.Continuousflashsuppressionreducesnegative879afterimages.NatNeurosci.2005;8:1096–1101.doi:10.1038/nn1500880

8. StewartLH,AjinaS,GetovS,BahramiB,TodorovA,ReesG.Unconscious881evaluationoffacesonsocialdimensions.JExpPsycholGen.2012;141:715–882727.doi:10.1037/a0027950883

9. MudrikL,BreskaA,LamyD,DeouellLY.Integrationwithoutawareness:884expandingthelimitsofunconsciousprocessing.PsycholSci.2011;22:764–885770.doi:10.1177/0956797611408736886

10. SklarAY,LevyN,GoldsteinA,MandelR,MarilA,HassinRR.Readingand887doingarithmeticnonconsciously.ProcNatlAcadSciUSA.2012;109:88819614–19619.doi:10.1073/pnas.1211645109889

11. MoorsP,BoelensD,OverwalleJvan,WagemansJ.SceneIntegration890WithoutAwarenessNoConclusiveEvidenceforProcessingScene891CongruencyDuringContinuousFlashSuppression.PsycholSci.2016;892956797616642525.doi:10.1177/0956797616642525893

12. AxelrodV,ReesG.Consciousawarenessisrequiredforholisticface894processing.ConsciousCogn.2014;27:233–245.895doi:10.1016/j.concog.2014.05.004896

13. HesselmannG,MoorsP.Definitelymaybe:canunconsciousprocesses897performthesamefunctionsasconsciousprocesses?FrontPsychol.2015;6:898584.doi:10.3389/fpsyg.2015.00584899

14. HaynesJ-D,ReesG.Predictingtheorientationofinvisiblestimulifrom900activityinhumanprimaryvisualcortex.NatNeurosci.2005;8:686–91.901doi:10.1038/nn1445902



https://doi.org/10.1101/050526


32

15. MoutoussisK,ZekiS.Therelationshipbetweencorticalactivationand903perceptioninvestigatedwithinvisiblestimuli.ProcNatlAcadSciUSA.9042002;99:9527–9532.doi:10.1073/pnas.142305699905

16. SterzerP,HaynesJ-D,ReesG.Fine-scaleactivitypatternsinhigh-levelvisual906areasencodethecategoryofinvisibleobjects.JVis.2008;8:10.1-12.907doi:10.1167/8.15.10908

17. HaynesJ-D,DriverJ,ReesG.Visibilityreflectsdynamicchangesofeffective909connectivitybetweenV1andfusiformcortex.Neuron.2005;46:811–821.910doi:10.1016/j.neuron.2005.05.012911

18. SchurgerA,PereiraF,TreismanA,CohenJD.Reproducibilitydistinguishes912consciousfromnonconsciousneuralrepresentations.Science.2010;327:91397–99.doi:10.1126/science.1180029914

19. SchurgerA,SarigiannidisI,NaccacheL,SittJD,DehaeneS.Corticalactivityis915morestablewhensensorystimuliareconsciouslyperceived.ProcNatlAcad916SciUSA.2015;112:E2083-2092.doi:10.1073/pnas.1418730112917

20. LingS,BlakeR.SuppressionDuringBinocularRivalryBroadensOrientation918Tuning.PsycholSciJAmPsycholSocAPS.2009;doi:10.1111/j.1467-9199280.2009.02446.x920

21. SchwarzkopfDS,ReesG.Interpretinglocalvisualfeaturesasaglobalshape921requiresawareness.ProcBiolSci.2011;278:2207–2215.922doi:10.1098/rspb.2010.1909923

22. HarrisJJ,SchwarzkopfDS,SongC,BahramiB,ReesG.Contextualillusions924revealthelimitofunconsciousvisualprocessing.PsycholSciJAmPsychol925SocAPS.2011;22:399–405.doi:10.1177/0956797611399293926

23. SobelKV,BlakeR.Subjectivecontoursandbinocularrivalrysuppression.927VisionRes.2003;43:1533–1540.928

24. GellatlyAR.Perceptionofanillusorytrianglewithmaskedinducingfigure.929Perception.1980;9:599–602.930

25. RamsdenBM,HungCP,RoeAW.Realandillusorycontourprocessingin931areaV1oftheprimate:acorticalbalancingact.CerebCortexNYN1991.9322001;11:648–665.933

26. ConciM,BöbelE,MatthiasE,KellerI,MüllerHJ,FinkeK.Preattentive934surfaceandcontourgroupinginKanizsafigures:evidencefromparietal935extinction.Neuropsychologia.2009;47:726–732.936doi:10.1016/j.neuropsychologia.2008.11.029937

27. MattingleyJB,DavisG,DriverJ.Preattentivefilling-inofvisualsurfacesin938parietalextinction.Science.1997;275:671–674.939



https://doi.org/10.1101/050526


33

28. VuilleumierP,LandisT.Illusorycontoursandspatialneglect.Neuroreport.9401998;9:2481–2484.941

29. VuilleumierP,ValenzaN,LandisT.Explicitandimplicitperceptionof942illusorycontoursinunilateralspatialneglect:behaviouralandanatomical943correlatesofpreattentivegroupingmechanisms.Neuropsychologia.9442001;39:597–610.945

30. GillamB,NakayamaK.Quantitativedepthforaphantomsurfacecanbe946basedoncyclopeanocclusioncuesalone.VisionRes.1999;39:109–112.947

31. HäkkinenJ,NymanG.Phantomsurfacecapturesstereopsis.VisionRes.9482001;41:187–199.949

32. LiuL,StevensonSB,SchorCM.Quantitativestereoscopicdepthwithout950binocularcorrespondence.Nature.1994;367:66–69.951doi:10.1038/367066a0952

33. vonderHeydtR,PeterhansE,BaumgartnerG.Illusorycontoursandcortical953neuronresponses.Science.1984;224:1260–1262.954

34. RubinN.Theroleofjunctionsinsurfacecompletionandcontourmatching.955Perception.2001;30:339–66.956

35. WangL,WengX,HeS.Perceptualgroupingwithoutawareness:superiority957ofKanizsatriangleinbreakinginterocularsuppression.PloSOne.2012;7:958e40106.doi:10.1371/journal.pone.0040106959

36. ParksTE.Onthemicrogenesisofillusoryfigures:afailuretoreplicate.960Perception.1994;23:857–862.961

37. ReynoldsRI.Perceptionofanillusorycontourasafunctionofprocessing962time.Perception.1981;10:107–115.963

38. RingachDL,ShapleyR.Spatialandtemporalpropertiesofillusorycontours964andamodalboundarycompletion.VisionRes.1996;36:3037–3050.965

39. StanleyDA,RubinN.fMRIactivationinresponsetoillusorycontoursand966salientregionsinthehumanlateraloccipitalcomplex.Neuron.2003;37:967323–331.968

40. BrainardDH.ThePsychophysicsToolbox.SpatVis.1997;10:433–6.969

41. FrisbyJP,ClatworthyJL.IllusoryContours:CuriousCasesofSimultaneous970BrightnessContrast?Perception.1975;4:349–357.doi:10.1068/p040349971

42. HalpernDF.Thedeterminantsofillusory-contourperception.Perception.9721981;10:199–213.973

43. VarinD.Fenomenidicontrastoediffusionecromaticanell’organizzazione974spazialedelcampopercettivo.[Contrastandchromaticdiffusion975



https://doi.org/10.1101/050526


34

phenomenaintheorganizationoftheperceptivefield].RivPsicol.1971;65:976101–128.977

44. RouderJN,SpeckmanPL,SunD,MoreyRD,IversonG.Bayesianttestsfor978acceptingandrejectingthenullhypothesis.PsychonBullRev.2009;16:979225–237.doi:10.3758/PBR.16.2.225980

45. MuiseJG,LeBlancRS,BlanchardLC,deWarnaffeA.Discriminationofthe981shapeofthemaskedinducingfigureprecedesperceptionoftheillusory982triangle.Perception.1993;22:623–628.983

46. AndersonBL.Thedemiseoftheidentityhypothesisandtheinsufficiency984andnonnecessityofcontourrelatabilityinpredictingobjectinterpolation:985commentonKellman,Garrigan,andShipley(2005).PsycholRev.2007;114:986470–487.doi:10.1037/0033-295X.114.2.470987

47. YangJ,YueZ,WuX.Independenceofthecompletioneffectfromthe988noncompletioneffectinillusorycontourperception.JVis.2015;15:6.989doi:10.1167/15.14.6990

48. KogoN,StrechaC,VanGoolL,WagemansJ.Surfaceconstructionbya2-D991differentiation-integrationprocess:aneurocomputationalmodelfor992perceivedborderownership,depth,andlightnessinKanizsafigures.993PsycholRev.2010;117:406–439.doi:10.1037/a0019076994

49. VandenbrouckeARE,FahrenfortJJ,SligteIG,LammeVAF.Seeingwithout995knowing:neuralsignaturesofperceptualinferenceintheabsenceofreport.996JCognNeurosci.2014;26:955–969.doi:10.1162/jocn_a_00530997

50. WokkeME,VandenbrouckeARE,ScholteHS,LammeVAF.Confuseyour998illusion:feedbacktoearlyvisualcortexcontributestoperceptual999completion.PsycholSci.2013;24:63–71.doi:10.1177/09567976124491751000

51. MoorsP,WagemansJ,vanEeR,de-WitL.Noevidenceforsurface1001organizationinKanizsaconfigurationsduringcontinuousflashsuppression.1002AttenPerceptPsychophys.2015;doi:10.3758/s13414-015-1043-x1003

52. BahramiB,CarmelD,WalshV,ReesG,LavieN.Spatialattentioncan1004modulateunconsciousorientationprocessing.Perception.2008;37:1520–10058.1006

53. KochC,TsuchiyaN.Attentionandconsciousness:twodistinctbrain1007processes.TrendsCognSci.2007;11:16–22.doi:10.1016/j.tics.2006.10.0121008

54. ZhaopingL.Attentioncapturebyeyeoforiginsingletonsevenwithout1009awareness--ahallmarkofabottom-upsaliencymapintheprimaryvisual1010cortex.JVis.2008;8:1.1-18.doi:10.1167/8.5.11011

55. LauJSF,CheungS-H.Illusorycontourformationsurvivescrowding.JVis.10122012;12:15.doi:10.1167/12.6.151013



https://doi.org/10.1101/050526


35

56. CoxMA,SchmidMC,PetersAJ,SaundersRC,LeopoldDA,MaierA.Receptive1014fieldfocusofvisualareaV4neuronsdeterminesresponsestoillusory1015surfaces.ProcNatlAcadSciUSA.2013;110:17095–17100.1016doi:10.1073/pnas.13108061101017

57. FfytcheDH,ZekiS.Brainactivityrelatedtotheperceptionofillusory1018contours.NeuroImage.1996;3:104–108.doi:10.1006/nimg.1996.00121019

58. GrosofDH,ShapleyRM,HawkenMJ.MacaqueV1neuronscansignal1020“illusory”contours.Nature.1993;365:550–552.doi:10.1038/365550a01021

59. HeiderB,MeskenaiteV,PeterhansE.Anatomyandphysiologyofaneural1022mechanismdefiningdepthorderandcontrastpolarityatillusorycontours.1023EurJNeurosci.2000;12:4117–4130.1024

60. KokP,deLangeFP.Shapeperceptionsimultaneouslyup-and1025downregulatesneuralactivityintheprimaryvisualcortex.CurrBiolCB.10262014;24:1531–1535.doi:10.1016/j.cub.2014.05.0421027

61. LeeTS,NguyenM.Dynamicsofsubjectivecontourformationintheearly1028visualcortex.ProcNatlAcadSciUSA.2001;98:1907–1911.1029doi:10.1073/pnas.0315799981030

62. MendolaJD,DaleAM,FischlB,LiuAK,TootellRB.Therepresentationof1031illusoryandrealcontoursinhumancorticalvisualareasrevealedby1032functionalmagneticresonanceimaging.JNeurosciOffJSocNeurosci.10331999;19:8560–8572.1034

63. Montaser-KouhsariL,LandyMS,HeegerDJ,LarssonJ.Orientation-selective1035adaptationtoillusorycontoursinhumanvisualcortex.JNeurosciOffJSoc1036Neurosci.2007;27:2186–2195.doi:10.1523/JNEUROSCI.4173-06.20071037

64. PanY,ChenM,YinJ,AnX,ZhangX,LuY,etal.Equivalentrepresentationof1038realandillusorycontoursinmacaqueV4.JNeurosciOffJSocNeurosci.10392012;32:6760–6770.doi:10.1523/JNEUROSCI.6140-11.20121040

65. PeterhansE,vonderHeydtR.Subjectivecontours--bridgingthegap1041betweenpsychophysicsandphysiology.TrendsNeurosci.1991;14:112–9.1042

66. vonderHeydtR,PeterhansE.Mechanismsofcontourperceptioninmonkey1043visualcortex.I.Linesofpatterndiscontinuity.JNeurosciOffJSocNeurosci.10441989;9:1731–1748.1045

67. ZhanCA,BakerCL.Criticalspatialfrequenciesforillusorycontour1046processinginearlyvisualcortex.CerebCortexNYN1991.2008;18:1029–10471041.doi:10.1093/cercor/bhm1391048

68. DavisG,DriverJ.ParalleldetectionofKanizsasubjectivefiguresinthe1049humanvisualsystem.Nature.1994;371:791–793.doi:10.1038/371791a01050



https://doi.org/10.1101/050526


36

69. Barlasov-IoffeA,HochsteinS.Illusory-contourfiguresprimematchingof1051realshapes.Perception.2009;38:1118–1131.1052

70. Seydell-GreenwaldA,SchmidtT.Rapidactivationofmotorresponsesby1053illusorycontours.JExpPsycholHumPerceptPerform.2012;38:1168–10541182.doi:10.1037/a00287671055

71. LinZ,MurraySO.Primingofawarenessorhownottomeasurevisual1056awareness.JVis.2014;14.doi:10.1167/14.1.271057

72. PratteMS,RouderJN.Atask-difficultyartifactinsubliminalpriming.Atten1058PerceptPsychophys.2009;71:1276–1283.doi:10.3758/APP.71.6.12761059

73. PoscolieroT,MarziCA,GirelliM.Unconsciousprimingbyillusoryfigures:1060theroleofthesalientregion.JVis.2013;13:27.doi:10.1167/13.5.271061

74. StanleyDA,RubinN.Rapiddetectionofsalientregions:evidencefrom1062apparentmotion.JVis.2005;5:690–701.doi:10.1167/5.9.41063

1064

1065

Appendix1066

1067

Primingexperiment1068

1069

Previousexperimentsused illusorycontourstocaptureattention inavisualsearch1070

task [68]or forpriming [69,70]. Inourprimingexperiment,we therefore tested if1071

Kanizsatrianglesrenderedinvisibleusingasimilarmaskingprocedurecouldbeused1072

as primes in a visual search task on a subsequent, visible stimulus array. Like1073

experiment 1, this experiment comprised two consecutive tasks. The first, again1074

called ‘Kanizsa’ task,wasan indirect test that investigatedwhetherprimingwitha1075

Kanizsa triangle in the location of the target improves detection of the target. A1076

secondtaskconsistedofa‘Visibility’taskthatdirectlytestedtheeffectivenessofthe1077

maskingtechnique.1078

1079

Studydesign1080

1081

Inthecontextofthisexperiment,primingreferstotheattentionalcueprovidedby1082

theprimeshapeaboutthelocationofthesubsequentlypresentedtarget inoneof1083



https://doi.org/10.1101/050526


37

fourpossiblelocationsthatinturnaffordsaboostinbehavioralperformanceonan1084

orthogonaldiscrimination task. In theprimedconditions the locationof the target1085

was always identical with the location of the prime, whereas in the unprimed1086

conditionstheprimeshapewasreplacedbyacontrolstimulus,whichlookedexactly1087

likethedistractors intheotherthree locations.Therefore, theunprimedcondition1088

provided a baseline that allowed the assessment of whether the triangle prime1089

boostedtheparticipants’performancebycapturingtheirattention.1090

1091

We tested priming with a 2x2x2 design with attentional priming (primed vs.1092

unprimed), the type of priming triangle (real vs. illusory), and prime visibility1093

(invisible vs. visible) as within subject factors. There were two different prime1094

shapes:arealtriangledefinedbyastrongluminancecontrastoranillusory(Kanizsa)1095

one.Thevisibilityofthecuewasmanipulatedbyvaryingpresentationtimesofthe1096

priming array that followed the mask. A short presentation time of 16.7ms was1097

selected for the invisible conditions to render the prime invisible. In the visible1098

conditions,however,theprimingarraywaspresentedfor300ms,whichleftenough1099

timefortheprimetoenterawareness.1100

1101

Everyparticipant completed twodifferent tasks. Each task comprised20blocksof1102

trials,whereoneblockconsistedof32trials.Acrossonetask,theunprimedvisible1103

and invisible conditions appeared for a total of 160 times each,whereas the four1104

primed conditions (real vs illusory and visible vs invisible)occurred80 timeseach.1105

Thus,halftheblockconsistedofunprimedtrialsandtheotherhalfofprimedones.1106

1107

Behavioral responseswere given on a forced-choice discrimination task by button1108

press (left or right arrow) on a standard computer keyboard. Each trial required1109

either a left or a right response (see experimental procedure) and each response1110

type appeared twice per block for each condition, except for the unprimed ones1111

whereitappearedfourtimesperblock.Conditionswereselectedpseudo-randomly1112

foreverytrialbutwerecounterbalancedwithineachblock.1113

1114

Participants1115



https://doi.org/10.1101/050526


38

1116

Participants in this experimentwere recruitedamong theUCL studentpopulation.1117

Seventeen (13 female, age range: 21-32 years,mean age 24.7±2.5 years) normal,1118

healthyparticipantstookpartintheexperiment,includingthetwoauthors.Allother1119

participantswereunawareoftheexperimentalhypothesis.1120

1121

Stimuli1122

1123

Inhalfthetrials,participantswerecuedwithoneoffourpossiblecues.Wecreated1124

shapestimulibyplacing fourblackdiscs (inducerelements,diameter=1.96°) in the1125

configurationof a square (width=2.8°). Inorder toproduce the illusory shapeof a1126

rightisoscelestrianglesittingontopofthesquare,awedgewasremovedfromeach1127

inducer element (Pacman).Wedges started at the center of their respective discs1128

andextendedtotheiredges.Onewedgemeasured90°andwasalwayspositioned1129

eitherinthebottomleftorbottomrightinducer.Thetwowedgesthatformedthe1130

apexes of the triangle measured 45°. Finally, the wedge of the fourth inducer1131

element measured 90°. This Pacman did not form part of the triangle and was1132

positionedopposite the hypotenuse of the right isosceles triangle. It always faced1133

the outside of the disc configuration.When the inducers were oriented with the1134

gaps forming the three corners of a triangle, an illusory light greyKanizsa triangle1135

wasvisibleontopoftheblackdiscs(FigureS1A,iandii).1136

1137

Real triangleswere identical to thecorresponding illusoryshapesexcept that their1138

contour was defined by a luminance contrast with a black line connecting the1139

cornersofthetriangle(FigureS1A,iiiandiv).1140

1141

ThecontrolstimuliwerecreatedbyalteringtheorientationsofthethreePacmenby1142

a systematic rotation of 180º. This modification broke the link between these1143

inducers and thus no illusory triangle was perceived (Figure S1B). Control stimuli1144

couldalsobehorizontalmirrorimagesoftheillustratedexamples.1145

1146



https://doi.org/10.1101/050526


39

Stimuli in thesearcharraywereright isosceles triangleswithacontourdefinedby1147

luminance (Figure S1C, i and ii). Target stimuliwere triangleswith the right angle1148

pointingdown,eithertotheleftortotheright,while indistractorstherightangle1149

pointedup(FigureS1C, iiiandiv).Participantswereaskedtodetectthetargetand1150

respondviabuttonpresswhether its rightanglewaspointing to the leftor to the1151

right.1152

1153

Themask consisted of a square of black lines connecting four discs, one at each1154

corner. Itwaswideenoughtocoverall fourPacmen,aswellasthelocationofthe1155

trianglewithinthem(FigureS1D,i).Inthe‘Visibility’task,therewasnosearcharray.1156

Instead,thestimulusarrayfollowingtheprimewasasquarecontainingacrossthat1157

gavenoclueaboutthelocationandorientationofthetargettriangle(FigureS1D,ii).1158

1159

The luminance of the background was 156 cd/m2. The luminance of inducer1160

elements, thecontourof the real triangleand theblackof themaskwas2cd/m2.1161

Finally,thegreyofthemaskinthefirsttaskwas58.2cd/m2.1162

1163

Notethatalthoughthereweredifferencesbetweenstimuli,the individual inducers1164

remained the same and there was no change in the local image properties (i.e.1165

square configuration of the four Pacmen and their position among the other 31166

configurations).Thisenabledanassessmentofthedifferenteffectsofprimingwith1167

theKanizsa shape, as compared to the corresponding real triangleandungrouped1168

Pacmenordistractorsintheunprimedconditions.1169

1170



https://doi.org/10.1101/050526


40

1171Figure S1. Illustration of the stimuli in the priming experiment.A)The fourprimeswere1172

formedoffourinducerelements,ofwhichthreecreatedeitherarealtriangleortheillusory1173impression of a triangle. In half the trials, participants were cued with one of the four1174

possible cues: (i) Kanizsa triangle pointing right, (ii) Kanizsa triangle pointing left, (iii) Real1175

triangle pointing right, (iv)Real triangle pointing left. B) Examples of control stimuli used1176

duringcuepresentation.Controlstimuliwerecreatedbyasystematic180°rotationof the1177

individual Pacmen, so that all gaps within the black discs pointed to the outside of the1178

formation. Control stimuli could also be the horizontal mirror image of the presented1179

examples.C)Thetwopossibletargets inthesearchtaskwereright isoscelestriangleswith1180

theirrightanglepointingdown,eithertotheright(i)ortotheleft(ii).Distractorstimuli in1181

the search array alwayspointedup, but variedbetweenpointing left and right.D) (i) The1182

maskusedtorendertheprimeinvisible.(ii)Theuninformativepost-stimulusmaskreplacing1183

thetargetarrayinthe‘Visibility’task.1184

1185

1186

1187

1188



https://doi.org/10.1101/050526


41

Procedure1189

1190

Kanizsatask,indirecttestassessingpriming:1191

Atfirst,participantswereinstructedthattheywouldseeanarrayoffourtrianglesof1192

which only one, the target, had a right angle pointing down and were asked to1193

decidewhether this anglewas pointing to the left or to the right by pressing the1194

respectiveresponsekeys.1195

1196

The20experimentalblockswereprecededbyasmanypracticeblocksasneededin1197

order to fully familiarize the participants with the task requirements. Participants1198

could initiateanewblockbypressinganybuttonon thekeyboardand, ifneeded,1199

takesmallbreaksbetweenblocks.1200

1201

Participantswereinstructedtofixateasmallblackdot(0.2°wide)thatwaspresent1202

inthecenterofthescreenthroughouttheexperiment.Oneachtrial,thefixationdot1203

wasdisplayedalonefor500ms.Subsequently,amaskarraywasdisplayedfor100ms1204

andwasimmediatelyfollowedbyoneoftheprimingarrays(seeconditionsinstudy1205

design),whichwaspresentedforeither16.7msor300ms.Thisarraycouldcontain1206

oneof the twopossibleprimeshapes (realor illusory triangle)ora simplecontrol1207

stimulusamongtheotherthreecontrolstimuli.Immediatelyafterthis,atargetarray1208

containingone targetamong threedistractorswasdisplayedand remainedon the1209

screenuntiltheparticipant’sresponse.1210

1211

The responsewas self-paced;however,participantswere instructed to respondas1212

quicklyaspossible.Responsesweremadewiththeindex(forleft)andthemiddleor1213

ring finger (for right), depending on how participants felt comfortable. For every1214

response, the fixationdotprovided feedback (100ms)bychanging itscolor (green:1215

correct; red: incorrect). Then, the next trial startedwithout any delay. Figure S2A1216

showsthegeneralparadigmfortheexperimentalprocedure.1217

1218

Visibilitytask,directtestassessingprimediscrimination:1219



https://doi.org/10.1101/050526


42

Thesecondtaskassessedtheeffectivenessofthemaskbymeasuringtheconscious1220

discriminabilityoftheprimesinadirectmanner.Forthispurpose,participantswere1221

askedtomakethesamedecisionasintheprevioustask,butthistimewithrespect1222

to theprime.Thus, insteadofdiscriminatingwhether the rightangleof the target1223

waspointingtotheleftortotheright,participantsdiscriminatedthisaspectofthe1224

primeshape.1225

1226

Procedure,stimuliandtimingofthetrialsequenceinthistaskwerekeptidenticalto1227

thepreviousindirecttest,withasingleexception:Insteadofatargetarray,anarray1228

comprising new identical mask shapes covered the four stimuli locations (Figure1229

S2B).1230

1231



https://doi.org/10.1101/050526


43

1232FigureS2. Illustrationofanexampletrialsequenceintheprimingexperiment.A)Kanizsa1233task:Eachtrialwascomposedoffourframes:fixationperiod,mask,cueingarrayandtarget1234

array. After the participant’s response, the fixation point provided feedback for 100ms1235

(green:correct; red: incorrect).Thedurationofeach frame isshownonthetime-line.The1236

duration of the cueing array was varied between conditions: 16.7ms in the invisible1237

conditionsand300msinthevisibleones.Notethatintheunprimedconditions,thecueing1238

arraydidnotcontainanyprimeshapeandwascomposedexclusivelyofcontrolstimuli.B)1239

Visibility task: Trial procedurewas identical to the one in the Kanizsa task except for the1240

fourth frame in which the search array was replaced by an array of four identical mask1241

shapes.1242



https://doi.org/10.1101/050526


44

1243

1244

Dataanalysis1245

1246

Response times (inmilliseconds)andperformanceaccuracy (inproportioncorrect)1247

weremeasuredforeachofthesixconditions.Toanalyzeprimingeffects,response1248

timeswerecalculatedoncorrecttrialsonly.1249

1250

FortheKanizsatask,accuracyofresponseswaslessinformative,asthetargetarray1251

was present until the participant’s response. We reasoned that priming in the1252

indirect Kanizsa taskwouldbe reflectedby fastermean response times toprimed1253

than unprimed conditions. Therefore the priming effect was calculated by1254

subtracting response times under priming from thosewithout priming for correct1255

responses.1256

1257

Primingeffectsfromallparticipantsforallconditionswereenteredintoatwo-way1258

repeated-measures analysis of variance (ANOVA) with prime visibility (visible vs1259

invisible) and prime stimulus (real vs illusory triangle) aswithin subject factors. In1260

addition to this, further comparisons between experimental conditions (visible or1261

invisible/real or illusory) and control conditions (unprimed) were assessed with1262

paired t-tests. We also determined the significance of the priming effect by1263

comparingitagainstzerowithone-samplet-tests.1264

1265

In the visibility task, the variableof interestwas theaccuracyof responses as this1266

allowed us to test whether participants could discriminate the orientation of the1267

primes. One-sample t-testswerecarriedoutat thegroup level foreachcondition1268

individually in order to assesswhether the participants’ level of performancewas1269

significantly above chance (>0.5) and paired-tests were used to compare1270

performanceindifferentexperimentalconditions.Asinexperiment1,forallt-tests1271

wealsocalculatedthedefaultBayesFactorinadditiontofrequentiststatistics.1272

1273

1274



https://doi.org/10.1101/050526


45

Results1275

1276

IntheprimingexperimentweaskedwhethermaskedKanizsashapescouldaffordan1277

attentionalprimingeffect.WetestedwhetheraKanizsastimuluspresentedaspart1278

ofanarrayof four stimuli couldprimeparticipants to the locationofa target ina1279

visual search array presented subsequently. The prime could either be a Kanizsa1280

triangleoratriangledefinedbyarealluminancecontour.Intheunprimedcondition1281

allfourstimuliintheprimingarraywerecontrolstimuliwithoutatriangleshape.1282

1283

KanizsaTask1284

1285

Overallparticipantsmadeveryfewerrorsindeterminingwhetherthetargetpointed1286

to the left or to the right. Indeed, themean accuracy in all conditionswas above1287

0.94.Therefore,foreachexperimentalconditionwequantifiedtheeffectofpriming1288

by the difference in response time with priming subtracted from that without1289

primingoncorrectlyansweredtrials(FigureS3A).1290

1291

1292FigureS3.Resultsofexperiment2.Primingeffect(responsetimeforunprimedtrialsminus1293primedtrials)ontheKanizsatask(A)andaccuracyfordiscriminatingtheinducerlocationin1294

theVisibility task (B) for visible or invisible (masked) trials. Eachdot represents themean1295

performanceof an individual participant in eachof the conditions. The large symbols and1296

error bars denote the mean ±1 standard error for each condition. Black: real luminance1297

contour.Red: illusory (Kanizsa)contour. InB)Grey:unprimed trialsdummycoded for real1298

contourprime.Orange:unprimedtrialsdummycodedforillusorycontourprime.1299



https://doi.org/10.1101/050526


46

1300

We conducted a two-way repeated-measures analysis of variance, with visibility1301

(visiblevsinvisible)andstimulus(illusoryvsrealtriangle)aswithinsubjectfactors,to1302

compare the priming effects (response time on unprimed minus primed trials)1303

between conditions. There was a significant main effect of visibility of the prime1304

(F(1,16)=60.06, p<0.001), and a significant main effect of stimulus type1305

(F(1,16)=13.82, p=0.002). The interaction between visibility and the stimulus type1306

wasnotsignificant(F(1,16=3.52,p=0.079).1307

1308

There was a clear difference in the strength of priming between the visible and1309

invisible condition. However, the critical test is whether the priming effect was1310

significant, that is, if itdiffered fromzero (i.e.apaired t-testbetweenprimedand1311

unprimed response times). Unsurprisingly, the very pronounced priming effect for1312

visibletrialswasextremelysignificantforbothconditions(real:M=336ms,t(16)=8.2,1313

p<0.001,BF10>4.1*104; illusory:M=275ms,t(16)=6.9,p<0.001,BF10>5.8*103).There1314

was also amuchmoremodest but nonetheless very significant priming effect for1315

both stimulus types in the invisible condition when inducers were masked (real:1316

M=43ms, t(16)=3, p=0.01, BF10=5.6; illusory: M=36ms, t(16)=4.4, p<0.001,1317

BF10=80.8).1318

1319

VisibilityTask1320

1321

As in experiment 1 we used a direct Visibility test to assess whether participants1322

mighthavehadresidualawarenessof theprimingstimuliundermasking.Thetask1323

procedure was the same as in the Kanizsa task except that the search array was1324

replaced with a foil in which all four stimuli were identical and non-informative1325

about the correct answer. We quantified the accuracy (proportion correct) with1326

whichparticipantscoulddiscriminatetheorientationoftheprimingtriangle(Figure1327

S3B).1328

1329

Unsurprisingly, in the visible condition performance for discriminating the1330

orientation of a real (M=0.88, t(16)=14.9, p<0.001, BF10=9.5*107) and an illusory1331



https://doi.org/10.1101/050526


47

triangle (M=0.87, t(16)=14,p<0.001,BF10=4.0*107)was significantly above chance.1332

Conversely,performance for theunprimedconditions, that is, stimulusarrays that1333

contained no triangle shape, performance was consistently at chance (“real”:1334

M=0.52, t(16)=1.3, p=0.21, BF10=0.52; “illusory”: M=0.53, t(16)=1.3, p=0.219,1335

BF10=0.5;butnotethatthesetwoconditionsweredummycodedinthiscaseasthey1336

containedthesamestimuli).1337

1338

Critically, performance in the invisible condition showed that discrimination in the1339

primedwith an illusory triangle conditionwas also significantly above chance (M=1340

0.55,t(16)=3.24,p=0.005,BF10=9.4),anddiscriminationofa real triangleshoweda1341

similarresulteventhoughitdidnotreachsignificance(M=0.53,t(16)=1.9,p=0.075,1342

BF10=1.1).Performanceforthetwoinvisibletrianglesalsodidnotdiffersignificantly1343

(t(16)=-0.9,p=0.385,BF10=0.35).Again,asexpectedbothoftheunprimedconditions1344

were at chance level (“real”: M=0.51, t(16)=0.6, p=0.534, BF10=0.3; “illusory”:1345

M=0.51,t(16)=0.7,p=0.494,BF10=0.31).1346

1347

1348

Discussion1349

1350

Taken together, the results of the priming experiment contradict those of1351

experiment 1 and 2 , as they suggest that Kanizsa triangles rendered invisible by1352

masking could afford an attentional priming effect in a subsequent visual search1353

task.However,acontrolexperimenttestingparticipants’abilitytodiscriminatethe1354

primeshapeorientationdirectlysuggestedthatthissmallprimingeffectforinvisible1355

trials could have been due to some residual awareness of the masked triangle1356

stimuli.Whenconductingsuchtestsofawareness,itisofparamountimportanceto1357

ensure that trials with masked and unmasked conditions are interspersed. When1358

onlymaskedconditionsaretestedforawareness,itispossiblethattheparticipant’s1359

performanceisatchanceeventhoughthereisinfactresidualawarenessinthemain1360

taskduetowhathasbeenreferredtoas“primingofawareness” [71].Conversely,1361

the ability to correctly identify purportedly subliminal prime stimuli (as in our1362

primingexperiment)mayhavebeenenhancedbytheinclusionoftrialswithclearly1363



https://doi.org/10.1101/050526


48

visibleprimes[72].Eitherway,becausewecontrolledforthispossibilityourvisibility1364

test should have provided a robust test of awareness. Therefore, the most1365

parsimonious interpretation of the priming results is that participants had some1366

residual awareness of the primes and that Kanizsa contours were probably not1367

processed when inducers were masked. Using a different masking technique, for1368

examplethefastcounter-phaseflickerusedinexperiment2couldbemoreeffective1369

and provide a better test of whether unconscious Kanizsa stimuli can act as1370

attentionalcues.1371

1372

Critically,however,becauseprimingexperimentslikethisareonlyanindirecttestof1373

an effect that is not specific to illusory contour formation, this would not be1374

informative. Even if we interpret the somewhat inconclusive results of this1375

experiment as showing thatmasked triangles produced priming effects, this does1376

notprovidedirectevidencethatthisadvantageisactuallycausedbythepresenceof1377

illusorycontours.AKanizsatrianglecanclearlyprovideasalientattentionalcuebut1378

this may be due to shared features, such as the collinearity of the edges in the1379

inducersortheimplicationofasurface.Thisisalsoconsistentwithpreviousreports1380

that priming of a shape discrimination task by subliminal stimuli depends on the1381

strength of the salient region [73] and that salient regions of such stimuli are1382

detected efficiently regardless of whether they are bounded by illusory contours1383

[39,74].Findingaprimingeffectthusonlysuggeststhatattentionalcuingcanoccur1384

without awareness but it does not rule out that other factors produced this1385

attentionalcapture.Conversely,theabsenceofaprimingeffectwouldonlyconfirm1386

thatwhatever acts as attentional cue is disrupted. Therefore, theonlymeaningful1387

testthatillusorycontoursareindeedformedisviaameasurethatisspecifictothis1388

percept.1389

1390

1391



https://doi.org/10.1101/050526


induction of kanizsa contours requires awareness of the ... · 7/14/2016 · 86 a kanizsa...

Documents