Journal of Experimental Psychology: General1980, Vol. 109, No. 1, 75-97

Temporal Integration in Visual Memory

Vincent Di LolloUniversity of Alberta, Edmonton, Canada

SUMMARY

Iconic memory has often been likened to a sensory store whose contents drainaway rapidly as soon as the inducing stimulus is turned off. Instances of short-livedvisible persistence have been explained in terms of the decaying contents of iconicstore. A fundamental requirement of this storage model is that strength of per-sistence should be a decreasing function of time elapsed since the cessation—notsince the onset—of the inducing stimulation. That is, strength of visible persistencemay be directly related—but not inversely related—to the duration of the inducingstimulus.

Two complementary paradigms were utilized in the present studies. In the firstparadigm performance was facilitated by visible persistence in that the task requiredthe bridging of a temporal gap between two successive displays. In the secondparadigm (forward visual masking by pattern), performance was impaired bylingering visible persistence of the temporally leading mask. Both paradigmsyielded evidence of an inverse relationship between duration of inducing stimulusand duration of visible persistence.

More specifically, in a task requiring temporal integration of a pattern displayedbriefly in two successive portions, performance was severely impaired if the dura-tion of the leading part exceeded about 100 msec. This suggests an inverse relation-ship between duration of inducing stimulus and duration of sensory persistenceand allows the inference that visual persistence may be identified more fittinglywith ongoing neural processes than with the decaying contents of an iconic store.In keeping with this suggestion, two experiments disconfirmed the conjecture thatlack of temporal integration following long inducing stimuli could be ascribed toemergence of unitary form separately in the two portions of the display or to thetriggering of some sort of discontinuity detection mechanism within the visualsystem. In added support of a "processing" model, two further studies showed thatthe severity of forward masking by pattern declines sharply as the duration of theleading mask is increased.

This pattern of results is equally unsupportive of a storage theory of iconicpersistence as of perceptual moment theory in any of its versions. This is so becauseboth theories regard interstimulus interval rather than stimulus-onset asynchronyas the crucial factor in temporal integration. Neither can the results be explainedin terms of receptor adaptation or of metacontrast suppression. The theory ofinhibitory channel interactions can encompass the more prominent aspects of theresults but fails to account for foveal suppression and for some crucial temporaleffects.

Copyright 1980 by the American Psychological Association, Inc. 0096-3445/80/090l-0075$00.75

75

76 VINCENT Di LOLLO

Why is it that two sequential visual dis-plays, separated by a brief temporal gap,often appear to be temporally contiguous oroverlapping? Two classes of answers havebeen suggested. One states that visual per-sistence, in the form of iconic memory(Neisser, 1967), is the basis for the per-ceptual bridging of the gap. According tothis theory, the contours of the temporallyleading display are stored in a sensoryregister (Atkinson & Shiffrin, 1968), alsocalled iconic store (Neisser, 1967), whosecontents fade over several hundred milli-seconds from the termination of the inducingstimulus. If the temporal gap does notexceed the duration of iconic persistence,the two displays will be seen as temporallyoverlapping, because the fading icon of thefirst will still be perceptually available at thetime the second is presented (Eriksen &Collins, 1967; Haber, 1971).

Alternately, the perceptual moment hy-pothesis assumes that the visual systemquantizes incoming stimulation into tempo-rally successive bundles, or "moments" offinite duration, often estimated at about 100msec (e.g., Allport, 1968). Visual eventsthat fall within the same moment are heldto be temporally integrated. Several versionsof the perceptual moment hypothesis havebeen proposed. Among the "discrete mo-ment" hypotheses (White, 1963), one ver-sion assumes that the train of moments isself-paced and that it is possibly linked tothe alpha rhythm of the electroencephalo-gram (EEC) (e.g., Kristofferson, 1967;Murphree, 1954; Walter, 1950). A closelyrelated version, noted by Haber and Hersh-enson (1973), is that the timing sequencemay be stimulus dependent in that its initia-tion may be time locked to the onset of theinducing stimulus. On this alternative, the

Portions of this article were presented at themeeting of the Lake Ontario Visionary Establish-ment, Niagara Falls, Ontario, Canada, February1977.

This research was supported by National Re-search Council of Canada Grant A0269.

Requests for reprints should be sent to VincentDi Lollo, Department of Psychology, BiologicalSciences Building, University of Alberta, Edmon-ton, Alberta, Canada T6G 2E9.

train of moments would begin at the onsetof a stimulus and would cease at the termina-tion of the moment containing the trailingedge of the inducing stimulus. A third ver-sion is Allport's (1968) "travelling mo-ment," which is conceptualized as a movingtemporal window containing all the informa-tion registered by the visual system duringthe preceding 100 msec or so; with thepassage of time, new events enter thewindow and old events drop out. Despitetheir differences, all versions provide thesame answer to the initial question: Twosequential stimuli separated by a temporalgap are perceived as temporally contiguousor overlapping if the trailing portion of thefirst stimulus falls within the same momentas the leading portion of the second.

Neither iconic theory nor the perceptualmoment hypothesis has remained entirelyunchallenged (e.g., Efron, 1970; Holding,1975). The results of the experimentsreported below are equally unsupportive ofa "storage" theory of iconic memory and ofany version of the perceptual moment hy-pothesis. Instead, the present results, as wellas other results reported in the literature,suggest that except for retinal afterimages,visual persistence may be more properlyregarded as the product of ongoing neuralprocesses of finite duration. This interpre-tation of sensory persistence is articulatedand examined in the remainder of this article.

Experiment 1

Experiment 1 is a replication and expan-sion of a recent study (Di Lollo, 1977) thatinvestigated perceptual integration of a con-figuration whose parts were displayed in twosequential portions separated by a temporalgap. The technique, first described by Hog-ben and Di Lollo (1974), employed an oscil-loscopic display consisting of a square matrixof 25 dots arranged in five rows and columns.One of the 25 dots, chosen randomly onevery trial, was not plotted. The subject'stask was to name the coordinates of themissing dot within the matrix. The displaysequence consisted of three steps: First, anaggregate of 12 dots, chosen randomly fromthe matrix, was displayed for a period that

TEMPORAL INTEGRATION 77

varied between 10 and 200 msec; next, a10-msec temporal gap was allowed to elapsewith no dots shown on the screen; andfinally, the remaining 12 dots were displayedfor 10 msec. The only temporal factor thatvaried across conditions was the duration ofthe first portion of the display. Since therewere no contextual cues outlining the frame-work of the matrix, identification of thelocation of the missing dot depended on thesimultaneous perceptual availability of thetotal matrix form (see Hogben & Di Lollo,1974).

Simply stated, success at this task dependsupon the bridging of the temporal gap be-tween the two portions of the display.According to iconic theory, the brief 10-msecgap should be easily filled by the iconic per-sistence of the first 12 dots. More impor-tantly, although a longer exposure durationof the leading portion of the display may welllead to improved performance (perhaps byestablishing a sharper icon), it should neverproduce an impairment in performance. Thisfollows from the joint considerations thatthe icon is said to begin fading from thetermination of the inducing display and thatthe duration of the temporal gap remainedunaltered throughout the experiment.

Predictions based on perceptual momenttheory depend on which version is con-sidered. According to the "travelling mo-ment" hypothesis (Allport, 1968), incre-ments in the duration of the leading displayshould neither facilitate nor impair perform-ance ; since the temporal separation betweenthe trailing edge of the first display and theleading edge of the second was the same inevery condition, both should appear withinthe "temporal window" for a period thatdepends on the width of the window, not onthe duration of the leading display. Accord-ing to "discrete moment" interpretations(White, 1963), increments in the durationof the leading display should yield a non-monotonic performance curve that oscillatesdiscretely between a high level when bothdisplays share portions of the same momentand a low level when the temporal gapstraddles two separate moments.

Neither iconic theory nor perceptual

moment theory could predict a rapid andsevere impairment in performance when theduration of the leading display exceededabout 100 msec. Yet this is what the resultsunambiguously showed.

Method

The equipment and method for producing thestimuli have been described in detail by Hogbenand Di Lollo (1974). Briefly, a display consistingof 24 dots of the 25 dots defining a 5 X 5 squarematrix was displayed on a gridless Tektronix 602point plotter equipped with fast PIS phosphor.Displays were generated by a PDP-8/L computer,which also performed all the timing and scoringfunctions.

The subject sat inside a light-proof cubicle andviewed the dimly lit display surface binocularlythrough a Tektronix Model 016-0154-00 viewinghood. From the viewing distance of 42 cm, the dotmatrix subtended a visual angle of approximately4.1°. The subject pressed a button to initiate atrial consisting of the following sequence of events:First, 12 matrix dots, chosen randomly on eachtrial, were displayed for either 10, 40, 80, 120, 160,or 200 msec; next, an interstimulus interval (ISI)of 10 msec was allowed to elapse; finally, another12 dots (chosen randomly from the remaining 13matrix locations) were displayed for 10 msec. Thesubject's task was to identify the location of themissing dot and to name its row and columncoordinates through an intercom to the experi-menter, who entered them on a teletypewriter.

An experimental session consisted of 120 ran-domly ordered trials composed of 20 presentationsof the dot matrix for each of the six experimentalconditions defined by the six durations of theleading portion of the display. The 120 trialsoccurred in a different random order for eachsession and were completed within 15 min. Datawere collected on five separate sessions from eachof three subjects: the author, a male graduate(RGB) and a female undergraduate student(SKP). All had normal or correct-to-normalvision.

Brightness cq^lalisation procedures. For stim-ulus durations shorter than about 100 msec, per-ceived brightness depends jointly on the intensity ofthe stimulus and on its duration. Time-intensityreciprocity over this range of stimulus durationshas been found at threshold levels of intensity(Bloch's law) and at suprathreshold levels(Butler, 1975).

In five of the six experimental conditionsdescribed above, the duration of the leading displayfar exceeded that of the trailing display. Theresultant brightness mismatch between the twoportions of the display could interfere with accuracyof performance (Eriksen & Collins, 1967). Moreimportantly, the level of interference due to

78 VINCENT Di LOLLO

100

90

80

70

60

50

40

30

20

10

ROB

40 8O ISO I6ODURATION OF FIRST PORTION

OF DISPLAY (MSEC)

200

Figure 1. Percentage of errors plotted separatelyfor each subject at each of the six durations of theleading portion of the display.

brightness mismatch would increase with incre-ments in the duration of the leading display andwould thus entirely confound the effects of theindependent variable.

Brightness mismatch was eliminated as animportant source of confounding through computer-controlled modulation of the intensity of eachdisplay. The longer displays were plotted at lowerintensity values so that all displays appeared of thesame brightness regardless of duration. Technically,this was made possible by an analog-to-digitalconverter on the z axis of the display interface,which allowed independent variation of the intensityof each dot from minimum (invisible) to maximum(flaring) in 256 steps. The intensity values appro-priate for each duration were determined separatelyfor each subject in a preliminary psychophysicaltask. The procedure, similar to that used byServiere, Miceli, and Galifret (1977) has beendescribed in detail by Di Lollo (1979).

Results and Discussion

Figure 1 shows the percentage of errorsat each duration of the leading display sep-arately for each subject. Performance wasessentially errorless at leading-stimulus dur-ations up to about 100 msec, but it deterio-rated rapidly and markedly thereafter. Thephenomenal appearance of the display shouldbe noted: At the briefer durations of theleading portion, all dots in the matrix wereseen simultaneously, but at the longer dur-ations the display was seen in two discreteparts separated by a noticeable temporal gap.

Casual observations in the laboratory showedthat the two portions of the display werestill perceived as discrete and that perform-ance remained severely impaired even atleading-stimulus durations of several sec-onds. This was true regardless of practice,whether the observer was a regular subjector a naive visitor in the laboratory. Also, itmust be noted that entirely similar resultswere obtained without employing the bright-ness-equalization procedure described above.Allowing the brightness of the leading dis-play to covary with its duration producedsome degree of brightness mismatch, butthe response curves remained essentiallyunaltered.

Surprisingly, apparent-motion effects wereweak or totally absent even at values ofstimulus-onset asynchrony (SOA) wellwithin the broad range of values (50-100msec) where optimal apparent motion hasbeen reported (Kahneman, 1967). Thiscould be due to the lack of any systematicspatial relationship between the locations ofthe dots in the first portion of the displayrelative to the dots in the second portion.Thus, apparent motion in one predominantdirection could occur only very rarely when,by chance, most dots in the trailing displayhappened to occupy matrix locations thatwere consistently to one side of the dots inthe leading display. On most trials, no clearimpression of motion was readily apparenteven between pairs of dots: At the briefervalues of SOA, the two sets of dots wereseen simultaneously; at the longer values ofSOA, the compelling perception was not oneof motion but one of replacement of the firstset of dots by the second set.

Since the experimental task was practicallyimpossible unless the two portions of thedisplay were actually seen as one, it seemslikely that some sort of visual persistencewas necessary for bridging the temporal gap.If some form of persistence is assumed, theresults strongly suggest that persistence wasavailable when the inducing display was briefbut not when it was long. Although thissuggestion is consonant with similar infer-ences drawn by other investigators (Efron,1973; Haber, 1971), it is entirely inconsistent

TEMPORAL INTEGRATION 79

with the view that visual persistence is pro-duced by an iconic store whose contentsbegin to fade upon termination of the induc-ing stimulus (Neisser, 1967; Sakitt, 1976;Sperling, 1960), Rather, this pattern ofresults suggests that visual persistence beginsat the onset of the display and continues fora fixed duration (about 100 to 150 msec)independent of the duration of the inducingstimulus.

Theoretical implications of this view ofsensory persistence are elaborated at the endof the present article, where the possibleroles of metacontrast masking and of inhibi-tory channel interactions (Breitmeyer &Ganz, 1976) are also considered. Here itmay be helpful to anticipate the main aspectsof the argument so as to permit a coherentstatement of the reasoning behind the experi-mental work. In this reasoning, visual per-sistence would be identified more fittinglywith the activity of sensory coding mech-anisms engaged in an early phase of visualinformation processing (here referred to asrecruiting phase) than with the contents ofan iconic store. The mechanisms in thisphase would be activated by the onset of adisplay, and activity would continue to theend of the phase whether or not the inducingstimulus was still on display. Thus, visualpersistence would be available only for induc-ing stimuli whose duration did not exceedthe period of early sensory coding activity,and its duration would correspond to thedifference between the duration of the induc-ing stimulus and the duration of the encod-ing process. On this account, the extent towhich two sequential stimuli are seen tooverlap temporally would depend on theperiod for which the two stimuli are con-currently active within the recruiting phase.

Support for the argument that the first 12dots were no longer visible at the termina-tion of the longer leading displays is suppliedby an examination of the incidence of errorsrelative to the two portions of the display.An error was made when the subject nameda matrix location where a dot had actuallybeen plotted instead of naming the matrixlocation that had been left empty. On anygiven trial, an incorrect response could

LEADING DISPLAY

w£

iK

UJU

o

I-111UKIDa

120 16O 200

DURATION OF FIRSTPORTION OF DISPLAY (MSEC)

Figure 2. Mean proportions of errors made byincorrectly identifying as missing a dot that hadactually been plotted in the first portion of thedisplay (filled bars) or in the second portion of thedisplay (unfilled bars). (The data were averagedover the three subjects, separately at each durationof the leading stimulus.)

result from naming a dot that had beenplotted either in the first or in the secondportions of the display. Figure 2 shows thepercentage of errors attributable to eitherthe first or the second portions of the dis-play at each of the six exposure durationsof the leading stimulus.

Proportions of errors in the leading andtrailing displays were approximately equalfor leading-stimulus durations up to about100 msec, but diverged sharply thereafter,with most errors being attributable to theleading portion. These results are clearlyconsonant with the present claim that thesensory persistence of the contents of theleading display diminished sharply afterexposure durations greater than about 100msec.

Before we pursue this line of reasoning,it should be recognized that failure of tem-poral integration at the longer exposuredurations could be brought about by factors

80 VINCENT Di LOLLO

other than termination of an early phase ofprocessing. The next two experimentsexamined the role of emergence of unitaryform and of visual masking as possibledeterminants of the effect.

Experiment 2

It may be suggested that at exposuredurations longer than about 100 msec, thevisual system can integrate the 12 elementsof the leading display into a unitary config-uration, which, in turn, may exhibit a tend-ency to remain perceptually segregated fromthe configuration formed by the 12 elementsof the trailing display. According to thisargument, lack of integration between thetwo portions of the display would be duenot to lack of visual persistence of localspatial detail (i.e., of the individual dot) butto the emergence of unitary form, whichwould override the effect of persistence ofthe component elements.

One way of examining this alternative isto vary the local spatial detail while leavingoverall configuration unaltered. In Experi-ment 2, each matrix element consisted of asmall equilateral triangle defined by threedots. Twelve elements, chosen randomly onevery trial, were displayed for the samedurations as in Experiment 1 and werefollowed, after an ISI of 10 msec, by theremaining 12 elements, which were plottedfor 10 msec. Allowing for the fact thattriangles instead of single dots were used asmatrix elements, this procedure was identicalto that of Experiment 1 and served as acontrol condition in Experiment 2. Themajor departure from the previous studywas provided by an experimental conditionidentical to the control condition except inone respect: During the last 10 msec ofexposure of the leading display, the threedots defining each of the 12 elements werereplaced by another three dots which,instead of an upward-pointing triangle,formed a downward-pointing triangle. Thisprocedure changed the local detail of thedisplay but left unaltered the overall config-urations that happened to be made by theleading 12 elements on any given trial. Itwas reasoned that, in so far as level of per-

formance is affected by emergence of unitaryform rather than by exposure duration oflocal detail, the two conditions should yieldsimilar results.

Method

Method, subjects, and procedures of Experiment2 were the same as for Experiment 1 with thefollowing exceptions: Each matrix element wasmade up of three dots arranged in the form of anequilateral triangle pointing upward; the angularseparation between pairs of dots in a triangle was.38°; and the total matrix subtended a visual angleof approximately 4.4°. The control condition was anexact replication of Experiment 1 with trianglesrather than single dots as matrix elements. Theexperimental condition was the same as the controlcondition except that 10 msec before the termina-tion of the leading display, the three dots in eachof the 12 elements were replaced by three new dotsthat defined an equilateral triangle with the tippointing downward. This was equivalent toinstantly rotating the triangles by 180°. The exactsequence of events at each of the six durations ofthe leading configuration in the experimental con-dition is illustrated in Figure 3. Data werecollected in five experimental and five controlsessions from each subject in a different randomsequence.

Results and Discussion

The results illustrated in Figure 4 areunambiguous: The performance curves ob-tained with the "unrotated" triangles (solidlines, Figure 4) were entirely similar to theresults of Experiment 1 (Figure 1); incontrast, the performance curves obtainedwith the "rotated" triangles (segmentedlines, Figure 4) were essentially flat through-out the domain. The results of the "rotated"condition suggest that sufficient visual per-sistence was available to permit temporalintegration of the total matrix form irrespec-tive of the duration of the leading (pre-rotation) display. In turn, these results sug-gest that visual persistence is an attribute ofeach individual dot rather than of the overallconfiguration formed by the first 12 elements.

The phenomenal appearance of the"rotated" displays was in accordance withthis conclusion. At the briefer durations ofthe leading portion, the appearance of thetotal display corresponded to a temporalintegration of all its parts; that is, each of

TEMPORAL INTEGRATION 81

C30J ,*.

(70). .

\ \ (110),*,

\ \ (ISO)/.

\ \ "BO) .*.

• ••

• ••

• •

•

• ••

• •

•

• ••

ouin

0

2

•• •

•• •

•• •

•• •

•• •

•• •

(10 MSEC) (10 MSEC)

ELEMENTS 1-12

Figure 3. Display sequence at each of the six stimulus durations in the experimental condition.(In the five sequences in which the duration of the leading configuration exceeded 10 msec, theinitial 12 matrix elements were in the form of upright triangles plotted for the number ofmilliseconds shown in parentheses. The upright triangles were replaced by downward trianglesfor the last 10 msec of the leading configuration. After an interstimulus interval [ISI] of 10msec, the remaining 12 elements were shown for 10 msec.)

the first 12 elements appeared as an aggre-gate of six dots resulting from the temporalintegration of an upright and a downwardtriangle. This aggregate was then temporallyintegrated with the trailing 12 elements toform a total representation similar to thatshown in Figure 5. At durations of the lead-ing display exceeding about 100 msec, thedisplay seemed to consist of two sequentialportions: 12 elements (upward triangles)were seen first and were then followed by24 other elements (12 upward and 12 down-ward triangles) from which the missingmatrix element could be chosen. Regardlessof the duration of the leading display and ofthe phenomenal appearance of the total con-figuration, identifying the location of themissing element was an invariably easy taskin the rotated condition.

As a whole, these results strongly impli-cate exposure duration of local detail ratherthan exposure duration of overall config-uration as the more potent determinant ofsensory persistence. A change in local detail(rotated condition) generated a new leaseof persistence, which enabled perceptualbridging of the temporal gap despite theunchanged overall configuration of the lead-ing display.

Experiment 3

Very similar experimental designs wereemployed in Experiment 1 and in the "un-rotated" condition of Experiment 2: In bothcases the display consisted of two flashesseparated by a temporal interval. Opera-tionally, this sequence of events is akin tothe sequence of test and masking stimuli inforward or backward visual masking.According to this account, it is necessary toconsider whether the impairment in per-formance found in Experiments 1 and 2could be ascribed to masking effects.

Since there is no a priori reason for defin-ing either portion of the display as teststimulus or mask, both forward and back-ward masking should be considered. It isimmediately obvious, however, that forwardmasking is an untenable alternative, because,as shown in Figure 2, almost 90% of allerrors were due to confusions between theempty matrix location and one of the ele-ments in the leading portion of the display.Clearly, if the results reported thus far areto be interpreted in terms of masking, theleading display must be designated as thetest stimulus and the trailing display as themask. This leads to the compelling but

82

I

100

90

80

70

60

50

30

20

10

0

ROTATED- VDL O O

R6B • •SKP A A

VINCENT Di LOLLO

NON-ROTATED

10 40 80 120 160

DURATION OF FIRST PORTIONOF DISPLAY (MSEC)

200

Figure 4. Percentage of errors plotted separately for each subject at each duration of theleading configuration in the experimental ("rotated") and in the control ("nonrotated")conditions.

hardly tenable proposition that in Experi-ments 1 and 2, the severity of maskingincreased as the exposure duration of thetest stimulus was increased.

Nevertheless, it is conceivable that adegree of perceptual interference may havearisen at the longer duration of the leadingdisplay, perhaps through the triggering ofsome form of discontinuity-detection mech-anism in the visual system that effectivelysegregated the two portions of the displayone from the other (Eriksen & Collins, 1968;Granit, 1947). This alternative was ex-amined in Experiment 3. The principal aimwas to examine the relationship betweenstimulus duration and duration of sensorypersistence under conditions designed toavoid the possibility of perceptual interfer-ence arising from the two-flash display modeof Experiments 1 and 2. This was achievedin Experiment 3 by altering the displaysequence: Instead of displaying the matrixin two flashes of 12 dots, each of the 24 dotswas briefly displayed only once in random

sequence at a regular inter-dot interval(with one exception as noted below). Ashas been shown by Hogben and Di Lollo(1974), if the inter-dot interval is such thatthe total duration of plotting of the 24 dotsexceeds about 120 msec, the matrix appearsto have several missing elements. Interest-ingly, Hogben and Di Lollo (1974) foundthat the apparently empty locations werethose where a dot had been plotted morethan about 120 msec before the terminationof the display; that is, the persistence of thedots plotted early in the sequence had van-ished before the last dot had been shown.

In Experiment 3, each dot was shownonly once for 1.5 //.sec, except for the 12thdot in the sequence, which was plotted for100 msec. It was reasoned that if sensorypersistence is an attribute of each separatedot and if its duration is inversely related tothe duration of the inducing stimulus, the12th dot should have negligible persistenceand should hence be easily confused with thetruly missing dot. Experiment 3 was a repli-

TEMPORAL INTEGRATION 83

cation and expansion of a recent study (DiLollo, 1977).

Method

The display consisted of the dot matrixdescribed in Experiment 1. In the control condi-tion, each of the 24 dots was plotted only once for1.5 jusec in a random sequence that changed onevery trial. The temporal interval between twosuccessive dots was 10 msec. This yielded a totalplotting interval of approximately 230 msec fromthe onset of the first dot to the termination of the24th dot. The experimental condition was thesame as the control condition except for the 12thdot in the plotting sequence, which remained onview from the time of plotting of the second dotuntil 10 msec after the plotting of the llth dot,a total of 100 msec. The brightness of the 12thdot was made indistinguishable from that of allother dots through the compensation proceduredescribed in Experiment 1. The plotting sequencewas randomized on every trial so that each of the25 matrix locations had an equal probability ofbeing the 12th. Data were collected from threesubjects (the author, a male student unaware ofthe purpose of the research, and an experimentallynaive paid female subject) in five experimentaland five control sessions sequenced randomly foreach subject. One session consisted of 100 trialsand was completed in about 10 min.

Results and Discussion

On every error trial the computer pro-gram identified the ordinal position in theplotting sequence of the dot that had beenerroneously named as missing. The propor-tion of trials on which a dot shown at eachof the 24 ordinal positions was incorrectlyidentified as missing is shown in Figure 6separately for each subject. In every case,the 12th dot in the plotting sequence wasnamed as missing much more frequentlywhen its duration was long than when it wasbrief (Figure 6, experimental vs. controlconditions). Similar results were obtainedwith several other subjects, and congruentresults were obtained with the "long" dot indifferent ordinal positions within the plottingsequence.

Spreading the dots regularly over theentire plotting interval obviated any maskingeffects possibly associated with the two-flashmode of presentation. In addition, the errordistributions plotted in Figure 6 deny that

Figure 5. Phenomenal appearance of the displaysequence in the experimental ("rotated") conditionat exposure durations of the leading configuration(10-80 msec) that permitted perceptual integrationof the total display sequence.

the termination of the 12th dot might havetriggered some sort of discontinuity-detec-tion mechanism within the visual system.Had this been so, all dots plotted before thetermination of the 12th dot would havebecome perceptually segregated from theremaining dots. On the contrary, the dotsthat were plotted just before or just afterthe 12th dot were named as missing muchless frequently than the 12th dot.

What the results strongly suggest is, first,that increasing the exposure duration of agiven element leads to a decrement in theduration of sensory persistence of that ele-ment, and second, that the effect is confinedto a given element and does not interferewith the sensory persistence or with theperceptual availability of temporally neigh-boring elements in the display sequence.

Though these considerations are con-sonant with a "process" theory of visualpersistence, it must be noted that activity ofsensory coding mechanisms at an early stageof processing is unlikely to have been thesole determinant of the results of Experi-ment 3. Had this been so, the proportion oferrors due to naming the 12th dot as missingshould have been precisely the same as theproportion of errors involving the second

84 VINCENT Di LOLLO

.16 [

.14

.12 •

.10-

.08

.06 •

.04 •

.02

0 .

SUBJECT PDS

6 B 1O 12 14 16 18 20 22 24

.22

.20

.18

in

I"Jj .14

IL

O .12

.08

.06

.04

.02

0

• • DURATION OF 12th DOT. 1OO MSECO O CONTROL

SUBJECT LDC

10 12 14 16 18 20 22 24

2 4 6 8 1O 12 14 16 18 2O 22 24ORDINAL SEQUENCE OF PLOTTING OF THE DOTS

TEMPORAL INTEGRATION 85

dot in the sequence (Figure 6). Instead,overall, the 12th dot was named as missingabout as frequently as the fourth dot. Thisimplies a discrepancy in duration of per-sistence of the order of about 20 msec andis clearly in need of explanation. But thesize of the discrepancy is minor in compari-son with the massive differences betweenexperimental and control conditions withrespect to the 12th dot. These differences arecounterintuitive and are consonant with a"process" theory—as distinct from a "stor-age" theory—of visual persistence.

Collectively, the results of Experiments1, 2, and 3 cannot be explained either interms of emergence of unitary form or interms of visual masking. This is not to saythat these and similar factors were entirelyirrelevant to the outcome. Indeed, the pos-sible effects of metacontrast and of inhibitorychannel interactions (Breitmeyer & Ganz,1976) are examined in detail later in thisarticle. On the other hand, the results areconsistent with the inference that duration ofvisual persistence is inversely related to theduration of the inducing stimulus. Thisinference, based on the present indirect esti-mates is entirely consonant with the resultsof direct estimates of persistence in visionand in other sensory modalities.

For example, Efron (1973) presented abrief, dimly lit disk and an auditory click inrapid succession and required the subjectsto synchronize the click with the terminationof the light disk. He found that when theduration of the disk was less than about130 msec, the subjects adjusted the click atabout 130 msec after the onset of the disk.Efron reasoned that the dark interval thatnecessarily elapsed between the two stimuliunder these circumstances was rilled by thesensory persistence of the first display. Ofeven greater interest was Efron's findingthat the subjects' performance was entirelyaccurate for first-stimulus durations greaterthan about 130 msec; that is, the subjects

adjusted the onset of the click to coincideexactly with the termination of the lightdisk. On the basis of these results, Efronsuggested that the duration of sensory per-sistence combined additively with the dur-ation of the inducing stimulus up to amaximum of about 130 msec. Thus, a stim-ulus of 10 msec duration will induce sensorypersistence of about 120 msec, but a 100-msec display will induce sensory persistenceof only 30 msec duration.

An inverse relationship between stimulusduration and sensory persistence has alsobeen reported by Haber (1971) and hisco-workers. Although the exact estimate ofsensory persistence varied with the experi-mental paradigm and may have been affectedby variations in response criterion (Haber &Hershenson, 1973), Haber reported that theduration of sensory persistence decreased asthe duration of the inducing stimulusincreased.

Some degree of perplexity, however, mayarise regarding the generality of these find-ings. The present studies and the studies ofEfron (1973) and Haber (1971) employedstructurally simple stimuli (e.g., the presentdot matrix or Haber's disk-shaped display).It may well be asked whether an inverserelationship between exposure duration andvisual persistence would also be obtainedwith more complex or more meaningfuldisplays. The next two experiments weredesigned to answer this question and toexamine the inverse relationship with yetanother experimental paradigm.

Experiment 4

Sensory persistence may be assumed tohave facilitated performance in the precedingthree studies by allowing perceptual inte-gration of successive stimuli. However,perceptual integration need not always behelpful: Visual persistence of a leadingdisplay may actually impair performance if

Figure 6. Temporal distribution of errors for each subject in the experimental and In the controlconditions. (The abscissa indicates the ordinal position of the dots in the plotting sequence,irrespective of spatial location within the matrix. The ordinate shows the proportion of trialsin which a dot plotted in the indicated ordinal position was incorrectly identified as missing.)

86 VINCENT Di LOLLO

the experimental task requires unencumberedperception of a trailing stimulus. Preciselythis requirement is made in studies of for-ward masking, in which a leading set ofmasking contours impairs perception of atrailing test stimulus. Indeed, a commoninterpretation of forward masking effects inpatterned displays is that the perceptualrepresentations of masking and test stimuliare temporally integrated as in a photo-graphic double exposure (Eriksen, 1966;Scheerer, 1973). In turn, identification ofthe test stimulus is impaired because thevisual system cannot disentangle the testcontours from the contours of the mask.

Perception of the test stimulus would befacilitated if the visual persistence of theleading mask could be reduced or eliminated.On a storage theory of visual persistence,this could be achieved by allowing a suffi-cient temporal interval between the termina-tion of the mask and the onset of the teststimulus. But an alternative and less intui-tive procedure is suggested by the results ofthe three experiments reported above: Theduration of visual persistence of the leadingmask may be reduced by increasing itsexposure duration. Since, according to thepresent viewpoint, long stimuli have little orno persistence, it should be expected thatforward masking due to perceptual integra-tion of contours would take place if the maskwere brief but not if it were long.

Method

Equipment and viewing conditions were the sameas in the previous studies. The test stimulus wasan uppercase alphabetic character, similar toLetraset No. 8831, chosen randomly on every trialfrom the full set of the English alphabet. Thecharacter was plotted within an imaginary squaresubtending approximately .5° of visual angle,immediately above a fixation dot located in thecenter of the screen. The masking stimulus wasan aggregate of portions of alphabetic characterscorresponding roughly to three whole charactersin amount of contour. The mask was shown withinthe same spatial confines as the test stimulus. Anew set of masking contours was assembled onevery trial.

In response to a button press by the subject, themasking stimulus was displayed for either 20, 40,80, 160, 320, or 640 msec and was followed immedi-ately by the test stimulus, which was displayed in

the same location as the mask for 20 msec in everycase. In an additional control condition, both thetest stimulus and the masking stimulus were dis-played simultaneously for 20 msec. Each of theseven conditions appeared 20 times in each experi-mental session. The 140 trials were sequenced in adifferent random order in every session and werecompleted in less than IS min. The brightness ofthe displays was equalized according to the pro-cedure described in Experiment 1. Data werecollected from four subjects: the author and threeundergraduate students unaware of the purpose ofthe research. Each subject served in five experi-mental sessions spaced over several days.

On any given trial, the sequence of events wasas follows: When the subject pressed a hand-heldbutton, the fixation point disappeared and theappropriate display was shown. The subject identi-fied the test character (or guessed if not sure) andcommunicated the response over an intercom to theexperimenter who entered it on a teletypewriter.The computer scored the response, set up the condi-tions for the next display, and then turned on thefixation point to indicate readiness for a new trial.

Results and Discussion

Figure 7 shows the results separately foreach subject. In every case, performanceimproved rapidly as the duration of the lead-ing mask was increased. These results areclearly inconsistent with a storage theory oficonic persistence, which would demand anundiminished level of masking throughoutthe domain. It must be noted that theimprovement in performance was the resultof increasing the duration of the mask, anoperation generally regarded as leading toincreased severity of masking in other para-digms. In this sense, the results are incon-sistent with a simple masking interpretationthough ostensibly consistent with the hypoth-esis of inhibitory channel interactions asdiscussed below.

Though they defy explanation in terms ofa storage theory of persistence, these resultscan easily be interpreted in terms of theduration of sensory coding events takingplace at an early (recruiting) phase ofinformation processing: As duration of theleading mask was increased, the intervalduring which the two stimuli were beingprocessed concurrently within the recruitingphase diminished. In other words, as theduration of the mask was increased, itsvisual persistence diminished while leaving

TEMPORAL INTEGRATION 87

the persistence of the test stimulus un-affected. At exposure durations of the maskthat exceeded the duration of the recruitingphase, the two displays did not overlap atall within that phase. This meant, in presentterms, that the mask's persistence was ex-hausted, and hence it could not merge with(and degrade) the representation of thetrailing test stimulus. The phenomenalappearance of the display was precisely aswould be expected on this interpretation: Atzero SOA, test stimulus and mask were seenas a single set of contours. The two displaysappeared to begin and to end simultaneously.As the SOA was lengthened, the contours ofthe test stimulus appeared to outlast the con-tours of the mask by an increasingly longmargin, which, in turn, facilitated identifica-tion of the test character.

If we pursue this line of reasoning a stepfurther, it becomes clear that test and mask-ing stimuli need not be displayed separatelyin time for the test character to be easilyidentifiable. Accurate performance should bepossible even if test and masking stimuliwere displayed simultaneously provided thatthe combined display was immediately pre-ceded by a display of the masking contoursalone for at least 100 msec. This follows

from the joint premises that duration ofvisual persistence is inversely related to theduration of the inducing stimulus and thatpersistence is an attribute of local detail,rather than of the overall configuration.Based on these premises, two displays havingdifferent durations but simultaneous termi-nations should exhibit the type of relation-ships illustrated in Figure 8. Namely, if bothstimuli are displayed briefly and simulta-neously, there should be entire overlap be-tween both stimuli and their persistence(Figure 8, upper portion). If the durationof the masking contours is increased, thesensory persistence of the mask will bediminished by a corresponding amount(Figure 8, middle portion). The diminishedeffective overlap should produce less inte-gration between the two sets of contoursand thereby less masking. Finally, if theleading masking contours are displayed forlonger than the duration of the recruitingphase (Figure 8, lower portion), the effec-tive overlap between the two stimuli wouldbe diminished to an irreducible minimum(i.e., the duration of physical overlap) andso would the severity of masking. Thesepredictions are examined in Experiment 5.

Displaying the stimuli in the sequence just

100

90

80

70

60

I 50

40

30

20

10

00 2040 80 160 320

STIMULUS-ONSET ASYNCHRONY (MSEC)

640

Figure 7. Percent correct responses plotted separately for each subject at each of the sevenlevels of stimulus-onset asynchrony.

VINCENT Di LOLLO

MSK

TS

T ~ ~~1

_ J

Experiment 5

Method

Figure 8. Schematic representation of the hypoth-esized relationship between exposure duration of astimulus (solid contours) and the duration ofvisual persistence induced by that stimulus (seg-mented contours). ("Effective overlap" refers tothe time interval during which test and maskingstimuli are held to be perceptually available at thesame time. TS = test stimulus ; MSK = maskingstimulus.)

described allows the introduction of a fur-ther control condition. It could be suggestedthat the improvement in performance inExperiment 4 may have been due, at leastin part, to an improvement in the visualsystem's readiness to process temporallytrailing contours. That is, at the longer dur-ations of the leading mask there may havebeen more time in which to perform theessential tasks of focusing, convergence, andrapid adaptation before the onset of the teststimulus. With this option, performance mayhave improved with increments in maskduration because the visual system, havingbeen primed for some time by the leadingdisplay, may have been more attuned toprocessing the contours of the test stimulus.In Experiment 5, this alternative was testedas follows: Test and masking stimuli werealways presented in a simultaneous displaypreceded by a variable-duration display con-taining either the contours of the maskingstimulus (experimental condition) or a setof contours totally unrelated to the maskingcontours (control condition). It was rea-soned that performance in the control condi-tion would provide an index of whatevereffects priming per se might have on identi-fication of the test stimulus.

Subjects, method, and procedure in Experiment5 were the same as in Experiment 4 with thefollowing exceptions. In the experimental condi-tion all stimuli terminated with a combined 20-msec display of test and masking contours. Thecombined display was immediately preceded by adisplay of the masking contours alone for either0, 20, 40, 80, 160, 320, or 640 msec. The controlcondition was the same as the experimental condi-tion except that an entirely new set of maskingcontours replaced the original set simultaneouslywith the onset of the test stimulus. In the sixtemporal conditions where the combined displaywas preceded by a display of the mask, one set ofmasking contours was shown during the initialperiod and a completely different set was shownfor the last 20 msec of the display. Every subjectserved in five experimental and five controlsessions in a different random sequence.

Results and Discussion

Individual results illustrated in Figure 9clearly show that temporal separationbetween masking and test stimuli is notnecessary for unimpeded perception of thelatter, provided the masking contours havebeen on view alone for some time. Thisresult is entirely consonant with the view-point outlined in the preceding discussionand illustrated in Figure 8, but not with asimple "storage" theory of visual persistence.That is, it is not immediately obvious why,of two stimuli that terminate simultaneously,one (the briefer one, at that) should havegreater iconic prominence and longer per-sistence than the other.

Changing the masking contours simul-taneously with the onset of the test stimuluscompletely nullified the facilitatory effects ofa long leading display (Figure 9, segmentedlines). This strongly suggests that "prim-ing" action by the longer leading displayswas a negligible determinant of the improve-ment in performance in the experimentalcondition (Figure 9, continuous lines).

Expectations from the perceptual momenthypothesis are uniformly disconfirmed. Teststimuli were always shown embedded withinthe masking stimuli or, to say it with Turvey(1978), the two configurations always borea "post-during" relationship one to the

TEMPORAL INTEGRATION 89

other. Under these display conditions thetwo stimuli would invariably coincide withinthe same moment and should thus be max-imally integrated. These would be just theconditions in which impairment in identifica-tion of the test stimulus should be mostsevere. Yet, through most of the domain(Figure 9), performance was virtuallyunimpeded. This result as well as the resultsof the preceding four studies are utterlyunexplainable in terms of perceptual mo-ments. Additional evidence strongly refut-ing the perceptual moment hypothesis hasbeen reported by Di Lollo and Wilson(1978).

More generally, the difficulty encounteredby the perceptual moment hypothesis is thatit regards temporal integration as principallydependent upon the interstimulus interval(ISI), whereas the data reported in the

foregoing experiments clearly implicate stim-ulus-onset asynchrony (SOA). (The ISIis the "blank" temporal interval elapsingbetween the termination of a leading stim-ulus and the onset of the next stimulus;SOA refers to elapsed time between theonsets of two successive stimuli irrespectiveof stimulus duration.) Given that temporalintegration follows an SOA—rather than anISI—law, no extant version of the percep-tual moment hypothesis can remain tenable.

General Discussion

Throughout this article the main theo-retical rationale for the experimental workwas provided by a juxtaposition of twotheories of sensory persistence: a simplestorage theory and a processing theory. Dis-cussion of other plausible explanations of

100 r

90

1°IM80

70

60

50

40

Sg 30

\20

10

A A JEFF

O O BCN

• • DBN

O

0 2040 80 160 320

STIMULUS-ONSET ASYNCHRONY (MSEC)

640

Figure 9. Percent correct responses plotted separately for each subject at each of the sevenlevels of stimulus-onset asynchrony in the experimental (continuous lines) and in the control(segmented lines) conditions. (In the experimental condition, the two sets of masking contours[Ml and M2] were identical; in the control condition, they were different, T = test stimulus.)

90 VINCENT Di LOLLO

the results has been postponed in order tohighlight the juxtaposition and to expeditereporting of the data.

In a nutshell, the finding to be explainedis the immediate loss of perceptual availabil-ity upon termination of a display that hasbeen on view for longer than about 100msec. Besides the hypotheses already dis-cussed, two alternative explanations suggestthemselves: receptor adaptation and per-ceptual suppression.

Receptor Adaptation

Explanation of the present results interms of receptor adaptation is based on theevidence that a receptor's level of respondingdiminishes as the duration of the inducingstimulus is increased. On this basis, theinverse relationship between duration ofleading display and degree of perceptualintegration with the trailing display isexplained by assuming that the responsive-ness of the receptors exposed to the leadingstimulus had waned by the time the secondstimulus was displayed.

But this interpretation is severely weak-ened by the fact that receptor adaptationtypically requires levels of stimulation eitherof far greater intensity or of far greaterduration than the stimuli used in these stud-ies (e.g., Brown, 1965). That is to say, ifreceptor adaptation played a role in thepresent work, that role must be regarded asquite limited because the phenomenaappeared conspicuously, despite the lowlevels of stimulus intensity, even when theduration of the first display was only 100msec.

Incidentally, it should be noted thatreceptor aftereffects provide a somewhatambiguous explanatory principle, in thatcontinued activity could be expected just asplausibly as underactivity upon terminationof the inducing stimulus. Consider the briefyet clear positive afterimages produced byprolonged or intense stimulation of photo-receptors. Such aftereffect would obviouslyhave a facilitatory rather than a hinderingeffect on temporal integration of successivedisplays. At any rate, bearing in mind thebrief durations and the relatively low

intensities of the stimuli, neither type ofaftereffect is likely to have played a majorrole in the present experiment.

Let us now consider the option that failureof perceptual integration at the longerdurations of the leading stimulus may beascribed to an active process of perceptualsuppression induced by the presentation ofthe second stimulus. Three classes of eventswhere some form of perceptual suppressionhas been implicated are examined below.They are apparent motion, metacontrastmasking, and inhibitory channel interactions.

Apparent Motion and Metacontrast

Suppressive effects of apparent motionand of metacontrast masking appear tofollow similar time courses and, largely forthis reason, have been linked theoreticallyone to the other (Fehrer, 1966; Kahneman,1968). Although, as noted above, there wasvery little phenomenal evidence of apparentmotion in the visual displays, it could beargued that some perceptual suppressionmight nevertheless have taken place. How-ever, the temporal parameters of apparentmotion do not seem to fit well with thepresent data. Optimal apparent motion isobtained within SOA ranges of 50-100 msec(Kahneman, 1968) with the effect lesseningthereafter. Now consider the results ofExperiments 1 and 2, where deterioration inperformance did not even begin until SOAsof at least 100 msec and showed no sign ofabating at much longer SOA values. Clearly,if these results are to be ascribed to thesuppressive effect of apparent motion, thenit must also be explained why the suppres-sive effects were weakest at those SOAswhere apparent motion would be expectedto be strongest, and vice versa.

Similar considerations appear to conflictwith an interpretation of these results interms of metacontrast masking. Although, inkeeping with the present data, severity ofmetacontrast masking is known to be a func-tion of SOA rather than ISI, the most severemasking is obtained at SOAs of 50-100msec (Kahneman, 1968). Thus, were theresults of Experiments 1 and 2 to beexplained in terms of metacontrast masking,

TEMPORAL INTEGRATION 91

it should be expected that performance wouldbe the worst when the second portion of thedisplay was presented 50-100 msec after theonset of the first portion. But the data (Fig-ures 1, 2, and 4) offer a strikingly differentpicture.

Additional considerations make a meta-contrast explanation even less likely. It hasbeen known for some time (Alpern, 1953)that metacontrast suppression is weak orentirely absent for stimuli displayed withinthe fovea. This would create difficulties fora metacontrast interpretation of Experi-ments 4 and 5, in which the stimuli wereentirely foveal. Furthermore, informal repli-cations of Experiment 1 using a dot matrixno larger than 2° of visual angle yieldedresults entirely comparable to those obtainedwith the larger stimuli. Finally, even thoughboth the stimuli and the tasks employed inthe present studies are unlike those used inthe most common class of metacontrastexperiments (cf. Weisstein, 1972), it isnotable that the curves obtained in thepresent work are "Type A" monotonic func-tions (Breitmeyer & Ganz, 1976; Kolers,1962), whereas metacontrast curves areinvariably "Type B" nonmonotonic func-tions.

Inhibitory Channel Interactions

Information-processing theories have beendeveloped in which the visual system isregarded as a multichannel processor (Breit-meyer & Ganz, 1976; Merikle, 1977; Weis-stein, Ozog, and Szoc, 1975). These theoriescan explain instances of perceptual suppres-sion in terms of patterns of interferencearising within and between processingchannels.

It is known that (e.g., Breitmeyer & Ganz,1976), as part of the total neural responseinitiated by a visual display, at least twoseparate neural networks are activated withinthe visual system: the transient and thesustained networks. Transient cells havevery brief latencies and are attuned to lowspatial frequencies, which mediate detectionof a visual event but are not attuned to thehigh spatial frequencies that mediate percep-tion of the contours of a display. In contrast,

sustained cells have longer latencies (up toseveral hundred milliseconds) and areattuned to higher spatial frequencies. It isalso known that a burst of activity in atransient channel inhibits activity in aneighboring sustained channel (Singer &Bedworth, 1973). Such interchannel differ-ences in response latencies have been utilizedto explain perceptual suppression of the firstof two rapidly successive displays (Breit-meyer & Ganz, 1976).

For example, consider two patternedvisual stimuli presented in rapid succession.The temporally leading stimulus (StimulusA) is shown for, say, 150 msec and is fol-lowed immediately by the second stimulus(Stimulus B) which is shown for, say, 20msec. According to theory, the onset ofStimulus A generates an immediate andshort-lived burst of activity in the transientchannels that subsides within less than 100msec from the onset of the stimulus. Underthese circumstances, transient-channel activ-ity would have no inhibitory consequencesbecause no sustained channels would beactive at the same time (i.e., the field of viewhad been blank just before the onset ofStimulus A). About 100 msec after theonset of Stimulus A, activity is initiated inthe sustained channels in response to thehigher spatial frequencies (i.e., contours oredges) of the display. Undisturbed, activityin the sustained channels would continueand would ultimately permit identificationof the contours of the stimulus, but the on-set of Stimulus B generates a burst oftransient-channel activity that inhibits on-going activity in the sustained channelsconcerned with Stimulus A and thus impairsperception of its contours. As a consequenceof such inhibitory interchannel interaction,at the end of the display sequence, the visualsystem has available the contours of onlyStimulus B.

Inhibitory interactions of this kind wouldnot arise were the onsets of A and B simul-taneous or within a few milliseconds fromeach other. In this case, all transient-channelactivity would have subsided before thebeginning of activity in the sustained chan-nels ; and, according to theory, most within-

92 VINCENT Di LOLLO

channel interactions are integrative ratherthan inhibitory.

The parallel between this example and theexperiments reported here need hardly beelaborated: If two visual displays in thesestudies started within less than about 100msec from each other, they tended to beperceived in the form of a composite stimulusas a result of integrative within-channelinteractions; if, however, the SOA betweenthem exceeded about 100 msec, perceptionof the leading contours was terminatedinstantly as a result of inhibitory between-channel interactions.

Even Experiment 2 (Figures 3, 4, and 5)can easily be explained in terms of inter-channel inhibitory effects. In the control(nonrotated) condition, performance washighly accurate at brief levels of SOA butbegan to deteriorate at levels of SOA greaterthan about 100 msec. This would be expectedon the basis of intrachannel integrativeeffects at the briefer SOAs and interchannelinhibitory effects at longer SOAs, as ex-plained above. In the experimental (rotated)condition, performance was never impairedbecause, given their temporal relationship(Figure 3), the micropatterns shown duringthe last 10 msec of the leading display alwaysbore an intrachannel integrative relationshipand never an interchannel inhibitory rela-tionship with the micropatterns of thetrailing display.

Interpretation of the results of Experi-ment 3 (Figure 6) would follow much thesame line of reasoning. Transient-channelactivity induced by any given dot wouldinhibit sustained-channel activity induced bydots plotted about 100 msec earlier. Amoment's reflection will show that underthese circumstances, only the dots plottedduring the last 100 msec would remainperceptually available at the end of thedisplay sequence. And, since interchannelinhibition follows SOA (not ISI) rules, the"long" dot in the experimental conditionwould be subject to the same level ofinhibition by later plotted dots as a "brief"dot plotted simultaneously with its onset.

Clearly, the theory of inhibitory channelinteractions can provide a satisfactory

account of the more prominent aspects ofthe present results. Yet, serious difficultiesare encountered with other less obviousfeatures. For example, it has been known forsome time that metacontrast suppressioneffects, although robust in the periphery ofthe retina, are weak and highly unstable atthe fovea (Alpern, 1953; Kolers & Rosner,1960). From the viewpoint of inhibitorychannel interactions, this result is to beexpected because the concentration of tran-sient channels in the fovea is extremely low,whereas that of sustained channels is veryhigh (Fukuda & Stone, 1974; Hoffman,Stone, & Sherman, 1972). Such imbalancein relative concentrations ensures that inter-channel inhibitory effects would be negligibleat the fovea but pronounced at the peripherywhere the opposite imbalance in relativeconcentrations is found.

Paucity of foveal transient channels is anatural empirical basis on which to explainthe fragility of interchannel inhibitoryeffects at the fovea. Now, consider the resultsof Experiments 4 and 5, bearing in mindthat the stimuli were displayed entirelywithin the fovea. It is obvious that thepowerful effects obtained in these experi-ments (Figures 7 and 9) cannot be ex-plained in terms of interchannel inhibitionwithout marked theoretical inconsistency. Infact, it seems reasonable that the thinlyscattered foveal transient channels shouldinduce only weak interchannel inhibitoryeffects, such as are found in foveal meta-contrast—not the powerful effects thatwould have to be hypothesized to accountfor the present data. This is not to deny theimportance of interchannel inhibitory effects,particularly in the processing of parafovealstimuli, but the effects obtained in Experi-ments 4 and 5 are quantitatively beyondwhat could be fully interpreted on the basisof inhibitory channel interactions.

A second comparison not easily handledby multichannel theory is between Experi-ments 1 and 2 on the one hand and Experi-ments 4 and 5 on the other. As can be seenin Figures 7 and 9, recovery from maskingis evident at SOAs as brief as 20 msec andis well developed at 40 msec. In contrast,

TEMPORAL INTEGRATION 93

decisive evidence of impairment in thematrix task did not develop until SOAs ofabout 100 msec or beyond (Figures 1 and4). If both sets of results are to be ascribedto perceptual suppression arising frominhibitory channel interactions, then anexplanation should also be given of the pro-nounced discrepancy in the time courses ofthe two sets of events.

It could be argued that the effects in Ex-periments 4 and 5 might have been enhancedby the close spatial proximity between thecontours of the leading and of the trailingdisplays. But this argument would ignorethe scarcity of transient channels in thefovea. Indeed, contrary to the findings, anexplanation based on interchannel inhibi-tion could well favor more vigorous effectsin Experiments 1 and 2 on the grounds that,unlike the alphabetic stimuli in Experiments4 and 5, the matrix configuration extendedto parafoveal areas where the concentra-tion of transient channels is known to be fargreater than in the fovea. At any rate, thesearguments speak principally to the relativemagnitudes of the effects, not to their rela-tive time courses; multichannel theory hasno ready way of handling the latter.

In summary, the present argument is notthat the theory of inhibitory channel inter-actions is disconfirmed by the present data,nor even that it was totally irrelevant to theoutcome of the experiments. Rather, thesuggestion is offered that the effects re-ported here are, in varying degrees, outsidethe class of events that the theory purportsto encompass. It is suggested instead thata more coherent account of this pattern ofresults can be given by regarding sensorypersistence as an outcome of neural activityat an early phase of visual information pro-cessing. This viewpoint was outlined earlierin this article and is elaborated below.

A Sensory-Coding Approach

Perceptual events whose principal tem-poral correlate is SOA rather than ISI havebeen explained in terms of information-pro-cessing events that take place during theSOA (e.g., Kahneman, 1967; Sperling,

1971). In broad outline, the same approachis taken in this article with respect to theduration of visual persistence.

Assume that perception of form emergesfrom a number of processing operations—some performed serially, others in parallel—and that such operations can be broadlycategorized in two successive phases of sen-sory coding, recruiting and interpretingphases. In the first phase (recruiting), thestimulus configuration becomes encoded interms of as yet meaningless features such asdots, bars, edges, and discontinuities. The"feature-encoded" stimuli that emerge fromthe recruiting phase are available to thesecond (interpreting) phase of sensory cod-ing in which identification and categorizationoperations are performed. The "meaning-encoded" items that emerge from the inter-preting phase may then be available formore complex processes of organization andfor more permanent forms of memory. Theprincipal function of each phase is to producea new level of sensory coding of the initialdisplay. Ultimately, the purpose of the pro-cessing sequence is to arrive at a level ofcoding that permits comparison betweenthe newly encoded stimulus and other long-term memories.

Although it is assumed that the two phasesoccur sequentially in the flow of informationprocessing, the possibility is not discountedthat processing of some stimulus dimen-sions (e.g., brightness, color, contour) maybe completed before others (cf. Cheatham,1952; Fehrer & Raab, 1962; Kahneman,1967). Nor is the possibility discounted offeedback loops between stages.

Some of the distinguishing characteristicsof the recruiting phase are held to be (a)faithful maintenance of stimulus geometrywith initial retinal configuration; (b) virtu-ally unlimited processing capability; (c)mainly parallel processing; (d) maskingthrough integration of contours (as distinctfrom interruption of processing; cf. Scheerer,1973) ; and (e) sensory persistence (namelythe maintenance of physical, as distinct frominformational, aspects of the display). Thecorresponding characteristics of the inter-preting phase are held to be (a) mainly

94 VINCENT Di LOLLO

serial processing; (b) little if any corre-spondence between stimulus geometry andinitial retinal configuration; (c) maskingby interruption of processing (cf. Scheerer,1973) ; and (d) maintenance of informationalrather than physical aspects of the display(i.e., absence of visual persistence). Theinterpreting phase may be regarded asperforming some of the functions of the lim-ited-capacity "central processor" hypothe-sized by Posner and Klein (1973). Exam-ples of events taking place at the level of theinterpreting phase are the instances of visualmasking without spatial proximity of testand masking stimuli (Di Lollo, Lowe, &Scott, 1974) and the incidence of backwardmasking of ISIs exceeding about 100-150msec (Scheerer, 1973; Spencer & Shuntich,1970).

Conceptually, this schema is clearly re-lated to other multistage formulations (e.g.,Atkinson & Shiffrin, 1968; Dick, 1974;Haber, 1971; Neisser, 1967; Scheerer, 1973;Turvey, 1973, 1978). There are, however,two distinctive aspects. First, the suggestedcorrelative classification between levels ofsensory coding (recruiting and interpretingphases) and types of visual masking (inte-gration or interruption) provides a rationalefor the reconciliation of the two theories ofmasking and for subsuming processing andmasking effects within the same class ofevents. And second, visual persistence is re-garded not as the content of a rapidly fadingstore but as the product of the activity ofsensory coding mechanisms engaged in theformation of "feature-encoded" stimuli.

Defining visual persistence in terms ofactivity at an early phase of sensory process-ing raises a problem with respect to stimuliwhose duration exceeds the duration of theprocessing phase. To be precise, it has beensuggested in this article that, assuming arecruiting phase of 100 msec, a 20-msecdisplay would have visual persistence of 80msec, whereas a 130-msec display wouldhave no persistence at all. However, in thecase of the 130-msec display, stimulationwould keep on entering the visual systemup to 30 msec beyond the presumed termina-tion of the recruiting phase. Should this ad-

ditional stimulation be expected to initiate anew processing cycle and thus generate anew period of persistence lasting about 70msec? The data clearly deny this prospect.Were this to be so, the relationship betweenduration of inducing stimulus and durationof visual persistence would be a recurringsawtooth function with period correspondingto the duration of the recruiting phase. Theresults of the present studies offer no hintof such a relationship. It must thereforeprovisionally be concluded that, once pro-cessed, a given display does not keep on be-ing reprocessed if it remains on view be-yond the hypothesized duration of the re-cruiting phase.

This is tantamount to saying that thevisual system can distinguish "new" from"old" contours (i.e., contours that have justentered the recruiting phase from contoursthat have been "feature-encoded" but thatare still externally on display). Separationbetween new and old contours could beachieved in a variety of ways. An appealingpossibility is some form of autocorrelationperformed on incoming stimulation. In es-sence, this would correspond to setting upan input filtering mechanism such that in-coming features that do not match those ofthe filter (i.e., new contours) would be senton for processing at the recruiting level; allother features would bypass the recruitingphase and would serve as a signal to higherlevels in the visual system that nothing haschanged.

Neither autocorrelation nor filtering isa new notion in visual information process-ing. Dodwell (1971) and Uttal (1975) havesuggested autocorrelation as the processingbasis for the emergence of contour. VonHoist's (1954) description of a perceptualfilter was as follows:

The efference (to the effector) leaves an "image"of itself somewhere in the CNS to which the re-afference . . . compares as the negative of a photo-graph compares to its print; so that . . . when theefference copy and the reafference exactly com-pensate one another, nothing further happens. When,however, the efference is too small or lacking [or]too great . . . the difference . . . can ascend to ahigher centre and produce a perception, (p. 91)

TEMPORAL INTEGRATION 95

Given some such filtering mechanism,"old" contours would have no persistencebecause of absence of corresponding activityat the recruiting level; at the same time, thecontours would remain phenomenally visiblethrough stimulation reaching higher centersdirectly from the filter. It may be notedthat a plausible physiological mechanismthat could underlie this class of information-processing events has recently been pro-posed by Singer (1977).

Concluding Remarks

Throughout the studies reported here, per-ceptual integration of successive visual dis-plays was found to be a function of SOA, asdistinct from ISI. In the discussion of Ex-periment 5, it was stressed that this patternof results is entirely inconsistent with pre-dictions from perceptual moment theorywhere ISI, rather than SOA, is regarded asthe basis for temporal integration.

Essentially the same difficulty is encoun-tered by a "storage" theory of iconic per-sistence. The icon has been likened to areservoir that is charged instantly and be-gins to discharge as soon as the chargingagent (i.e., the inducing stimulus) is turnedoff. However, the analogy breaks down whenit is shown that the state of charge of thestore (i.e., the availability of sensory per-sistence) cannot be predicted from a knowl-edge of the time elapsed since the removal ofthe charging agent without knowledge of theexact duration of the charging process; thenthe relationship is just the opposite of whatshould be expected between duration ofcharging agent and degree of saturation ofthe recipient store. It must be concludedthat a simple "storage" theory of visual per-sistence is unambiguously contradicted andshould be regarded as a wholly inadequateexplanatory model. The same conclusion hasbeen reached independently in other investi-gations (e.g., Di Lollo, 1978; Holding, 1975;Meyer, 1977; Meyer, Lawson, & Cohen,1975).

A final point needs to be added. The pres-ent viewpoint that persistence is a form ofvisual memory associated with a particular

phase of processing need not be limited tothe class of phenomena examined here. Otherforms of short-term memory could also beregarded as outcomes of activity at otherphases of processing. In a general sense, asthe coding of the initial stimulus proceedsfrom energy transduction at the retina tothe emergence of meaning and the ramifica-tion of associations at higher centers, con-comitant short-lived representations wouldensue, each stemming from, and coded interms of, the prevalent processing activitytaking place during a given phase. In thissense, the terms memory production andmemory representation would be regarded asvirtually synonymous (cf. Mandler, 1975).

To illustrate, let us pursue the distinctionmade earlier between recruiting and inter-preting phases of processing. The formerwas said to be concerned with the extractionof meaningless features from the inducingstimulus, the latter with the emergence ofstructure and with more symbolic aspectsof the display. The type of sensory per-sistence arising from these two phases shouldbe expected to reflect the differences in pro-cessing activities taking place within them.Notably, activity in the interpreting phasemay give rise to a form of persistence which,while arising from visual stimulation and stillmaintaining structural information about thedisplay, need no longer be visible. Following aclose examination of the literature on iconicmemory, Turvey (1978) found that justthis kind of nonvisible persistence wasstrongly suggested by the experimental evi-dence. He named it "schematic persistence"and distinguished it from visible persistencealong several dimensions.

Patently, this conceptual framework im-plies that sensory persistence is not a uni-tary phenomenon labeled "iconic memory."Rather, it suggests that there are as manyexpressions of sensory persistence as thereare discernible phases of information pro-cessing. In this respect, the present approachto the problem of sensory persistence sharesbroad yet explicit similarities with Craikand Tulving's (1975) approach to the moreextended problems of memory and recall.

96 VINCENT Di LOLLO

References

Allport, D. A. Phenomenal simultaneity and theperceptual moment hypothesis. British Journalof Psychology, 1968, 59, 39S-406.

Alpern, M. Metacontrast. Journal of the OpticalSociety of America, 1953, 43, 648-657.

Atkinson, R. C, & Shiffrin, R. M. Human mem-ory : A proposed system and its control pro-cesses. In K. W. Spence & J. T. Spence (Eds.),Psychology of learning and motivation: Advancesin research and theory (Vol. 2). New York:Academic Press, 1968.

Breitmeyer, B. G., & Ganz, L. Implications ofsustained and transient channels for theoriesof visual pattern masking, saccadic suppression,and information processing. Psychological Re-view, 1976, 83, 1-36.

Brown, J. L. Afterimages. In C. H. Graham (Ed.),Vision and visual perception. New York: Wiley,1965.

Butler, T. W. Luminance-duration relationshipsin the photopic ERG and the apparent bright-ness of flashes. Vision Research, 1975, IS, 693-698.

Cheatham, P. G. Visual perceptual latency as afunction of stimulus brightness and contourshape. Journal of Experimental Psychology, 1952,43, 369-380.

Craik, F. I. M., & Tulving, E. Depth of process-ing and the retention of words in episodic mem-ory. Journal of Experimental Psychology: Gen-eral, 1975,104, 267-294.

Dick, A. O. Iconic memory and its relation toperceptual processing and other memory mecha-nisms. Perception & Psychophysics, 1974, 16,575-596.

Di Lollo, V. Temporal characteristics of iconicmemory. Nature, 1977, 267, 241-243.

Di Lollo, V. On the spatio-temporal interactionsof brief visual displays. In R. H. Day & G. V.Stanley (Eds.), Studies in perception. Perth,Australia: University of Western AustraliaPress, 1978.

Di Lollo, V. Luminous calibration of oscilloscopicdisplays. Behavior Research Methods and In-strumentation, 1979, 11, 419-421.

Di Lollo, V., Lowe, D. G., & Scott, J. P., Jr.Backward masking and interference with theprocessing of brief visual displays. Journal ofExperimental Psychology, 1974, 103, 934-940.

Di Lollo, V., & Wilson, A. E. Iconic persistenceand perceptual moment as determinants of tem-poral integration in vision. Vision Research,1978,18, 1607-1610.

Dodwell, P. C. On perceptual clarity. PsychologicalReview, 1971, 78, 275-289.

Efron, R. The relationship between the durationof a stimulus and the duration of a perception.Neuropsychologia, 1970, 8, 37-55.

Efron, R. An invariant characteristic of perceptualsystems in the time domain. In S. Kornblum

(Ed.), Attention and performance IV. NewYork: Academic Press, 1973.

Eriksen, C. W. Temporal luminance summationeffects in backward and forward masking. Per-ception & Psychophysics, 1966, 1, 87-92.

Eriksen, C. W., & Collins, J. F. Some temporalcharacteristics of visual pattern perception.Journal of Experimental Psychology, 1967, 74,476-484.

Eriksen, C. W., & Collins, J. F. Sensory tracesversus the psychological moment in the temporalorganization of form. Journal of ExperimentalPsychology, 1968, 77, 376-382.

Fehrer, E. Effect of stimulus similarity on retro-active masking. Journal of Experimental Psy-chology, 1966, 71, 612-615.

Fehrer, R., & Raab, D. Reaction time to stimulimasked by metacontrast. Journal of ExperimentalPsychology, 1962, 63, 143-147.

Fukuda, Y., & Stone, J. Retinal distribution andcentral projections of Y-, X-, and W-cells ofthe cat's retina. Journal of N euro physiology,1974, 37, 749-772.

Granit, R. Sensory mechanisms of the retina. NewYork: Oxford University Press, 1947.

Haber, R. N. Where are the visions in visual per-ception? In S. J. Segal (Ed.), The adaptativefunction of imagery. New York: Academic Press,1971.

Haber, R. N., & Hershenson, M. The psychologyof visual perception. New York: Holt, 1973.

Hoffman, K. P., Stone, J., & Sherman, S. M. Relayof receptive-field properties in dorsal lateralgeniculate nucleus of the cat. Journal of Neuro-physiology, 1972, 35, 518-531.

Hogben, J. H., & Di Lollo, V. Perceptual inte-gration and perceptual segregation of brief visualstimuli. Vision Research, 1974, 14, 1059-1069.

Holding, D. H. Sensory storage reconsidered.Memory & Cognition, 1975, 3, 31-41.

Hoist, E. von. Relations between the centralnervous system and the peripheral organs.Animal Behaviour, 1954, 2, 89-94.

Kahneman, D. An onset-onset law for one case ofapparent motion and metacontrast. Perception &Psychophysics, 1967, 2, 577-584.

Kahneman, D. Method, findings, and theory instudies of visual masking. Psychological Bulletin,1968, 70, 404-425.

Kolers, P. A. Intensity and contour effects invisual masking. Vision Research, 1962, 2, 277-294.

Kolers, P. A., & Rosner, B. S. On visual masking(metacontrast) : Dichoptic observations. Amer-ican Journal of Psychology, 1960, 73, 2-21.

Kristofferson, A. B. Attention and psychophysicaltime. Acta Psychologica, 1967, 27, 93-100.

Handler, G. Consciousness: Respectable, useful,and probably necessary. In R. L. Solso (Ed.),Information processing and cognition. Hillsdale,N.J.: Erlbaum, 1975.

Merikle, P. M. On the nature of metacontrast withcomplex targets and masks. Journal of Experi-

TEMPORAL INTEGRATION 97

mental Psychology: Human Perception andPerformance, 1977, 3, 607-621.

Meyer, G. E. The effects of color-specific adapta-tion on the perceived duration of gratings.Vision Research, 1977, 17, 51-56.

Meyer, G. E., Lawson, R., & Cohen, W. Theeffects of orientation-specific adaptation on theduration of short-term visual storage. VisionResearch, 1975, 15, 569-572.

Murphree, O. D. Maximum rates of form percep-tion and the alpha rhythm: An investigation andtest of current nerve net theory. Journal ofExerimental Psychology, 1954, 48, 57-61.

Neisser, U. Cognitive psychology. New York:Appleton-Century-Crofts, 1967.

Posner, M. L, & Klein, R. On the function ofconsciousness. In S. K. Kornblum (Ed.),A ttention and performance V, Amsterdam:North Holland, 1973.

Sakitt, B. Iconic memory. Psychological Review,1976, 83, 257-276.

Scheerer, E. Integration, interruption and process-ing rate in visual backward masking. Psycho -logische Forschung, 1973, 36, 71-93.

Serviere, ]., Miceli, D., & Galifret, Y. A psycho-physical study of the visual perception of"instantaneous" and "durable." Vision Research,1977,17, 57-63.

Singer, W. Temporal aspects of subcortical con-trast processing. Neurosciences Research Prog-ress Bulletin, 1977, IS, 358-369.

Singer, W., & Bedworth, N. Inhibitory inter-actions between X and Y units in the cat lateralgeniculate nucleus. Brain Research, 1973, 49,291-307.

Spencer, T. ]., & Shuntich, R. Evidence for aninterruption theory of backward masking. Jour-nal of Experimental Psychology, 1970, 85, 198-203.

Sperling, G. The information available in briefvisual presentations. Psychological Monographs,1960, 74(11, Whole No. 498).

Sperling, G. Information retrieval from two rapidlyconsecutive stimuli: New analysis. Perception &Psychophysics, 1971, 9, 89-91.

Turvey, M. T. On peripheral and central processesin vision: Inferences from an information-pro-cessing analysis of masking with patternedstimuli. Psychological Review, 1973, 80, 1-52.

Turvey, M. T. Visual processing and short-termmemory. In W. K, Estes (Ed.), Handbook oflearning and cognitive processes (Vol. 5). Hills-dale, N.J.: Erlbaum, 1978.

Uttal, W. P. An autocorrelation theory of formdetection. Hillsdale, N.J.: Erlbaum, 1975.

Walter, W. G. The function of the electricalrhythms of the brain. Journal of Mental Science,1950, 96, 1-31.

Weisstein, N. Metacontrast. In D. Jameson & L.M, Hurvich (Eds.), Handbook of sensory physi-ology (Vol. 7), Part IV: Visual psychophysics.New York: Springer, 1972.

Weisstein, N., Ozog, G., & Szoc, R. A comparisonand elaboration of two models of metacontrast.Psychological Review, 1975, 82, 325-343.

White, C. T. Temporal numerosity and the psy-chological unit of duration. Psychological Mono-graphs, 1963, Whole No. 575.

Received August 23, 1978 •