Journal of Experimental Psychology:Human Perception and Performance1985, Vol. I I , No. 3, 317-328

Copyright 1985 by the American Psychological Association, Inc.0096-1523/85/$00.75

Can Mental Images Be Ambiguous?

Deborah Chambers and Daniel ReisbergNew School for Social Research

Although much recent research has emphasized the equivalence between imageryand perception, there are critical differences between these activities: Perception,

initiated by an external stimulus, is to a large extent concerned with theinterpretation of that stimulus. Images, in contrast, are created as symbols of

something and hence need no interpretive process. Without a construal process,images do not allow reconstrual. In support of this argument, we ask whethersubjects can reverse an ambiguous figure in mental imagery. In three experiments,

no subject was ever able to reverse a mental image. In contrast, all of the subjectswere able, immediately after this failure, to draw a picture from their mentalimage and then reconstrue the figure in their own drawing. This failure to reverse

images occurs despite hints to the subject, some coaching, and a moderate amountof training in figural reversal.

A great deal of recent research has focusedon the "equivalence" of mental imagery andof perception. In addition to the obviousphenomenal similarities between imaging andperceiving, these activities share a number offunctional properties. When asked to scanacross an image or rotate one or zoom in onone (to inspect detail) or zoom back fromone (to make gross comparisons), subjects'response times are all as one would expect ifsubjects were inspecting an image that wasin many ways picturelike, and, in particular,that preserved the metric properties of space.(For reviews of this large literature, see Kos-slyn, 1980; Shepard & Cooper, 1983.)

In the same vein, the aftereffects of inspect-ing images seem comparable to those ofinspecting visually present scenes. Finke andSchmidt (1977) obtained color aftereffectsfrom inspection of images; Finke (1979) re-ported motor aftereffects from imaged "prism

This study emerged from discussions in a seminar onmental imagery that was made possible by a grant fromthe Exxon Foundation for the support of jointly taughtgraduate courses.

We want to thank the participants in that seminar,especially Jerome Bruner and Veronique Foti. FriderikeHeuer, David Smith, Robert Melara, Carol Staszewski,and Peter Monk all made helpful comments on earlierdrafts. We also thank Lynn Winters for her help inrunning Experiment 3, and Kriss Grisham for photo-graphing the stimuli.

Requests for reprints should be sent to Daniel Reisberg,Department of Psychology, New School for Social Re-search, 65 Fifth Avenue, New York, New York 10003.

adaptation." If we attribute these effects tophenomena in the visual system (such assome sort of recalibration or fatigue), thesedemonstrations appear to be a strong argu-ment that visual imagery draws on the samemental processes and structures as does visualperception.

This claim of shared mental resources isfurther strengthened by findings of interfer-ence between perceptual activity and imagi-nation. In an extension of Perky's work(1910), Segal and Fusella (1970) showed thatimaging interferes with the detection of near-threshold stimuli. Likewise, Brooks (1968)demonstrated that imaging interferes withvisually guided responses but leaves verbalresponses unimpaired, indicating that the actof imaging and the demands of visual percep-tion draw on common mental structures.

Despite these (and other) demonstrationsof commonality between imaging and per-ceiving, there is good reason to set boundson this apparent functional equivalence. ThePerky effect notwithstanding, we have in gen-eral little trouble distinguishing images andpercepts phenomenally, a fact that suggeststhat there must be important differences be-tween these activities. (See Kolers, 1983;Neisser, 1978; and, for an extensive phenom-enological discussion of imagination, Casey,1976.) In addition, visual images do not seemuniformly to respect laws of optical occlusion,so that "hidden" objects in an image seemto be part of the image. (For evidence that

317

318 DEBORAH CHAMBERS AND DANIEL REISBERG

images can behave in this fashion, see Kerr& Neisser, 1983; Neisser & Kerr, 1973. Forevidence that images do not always behavein this way, see Keenan, 1983; Keenan &Moore, 1979.) Finally, images are sometimesindeterminate on properties that cannot beunspecified in a visual stimulus, such as theexact number of stripes on an imaged stripedtiger. (See Dennett, 198la, 1981b.)

One important source of these differencesbetween images and percepts lies in the wayeach of these comes into being. Perception,initiated by stimulation from an externalobject, is largely concerned with the interpre-tation of that object. Percepts are thus shapedand constrained by a stimulus; the perceivermight decide to reinterpret some ambiguousaspect of the stimulus but cannot will thestimulus to be other than it is.

Images, in contrast, are constructed as animage of some particular thing or scene. Wedo not mean to suggest that images arealways deliberately constructed; obviouslyimages can be evoked by external causes orcan arise unbidden. The claim is that, whetheror not an image was intentional in the ordi-nary sense of the term, it is intentional inthe sense described by Brentano (1874/1973,p. 88): "Every mental phenomenon is char-acterized . . . by what we might call . . .reference to a content, direction toward anobject. . . ." If a loud noise, for example,summons forth an image of someone falling,that image comes into being as an image ofsomeone falling, not (to borrow an argumentfrom Fodor, 1981) as an image of someonediving forward to grab something on thefloor. Pictures of these two scenes might beindistinguishable, but the image is, for theimager, unmistakably one and not the other.

Kolers (1983) makes essentially the samepoint from a different perspective. Despitethe misleading tendency in the literature tocontrast images and symbols, images aresymbolic. As such, images refer, not by someresemblance relation (which might be ambig-uous), but by the conventions of the creatorof the symbol, namely the imaginer. There-fore, the interpretation of the image is essen-tially transparent to the imaginer, who createdthe image to symbolize some particular thing.

These kinds of considerations led Fodor(1981) to argue that mental images are "im-

ages under description." There is no issue of"reading" or "interpreting" an image. Theimage is created as a symbol of some partic-ular thing, and so the interpretation is thereat the outset. As one consequence of thisview, Fodor notes, one's image need notclosely resemble what is being imagined:One's image of a tiger does not need to lookmuch like a tiger, by the fact that the imagerunderstands it as a tiger image, it is one. Asanother consequence, images, unlike thestimuli that give rise to perception, cannotbe reinterpreted. Images can certainly bewilled to be different, or can be replaced, butwithout a construal process, there is no pos-sibility for reconstrual. Gasey (1976), workingwithin a very different philosophical tradition,shares this conclusion with Fodor: One regardin which images are not picturelike is thatthere is no such thing as an ambiguousimage. "To imagine something differently isto imagine something different" (Casey, 1976,p. 159).

However, some observations suggest thatthese views may be overdrawn. First, thesearguments seem odd in the face of imageryreports of fluid, changing images (includingCasey's own in his phenomenological study).Second, in many of the mental rotation pro-cedures, one must rotate an imagined stimulusin order to decide whether that stimulus is aletter or a mirror reversal of a letter (Shepard& Cooper, 1983). In these studies, one seemsfirst to have the image and only later todiscover how to construe that image, contraryto Fodor or to Casey. Slee (1980) also pre-sented evidence that suggests subjects canreinterpret mental images, or, more precisely,can interpret images in a way that differsfrom that intended when the image was con-structed. Employing a procedure developedby Reed (1974; also Reed & Johnson, 1975),Slee briefly presented subjects with a patternthat was perceived as two adjacent Romannumeral 10s, enclosing a diamond-shapedspace. In a later cued-recall test, subjects withvivid mental imagery (assessed by Slee's Vi-sual Elaboration Scale, or VES) were able, toreconstruct the figure around either one ofthe Roman numeral 10s or a parallelogram(which was embedded, but not perceived, inthe original figure). Reed's (1974) and Reedand Johnson's (1975) subjects were also able

AMBIGUOUS IMAGES? 319

to do this and comparable tasks. Althougherror rates were quite high in their studies,their subjects were above chance performancein detecting the parts of imaged stimuli.(Judging from Slee's data, the high error ratespresumably reflect the inclusion in these ear-lier procedures of subjects with low imageryvividness.) Apparently, then, subjects (withthe aid of a cue) are able to reorganize theirmemory image of the initial stimulus. Finally,we suspect that many readers of this articlecan easily reverse a mental image of a Neckercube or another ambiguous figure, a likelihoodthat again suggests the possibility of recon-struing images. In all of these cases, contraryto Casey or Fodor, images appear to sharewith percepts a construal process and, inaddition, a possibility for reconstrual.

There are a number of ways one mightreconcile these findings and the claim thatimages lack a construal process (and hencethe possibility of reconstrual). In most ofthese cases, the "new" construal of an imageinvolves the discovery of a familiar form—aletter, a previously cued shape, or the familiarview of the alternative perspective on a Neckercube. Thus, it is possible that the imager hassimply replaced one image with another, has(in Casey's words) imagined something dif-ferent rather than imagined something differ-ently. This distinction does have an empiricalconsequence: In order to replace an image,one must know either with what to replaceit or how specifically to alter the currentimage. Thus, the critical test of whetherimages can be reconstrued hinges on whethersubjects can discover an unanticipated, un-cued shape in an image. To find out if this ispossible, subjects in the present experimentswere asked to imagine an ambiguous figure(the duck/rabbit) and then asked to inspecttheir image for an interpretation of the figureother than that they had initially seen. Weask whether subjects can discover the alter-native construal by inspecting the image, thatis, whether the first reversal of this figureoccurs while they are inspecting the image.

We chose the duck/rabbit (Figure 1A) forour test stimulus for three reasons. First, weexpected that this simple outline drawingwould be quite easy to image. In this way,any failures to reverse the figure could notbe attributed to difficulties in creating or

maintaining the mental image. (This expec-tation appears to have been correct: Oursubjects, informally polled, indicated thatthey had no difficulty in forming an imageof the figure.) Second, this figure seems lesswell known than some other ambiguous fig-ures, making it easier, we hoped, to findsubjects who were attempting to reverse itfor the first time. (Note that our subjects donot need to be completely naive to ambiguousfigures but only to the test figure.) Third, inorder to maximize our chances of observingfigural reversals, we wanted our instructionsto be as explicit as possible. In particular, wewanted to acquaint subjects with an exemplarthat would ensure that they understood ex-actly what was requested of them. There issuch a parallel exemplar for the duck/rabbit,namely the chef/dog figure (Figure IB). Bothfigures are simple line-drawings, both dependon orientation, and both are heavily influ-enced by the subjects' visual fixation. Theprocedure exploits these similarities to in-struct subjects and thus to facilitate recon-struals of the mental image.

The logic of our procedure requires thatsubjects do not reconstrue the figure duringthe initial exposure of the stimulus, so thatreversals of the imaged stimulus (if they occur)will be first reversals for that figure. With thisin mind, we presented the stimulus to subjectsfor a brief (5 s) exposure. Pilot data indicatedthat this was ample time to encode thesesimple figures yet not enough time for naive.subjects to find the reconstrual.1 Subjects'performance in the reported experimentsconfirms both of these claims.

Experiment 1

Method

Subjects. Fifteen subjects were recruited from variousuniversities in the New York area. Psychology studentswere excluded from the subject pool (for this and all ofthe procedures reported), because we feared they wouldbe familiar with our test stimuli. None of the subjects

we used indicated that they had seen ambiguous figures

' This is, of course, consistent with much evidence inthe ambiguous figures literature. Although reversals sub-sequent to the first happen quickly, the first reorganizationoften requires more than 5 s. See, among others, Girgus,Rock, and Egatz. 1977; Reisberg and O'Shaughnessy,1984.

320 DEBORAH CHAMBERS AND DANIEL REISBERG

before, nor did any, in our training or instruction, giveany indication that they were familiar with the figures.

Stimuli and materials. Slee's Visual Elaboration Scale(VES; Slee, 1980) was employed to assess subjects' imageryvividness. Instructions for this scale urge subjects to thinkabout certain objects (e.g., an animal skin hanging on a

wall); subjects are then probed to determine how muchthey had spontaneously elaborated the image ("Was theskin a particular distance off the floor, or hadn't youthought about that?"). Subjects received one point foreach of the elaborations they included; scores on theVES ranged from 0 to 16.

Subjects saw both test and training stimuli in this (andthe subsequent) procedures; the training stimuli weredesigned to acquaint subjects with figural reversal. Allstimuli were drawn in black on white paper, then pho-tographed. During the procedure, subjects used a hand-held slide viewer to see the stimuli, all of which were on1' X 1* transparencies. In the first experiment, our am-biguous test figure was the duck/rabbit; our trainingfigures were the chef/dog, the Necker cube, Mach's

"book," and the vase/face.Procedure. All subjects were run individually. The

procedure consisted of six steps. First, the VES wasadministered. Second, subjects were shown the Neckercube, the Mach book, and the vase/face slides to acquaintthem with ambiguous figures and to ensure that eachsubject was able to reverse these test figures. In both thepractice and the test trials, the subjects were instructedto announce when they had discovered the alternate viewof the figure; they were then required to describe both

construals to ensure that the reversal had in fact occurred.Third, subjects were shown the duck/rabbit slide, ourtest stimulus, for 5 s and were explicitly instructed toform a "mental picture" of this slide so that they wouldbe able to draw it later. Although the slide was presentedfor only 5 s, subjects were given as much time as they

needed to complete their mental picture. Fourth, subjectswere shown the chef/dog figure, which, like our teststimulus, depends for its interpretations on orientation.Subjects were explicitly given the hint that shifting theirvisual fixation from the tower left corner of the figure tothe upper right corner might help them to find the

alternative construal. We hoped that the presentation ofthis figure, together with our hint, would instruct subjects

in how best to look for a new construal of the duck/rabbit image.

Fifth, after the subjects had successfully found bothconstruals of the chef/dog figure (all did), they wereasked to inspect their image of the previous slide (duck/rabbit). Subjects were given as much time as they neededto form the image; all subjects reported having the imagein a few seconds. The subjects had all seen either theduck or the rabbit initially; they were now asked toidentify the alternate view. To aid subjects in reconstruingthe image, they were given a standardized series ofprompts. They were first asked if they had a "clearmental picture of the previous slide," then whether theycould find an alternative interpretation "in the same waythat [they] did for the chef/dog." If subjects still couldnot reverse the figure, they were again asked whetherthey '%tiU had a clear picture of the figure" and wereurged to look at the "east corner of the figure." Whenthe subjects had shifted their mental fixations, they wereasked what the figure "most resembled"; this was thenrepeated with the "west" corner. (These are in fact thefixations that facilitate reinterpretation with a perceptuallygiven stimulus.) This entire series of prompts was repeatedthree times for each subject so that the entire imageryportion of the experiment took approximately 3 to 4min.

Finally, in the last step of the procedure, subjects wereasked to draw a picture of the imaged stimulus and toinspect their drawing until they found the alternate

y/Figure I. Stimulus A was our test stimulus for Experiments 1 and 2; Stimulus B was used to acquaintsubjects with reversal; Stimuli C and D were used as test stimuli in Experiment 4.

AMBIGUOUS IMAGES? 321

construal. If they did not spontaneously report bothconstruals, they were coached in the same fashion as

they had been with the mental image.

Results

The results are summarized in Table 1.Our subjects spanned a range of imageryvividness, with a mean VES score of 8.23(SD = 2.87) and a range from 2 to 13.

Despite the inclusion of several "high viv-idness" imagers, none of the 15 subjectstested was able to reconstrue the imagedstimulus, despite the familiarization with am-biguous figures in general and the chef/dogin particular and despite our hint aboutfixation. In sharp contrast, all 15 of thesubjects were able to find the alternate con-strual in their own drawings. This makesclear that subjects did have an adequatememory of the duck/rabbit figure and thatthey understood our reconstrual task.

Although we did not time subjects in draw-ing, all subjects drew the figure immediatelyand fluidly. That is, subjects did not constructtheir drawings in a piece-meal fashion buttypically drew the outline without lifting theirpencil from the paper.

It is at least possible that subjects' memoryimages were biased by their original interpre-tation of the figure, making reconstrual ofthe image much more difficult. We knowfrom Bartlett (1932) and Carmichael, Hogan,and Walters (1932) that these biasing effectsare quite strong: Images are distorted in away that makes them more consistent withour construals of them. To rule out thepossibility that this accounts for our negativeresult, Experiment 2 is designed to minimizethis possible memory distortion and hence tomaximize our chances of detecting an ambig-uous image.

Experiment 2

The second experiment is identical to thefirst except for one important change in theinstructions. Prior to the presentation of theduck/rabbit figure, Bartlett's experimental re-sults (and accompanying figures) showingmemory distortions of images were explainedto subjects. Subjects were explicitly asked toavoid these types of errors when encodingtheir image and were instructed to create an

Table 1Summary of Results

Number of reversals

Experiment andtest stimulus

1 Duck/rabbit2 Duck/rabbit4 Duck/rabbit

Necker cubeSchroder staircase

N

1510101010

Fromimage

00000

From owndrawing

151010106

"unbiased" literal copy of the slide. (Giventhe power of memory distortion, this manip-ulation may not be completely effective.However, the third experiment will provide ameans of assessing the fidelity of our subjects'memories.)

Method

Subjects. Ten new subjects were recruited from acommunity college in New Jersey. As before, psychologystudents were excluded; all subjects indicated that theyhad never seen the ambiguous figures. None gave anyindication during the procedure of familiarity with thefigures.

Procedure. With the exception of the instructions,the procedure was identical to that of the first experiment.Before subjects were shown the duck/rabbit slide (andafter they had seen the practice ambiguous figures),subjects were told that people have a strong tendency toremember biased versions of these figures and wereshown drawn examples (the famous eyeglass/barbell figure)of how these bias effects are manifested. Subjects wereencouraged to remember the slide exactly as it appearedso that they would later be able to draw a literal copy.

Results

Our subjects' VES scores ranged from 1 to15, with a mean of 8.87 (SD = 2.92).

None of the 10 subjects was able to findmore than one construal of their image; allof the 10 subjects were able to find thealternate construal, a few moments later, intheir own drawing. As in the first experiment,all subjects drew the figure fluidly and quickly.

Experiment 3

Were our instructions to subjects to forman unbiased, literal picture of the stimulussuccessful? One way to ask this is by askingwhether subjects' drawings of the figure were

322 DEBORAH CHAMBERS AND DANIEL REISBERG

in fact ambiguous, that is, could supportboth construals. In the third experiment, wetake the drawings produced by our Experi-ment 2 subjects, show them to a new groupof subjects, and ask them to find both con-struals of the figure.

Method

Subjects. Thirty subjects were recruited for this ex-periment; as before, the subjects were all naive to thetest figures.

Stimuli and procedure. The practice stimuli in thisprocedure were the same as those used earlier; our teststimuli were the drawings produced by the Experiment

2 subjects. The procedure is identical to that of Experi-ments 1 and 2 except for four changes—we omit the

VES and the procedural step in which subjects were firstasked to encode the image; we substitute the drawing in

place of an image of the duck/rabbit; and we omit therequirement to draw the figure. As before, subjects wereshown the practice figures: the Necker cube, the vase/face, the Mach book and the chef/dog. Each subject was

then shown a drawing produced by one of the Experiment2 subjects. Because each subject in the current procedure

saw only one of these drawings, our 30 subjects allowedus to show each drawing to 3 new subjects. If subjects

did not spontaneously note both construals of the drawing,they were coached in the same way as the earlier subjects,and, in particular, were instructed to try fixating differentcorners of the figure. As before, once subjects announcedthe discovery of both construals, they were required todescribe both views of the figure.

Results

The results indicate that our earlier subjectsdid encode a reasonably unbiased version ofthe stimulus, as the majority of their drawingssupported both construals for new viewers.Four of the drawings are shown in Figure 2.(We include the three "best" drawings andthe "worst," as judged by both the authorsand as reflected in the Experiment 3 results.)Of the 10 drawings, 8 were reversed by atleast one subject, 6 by at least 2 subjects, and5 by all 3 of the subjects who viewed them.In this procedure, we accepted responses ofbird or chicken as equivalent to duck. Wealso accepted goat and giraffe as substitutesfor rabbit, because subjects were able toidentify the animals' ears, nose, eye, andneck—all features shared with the rabbitconstrual. (These alternate labels were usedby only 2 of the 30 subjects; the modalresponse was, as expected, rabbit.) We didnot, by this logic, accept the whale or footresponses, each offered by 1 subject.

It should be noted that these data, althoughinstructive, necessarily provide a weak test ofwhether the information in our subjects' im-age was an unbiased rendition of the teststimulus. Even though 8 subjects succeededin creating ambiguous drawings, we have noguarantee that these drawings were "fromtheir image."2 Nonetheless, the third experi-ment provides an additional line of argumentfor two critical claims, first, that subjects inExperiment 2 adequately encoded the testfigures, and, second, that they did, in fact,discover the alternative construal in their owndrawings.

Experiment 4

Although the results so far seem to arguethat images are not subject to reconstrual,one might challenge this conclusion on anyof three grounds. First, it is conceivable thatour population of subjects does not includean adequate number of "vivid imagers." Thisis, of course, tantamount to denying thevalidity of the VES assessment. This criticismseems unlikely, because our subjects includedseveral artists and architects, members ofprofessions likely to select systematically peo-ple with rich mental imagery. Nonetheless,the final procedure addresses this possibleproblem.

Second, our results may be peculiar to theuse of the duck/rabbit figure, which maysimply be difficult to reconstrue. Third, it isconceivable that we have not given subjectssufficient familiarization with figural reversal,again making it overly difficult to reconstruethe test stimulus. The last experiment answersthese possible criticisms.

2 There are many reasons why the other 2 subjectsmight have produced a drawing that is ambiguous forthe drawer but unambiguous for some other observer. Adrawer might realize, for example, that parts of the figurewere drawn with less confidence, or with a different scale,and might compensate for this in inspecting his/her owndrawing; a new observer would have no way of makingthis compensation. Alternatively, what is an adequaterepresentation for one person may not be adequate foranother, so that a drawing that "works" for its creatormight be insufficient for another person. This is certainlyconsistent with the fact that some of the drawings wereambiguous for only some of the subjects in Experiment3. Whether these speculations or some others turn out tobe correct, however, is not relevant to our overall argument.

AMBIGUOUS IMAGES? 323

Figure 2. Four of the drawings produced by subjects in Experiment 2. (All of these drawings except theone on the bottom were ambiguous for all of the subjects who viewed them in Experiment 3.)

Method

Subjects. Ten new subjects were recruited from various

New York area campuses; again, none was a psychologystudent and alt were naive to ambiguous figures.

Stimuli and materials In addition to the materialsalready described, the fourth experiment employed anadditional imagery vividness measure, the Vividness ofVisual Imagery Questionnaire (VVIQ; Marks, 1972).

All stimuli were presented as before, via the hand-heldslide viewer. We employ three test figures: the duck/rabbit from the first two experiments, a slide of theNecker cube, and a slide of the Schroder staircase. Inaddition, we employed 10 additional training figures, allline drawings of geometrically reversible figures.

Procedure. The procedure followed that for Experi-ment 2, including the instruction about forming literalimages, with the following changes. Subjects were askedto inspect, image, and then draw each of our three testfigures in turn. (The order of presentation for the test

figures was randomly varied.) In each case, the inspectionperiod (as before) was 5 s; subjects were encouraged to

find the alternate construal of their image and, failingthat, were asked to draw the figure and find the alternateconstrual of their drawing. As in the first two procedures,subjects announced when they had discovered the secondconstrual and were then required to describe both views.

Results

The mean VES score was 8.83 (SD =2.96); the scores ranged from 1 to 15. Themean VVIQ score was 33.3 (SD = 18.95),with a range of 20-70. These two imagerymeasures were correlated significantly (r -—.86, p < .01). (The negative correlation isdue to the way the VVIQ is scored—"high"imagery on this test yields a low score.) There

is thus no basis in our data for arguing thatwe have an inadequate sample of vivid imag-ers. There is likewise no basis for distrustingthe VES scores; this argues that our earlierprocedures also had an adequate group ofvivid imagers.

Once again, none of the 10 subjects re-ported a reconstrual of an image, despite thepresence of three opportunities for each sub-ject. All subjects were able to draw and thenreconstrue both the duck/rabbit and theNecker cube stimuli. Six of our 10 subjectswere able to draw and then reconstrue theSchroder staircase. The other 4 subjects haddifficulty in drawing the staircase and reportedthat their drawing did not accurately representtheir image of this figure. The results aresummarized in Table 1.

General Discussion

According to our results, images are notambiguous. None of our 35 subjects reporteda single reversal on any of the test trials. Oursubject population contained many with vividimagery (by the standardized self-report mea-sures), including both artists and architects,so that there is no basis for claiming that oursubjects were weak in imagery skills. Mostimpressive, our subjects were uniformly ableto create an ambiguous picture from anunambiguous image, a finding that empha-sizes the difference between images and per-cepts.

324 DEBORAH CHAMBERS AND DANIEL REISBERG

Could our negative results be due to someprocedural insensitivity? One might claimthat subjects had failed to encode the teststimuli adequately or had not adequatelyunderstood what a figural reversal was. Thesepossibilities are, of course, ruled out by sub-jects' success in reversing their own drawing.Alternatively, it is possible that reversals fromimages are difficult simply because one cannotperceptually explore the figure, cannot shiftone's gaze. However, the same is true for theafterimages of briefly flashed, brightly litstimuli, and we know that such afterimagesof ambiguous figures reverse quite readily(Magnussen, 1970;Piggins, 1979). Third, onemight argue that our subjects were able toimage our stimuli only part by part, so that,without opportunity to confront the entirefigure, subjects were blocked from reversingthe figure. This seems unlikely, given boththe simplicity of the figures and subjects'informal self-reports (which, for our high-imagery subjects, indicated the presence of acomplete and vivid image). Moreover, it isnot obvious that one needs to see an entirefigure to discover a reconstrual. Hochberg(1970) showed subjects ambiguous figuresmoving behind a narrow slit, so that only anarrow strip of the figure was visible at anygiven time. Subjects were still able to reversethe figures under these viewing circumstances.

Some early readers of this manuscript havesuggested that the demand characteristics ofour procedure blocked subjects from reportinga figural reconstrual. This claim seems du-bious. For one thing, our subjects showedsigns of considerable frustration when unableto discover the alternative view of the image,particularly since they had easily reversed thepractice figures a few moments earlier. Thesubjects also expressed puzzlement at theirfailure when they subsequently quickly re-versed their own drawings. This seems not atall like subjects who, because of experimenterdemand, are initimidated into denying thefigural reversal. Furthermore, all overt aspectsof the procedure—our instructions to reportthe alternative view as soon as it was discov-ered and our coaching and hints about fixa-tion—communicated to the subject that areversal of the imaged stimulus was, in fact,possible. (Indeed, it was our expectation,prior to seeing our own results, that such

reversals were possible!) Likewise, it is difficultto conceive of experimenter demand strongenough to have produced our entirely uni-vocal results: One would expect at least 1recalcitrant subject to resist this pressure.

Finally, one might argue that our subjectssimply were not forming mental images butwere instead thinking of some verbal or prep-ositional code. In the extreme, a subjectmuttering "duck, duck" is unlikely to havethe label rabbit come to mind. By the sametoken, one could argue that subjects werecreating their drawings from some memoryrepresentation other than a mental image. Anumber of considerations, however, argueagainst this view. For one, our instructionsfor both the main task and for the VVIQexplicitly and repeatedly mention the creationof a mental picture. Likewise, our hints tosubjects to try moving their "mental gaze"across the figure or to look at a particularcorner presume a spatially laid-out represen-tation of some sort; subjects indicated noconfusion about this instruction. It is unclearhow this instruction (or, for that matter, ourrequest to find a reconstrual) could be un-derstood without reference to a mental image.Finally, aspects of subjects' behavior all seemto indicate the use of mental imagery: Subjectswould often close their eyes during the testor shift their gaze to some blank surface.Subjects also used the kinds of "visual lan-guage" that one associates with imagery, talk-ing about "looking at" or "seeing" the figureor what a portion of the figure "looks like."In short, there seems to be no basis forarguing that our subjects failed to generatemental images in this procedure.

There is, of course, a word of cautionneeded here. For obvious reasons, there is noway to assess directly what information sub-jects' images do or do not contain or whetherthe images provide an accurate facsimile ofthe test stimuli. As we noted earlier, althoughsubjects' drawings are ambiguous, we haveno guarantee that these drawings come "fromimagery." Likewise, although we have pro-vided a series of converging indications thatsubjects were imaging, there is no way toverify that subjects did indeed have a visualform in mind. In the same vein, subjects'self-reports of a vivid, complete image maybe inaccurate. In short, we are confronted by

AMBIGUOUS IMAGES? 325

what may be an intractable methodologicalproblem: There is no way to prove thatsubjects do have an image and that theirimage "has a particular shape." These prob-lems are not unique to this study but pervadethe mental imagery literature (and beyond).The best we can provide is a pattern ofconverging argument for our claims aboutthese phenomena, and that is what we havedone here.

If we thus set aside these alternative ac-counts of our data, the results strongly supportthe arguments offered by Casey, Kolers, andFodor that images cannot be reconstruedbecause images are not construed. Images arenot quasi-pictorial in the sense of needingsome interpretation. Instead, an image's ref-erent is specified by the imaginer. Thus al-though an image might in some fashionresemble more than one content (e.g., a duckor a rabbit, someone falling or someonediving forward), this does not result in am-biguity because it is not by resemblance thatimages refer.

These results also support our earlier claimsabout tasks like Slee's or Reed's. Our argu-ment here is simply that these tasks can beperformed without reconstrual. Thus althoughone could describe these tasks in terms ofreinterpreting images, there is nothing tocompel us to adopt such an account. Giventhis, given the philosophical arguments againstconstrual, and given our results, these earliertasks cannot constitute evidence that imagescan be reconstrued.

How might one perform the imageryembedded-figures task without reconstrual,for example, how might one decide whetheran imaged parallelogram could be embeddedin a Roman numeral 20? All one needs todo, first, is replace one image with another(i.e., replace the imaged parallelogram withone that simply adds or deletes specific parts).By a succession of such replacements, onemay come across a form isomorphic to thetarget figure. Second, one must be able torecognize this isomorphism between theimagined figure and the target. The imaginerunderstands the imagined parallelogram withadded line segments as being just that, andnot a Roman numeral 20, and hence noreconstrual has occurred. At the same timethe imaginer can detect that these two distinct

figures have a common form. Finally, oncethis isomorphism is discovered, subjectsshould be able at will to exchange the twoimages, the embellished parallelogram andthe Roman numeral, and may in this wayconfirm the correspondences.3

Contrast this with the present procedure.Our subjects could have in principle detectedthe isomorphism between imaged duck andimaged rabbit, but this detection presumesboth representations are already available forcomparison. Hence sensitivity to this iso-morphism is of no use in discovering thealternate interpretation of the image. Likewiseour subjects could have freely replaced theirduck image, perhaps with a rotated duck oran embellished duck, but these replacementsare all still imaged ducks and are no moreisomorphic to the rabbit than was the originalfigure. Hence the process we are describingfor the imagery embedded-figures task willnot succeed in a task like ours.

That a judgment of isomorphism is possiblein imagery is itself interesting and leads usto an important point. Intending to imagesome X does not imply that one knows whatthe image will look like. In our study, subjectscreated an image to be an image of a duck(or of a rabbit) but may have been surprisedby the size of the image, the color of thebackground, and so forth. These and otherparticulars of the image may be determinedby the relative ease of imagining one versionor another of the imagined object, certainconventions arising from experience, the spe-cifics of some canonical representation (cf.Kennedy, 1983), or, significantly, propertiesof the imagery medium, and so on. In short,

3 Two things should be noted about this account. First,it is by no means unique to pictorial stimuli. If oneimagines hearing the phrase "the right wright," one isfully aware that it is this phrase (and not some homo-

phonous one) being imagined. This knowledge of prep-ositional content, however, does not preclude detectingthe homophone.

Second, one can easily find variants of our account ofthe embedded-figures task. For example, rather than

imagining a succession of embellished parallelograms,one might imagine just one comer of the shape and thenconsider the target figure, seeking an isomorphic corner.We have no stake in specifying which variant is correct.What is critical is merely that one does not need recon-strual for this task. Because of this, subjects succeed inthe embedded-figures task but fail in ours.

326 DEBORAH CHAMBERS AND DANIEL REISBERG

the fact that an image is created as a symbolof some particular thing (and hence requiresno construal) does not imply that all of theparticulars of an image are anticipated. Be-yond these unanticipated particulars, one canalso be surprised by relations inside of animage. One might decide to imagine oneobject next to another yet still be surprisedby aspects of the juxtaposition.

In all of these ways, one can image withoutknowing in advance what an image will looklike. Thus, even though one does not needto inspect an image to learn what the imagerepresents, one does need to inspect theimage to learn how it appears. This plays animportant role in our account of Slee's orReed's data. It is also consistent with theinformal observation that one is often sur-prised by one's images (see also the numerousanecdotal reports of problem solutionsprompted by a mental image). Some experi-mental evidence makes the same generalpoint: Finke and Kosslyn (1980), in a studyof imagery "two-point acuity," reported thatsubjects can apparently learn from inspectingan image whether two points can or cannotbe seen as separate at a given retinal eccen-tricity. Pinker and Finke (1980) reported thatsubjects can discover "emergent two-dimen-sional patterns" in imaged three-dimensionalconfigurations. These observations are in noway inconsistent with the view we have beendeveloping in this article, for the reasons justdiscussed.

Finally, the present results begin to setlimits on just what are the "functional equiv-alences" of images and percepts (after Finke,1980). Beyond this, however, our results bearon how the question about equivalence shouldbe cast. Images might be seen as entering thestream of visual information processingevents, and so images and percepts have acommon final path. On this view, images andpercepts would share properties arising sub-sequent to this entry but would not shareproperties arising earlier in the processingsequence. Additionally, not all of the infor-mation in images needs to activate visualmechanisms but might be carried along inparallel with the processing mechanisms ofperception. (Cf. Finke, 1980, pp. 130-131.)We believe that this view (at least in thesimple version just described) is not consistent

with the present results. On the one hand,one might try to argue that reconstrual ofambiguous figures occurs early in the streamof visual processing (prior to the entry ofimages) and hence is a property of perceptionbut not imagination. Such a claim seemsimplausible in light of current "cognitive"views of figural reversal. (See Girgus, Rock,& Egatz, 1977; Hochberg, 1970; Reisberg,1983; Reisberg & O'Shaughnessy, 1984; Rock,1983; and others.) On the other hand, onecould claim that the mechanisms for inter-preting a visual stimulus are among thosenot activated by imagery. In this way, theunderstanding or interpretation of an imageis one of the properties maintained apartfrom the sequence of visual processing. Thedefect in this suggestion is that this sequenceof visual processing is, presumably, largelyconcerned with just such problems as figuralconstrual. If images do not enter this in-terpretive process, it is difficult to characterizewhat common final path for images andpercepts this leaves.

Our results can thus be taken as arguingagainst the claim that imagery and perceptionshare a common processing path. This argu-ment is all the more pressing given the ques-tions recently raised about the empirical baseof the common path claim. (See Broerse &Crassini, 1980, 1981; Harris, 1982; Intons-Peterson & White, 1981; Kolers, 1983; Kunen& May, 1981; Neisser, 1978). Nonetheless,the fact remains that, as we noted at theoutset of this article, there is considerablecommonality between perceiving and imag-ining. An alternative account of this com-monality, one consistent with our data, isreadily available. As Neisser (1978) notes, apopular view of perceiving treats it as a kindof information processing and emphasizes asequence of "bottom-up" events. This con-ception is implicit in the common processingpath claim; these bottom-up processes are(to some extent) shared by imagery and per-ceiving (cf. Kolers, 1983). In contrast, considerconceptions of perception emphasizing "top-down" or "cyclic" processes (e.g., Gregory,1970; Hochberg, 1970; Neisser, 1976; Rock,1983). These conceptions, all of which involvean "interpretive," "constructive," or "antici-patory" aspect of perception, provide an ob-vious basis for the properties shared by im-

AMBIGUOUS IMAGES? 327

aging and perceiving. Imaging is, or at leastdraws heavily on, the top-down or anticipatoryaspects of perceiving. Neisser (1978) has ex-plicitly described one way this view might bedeveloped: Perception, on Neisser's view, in-volves a "cyclic interaction" with the world,a critical part of which is "anticipation."Imagination is simply this anticipation in theabsence of the stimulus. In short, to image isto get ready to see.

A closely related view can be derived fromthe work of Rock (1983), who has arguedthat form perception is a process of achievinga "non-verbal description" of the stimulus, adescription that includes specification of whatis figure, what ground, that defines the properorientation of the figure, and that identifiesaxes of symmetry. The description also spec-ifies the depth arrangement of the figure, sothat two-dimensional drawings are assigneda three-dimensional interpretation. Thus forRock, percepts are in no sense "copies" ormere transcriptions of the input; they are, atthe least, interpreted representations or la-beled copies. This notion of percept-as-de-scription obviously resembles our claim thatimages are, to use Fodor's (1981) language,"images under description." Given Rock'sdescription of the deliberative, intelligent,thoughtlike nature of this process, it wouldnot be surprising if the process could beinitiated by the imaginer, and not only bythe presence of some stimulus. Hence, oursuggestion, following Neisser's lead, is thatthis description, occurring in the absence ofstimulus information, is what we call a mentalimage.

Thus imagery and perception will not sharea number of aspects, including those asso-ciated with bottom-up processes or thoseresulting from the interaction of top-downand bottom-up events. (Construal, the processof finding what a stimulus is, presumably isone such interactive process.) At the sametime, imagining will share at least some ofthe top-down processes and will reflect theproperties of those processes. This providesa ready means of accounting for the phenom-enal similarity between images and percepts.It is also easy to see how the activity ofimaging can both interfere with perceiving(Segal & Fusella, 1970) or, under differentcircumstances, pave the way for perceiving

(Posner, 1978). This view is also consistentwith findings that imagery and perceptionseem to share a mode of representation, amode that respects the metric properties ofspace. Finally, images should resemble per-cepts in how they interact with higher mentalprocesses. If we adopt Rock's view of perceptsas descriptions, then both images and perceptscarry with them a retrieval set, so to speak,that governs how they interact with memory.Our imagery results, in fact, echo some ofRock's perceptual data concerning orientationor figure/ground organization. How a stimulusis internally described governs what priorfigures are evoked from memory. Thus afamiliar shape (e.g., an outline drawing ofAfrica) will not be recognized if it is assigneda "top" other than the northern edge (Rock,1973, 1983). This obviously parallels ourfindings in which subjects, having assignedone interpretation, fail to discover (or evoke)a second.

This view of imagery forces on us a ques-tion that we earlier set aside. Regardless ofthe "language" of this internal representa-tion—analog or proposition, "description" or"anticipation" (Kosslyn, 1981; Pylyshyn,1981; Kolers, 1983)—we need to ask what isincluded in this representation. As alreadynoted, Rock (1983) argues for some of theaspects that must be included in percepts—specification of figure and ground, orientation,depth relations; surely these same minimalaspects must be included in images as well.The more difficult problem lies in askingwhat else is included, and, additionally,whether the specification of images is equiv-alent in form to that of percepts. These areclearly problems for further work, both onempirical and philosophical fronts.

References

Bartlett, F. (1932). Remembering: A study in experimental

and social psychology. Cambridge: Cambridge Univer-sity Press.

Brentano, F. (1973). Psychology from an empirical stand-point (A. Rancurello, B. Terrell, & L. McAlister,Trans.). London: Routledge & Kegan Paul. (Originalwork published 1874)

Broerse, J., & Crassini, B. (1980). The influence ofimagery ability on color aftereffects produced by phys-ically present and imagined induction stimuli. Percep-tion & Psychophysics, 28, 560-568.

Broerse, J., & Crassini, B. (1981). Misinterpretations of

328 DEBORAH CHAMBERS AND DANIEL REISBERG

imagery-induced McCollough effects: A reply to Finke.Perception & Psychophysics. 30, 96-98.

Brooks, L. (1968). Spatial and verbal components of theact of recall. Canadian Journal of Psychology, 22, 349-368.

Carmichael, L., Hogan, H., & Walters, A. (1932). Anexperimental study of the effect of language on thereproduction of visually perceived form. Journal of

Experimental Psychology, 15, 73-86.Casey, E. (1976). Imagining: A phenomenological study.

Bloomington, IN: Indiana University Press.Dennett, D. (198 la). The nature of images and the

introspective trap. In N. Block (Ed.), Imagery (pp. 51-

61). Cambridge, MA: MIT Press.Dennett, D. (1981 b). Two approaches to mental images.

In N. Block (Ed.), Imagery (pp. 87-107). Cambridge,

MA: MIT Press.Finke, R. (1979). The functional equivalence of mental

images and errors of movement. Cognitive Psychology,

II, 235-264.Finke, R. (1980). Levels of equivalence in imagery and

perception. Psychological Review, 87, 113-132.Finke, R., & Kosslyn, S. (1980). Mental imagery acuity

in the peripheral visual field. Journal of ExperimentalPsychology: Human Perception and Performance, 6,

126-139.Finke, R., & Schmidt, M. (1977). Orientation specific

color aftereffects following imagination. Journal of

Experimental Psychology: Human Perception and Per-

formance, 1, 599-606.Fodor, J. (1981). Imagistic representation. In N. Block

(Ed.), Imagery (pp. 63-86). Cambridge, MA: MIT

Press.Girgus, J., Rock, I., & Egatz, R. (1977). The effect of

knowledge of reversibility on the reversibility of am-biguous figures. Perception & Psychophysics, 22, 550-

556.Gregory, R. (1970). The intelligent eye. London: Weiden-

feld and Nicholson.

Harris, J. (1982). The VVIQ and imagery-producedMcCollough effects: An alternative analysis. Perception

& Psychophysics, 32, 290-292.Hochberg, J. (1970). Attention, organization and con-

sciousness. In D. Mostofsky (Ed.), Attention: Contem-porary theory and analysis (pp. 99-124). NY: Appleton-

Century-Crofts.Intons-Peterson, M., & White, A. (1981). Experimenter

naivete and imaginal judgments. Journal of Experi-

mental Psychology: Human Perception and Performance,

7, 833:843.Keenan, J. (1983). Qualifications and clarifications of

images of concealed objects. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 9, 222-

230.Keenan, J., & Moore, R. (1979). Memory of images of

concealed objects: A reexamination of Neisser andKerr. Journal of Experimental Psychology: HumanLearning and Memory, 5, 374-385.

Kennedy, J. (1983). What can we learn about picturesfrom the blind? American Scientist, 71, 19-26.

Kerr, N., & Neisser, U. (1983). Mental images of concealedobjects: New evidence. Journal of Experimental Psy-chology: Learning, Memory, and Cognition, 9, 212-221.

Kolers, P. (1983). Perception and representation. Annual

Review of Psychology, 34, 129-166.

Kosslyn, S. (1980). Image and mind. Cambridge, MA:Harvard University Press.

Kosslyn, S. (1981). The medium and the message inmental imagery: A theory. In N. Block (Ed.), Imagery

(pp. 207-244). Cambridge, MA: MIT Press.Kunen, S., & May, J. (1981). Imagery-induced Mc-

Collough effects: Real or imagined? Perception & Psy-chophysics, 30, 99-100.

Magnussen, S. (1970). Reversibility of perspective innormal and stabilized viewing. Scandinavian Journalof Psychology, 11, 153-156.

Marks, D. (1972). Individual differences in the vividnessof visual imagery and their effect on function. In P.Sheehan (Ed.), The function and nature of imagery(pp. 83-108). NY: Academic Press.

Neisser, U. (1976). Cognition and reality. San Francisco:Freeman.

Neisser, U. (1978). Perceiving, anticipating and imagining.In W. Savage (Ed.), Perception and cognition: Issues in

the foundations of psychology (pp. 89-105) Minneapolis:University of Minnesota Press.

Neisser, U., & Kerr, N. (1973). Spatial and mnemonicproperties of visual imagery. Cognitive Psychology, 5,138-150.

Perky, C. (1910). An experimental study of imagination.American Journal of Psychology, 21, 422-452.

Piggins, D. (1979). The influence of line reduction andimage-stabilization on Necker cube reversal. Perception,

8, 719-720.Pinker, S., & Finke, R. (1980). Emergent two-dimensional

patterns in images rotated in depth. Journal of Exper-imental Psychology: Human Perception and Perfor-

mance, 6, 244-264.Posner, M. (1978). Chronometric explorations of mind.

Hillsdale, NJ: Erlbaum.Pylyshyn, Z. (1981). The imagery debate: Analog media

versus tacit knowledge. In N. Block (Ed.), Imagery

(pp. 151-206). Cambridge, MA: MIT Press.Reed, S. (1974). Structural descriptions and the limitations

of visual images. Memory & Cognition, 2, 329-336.Reed, S., & Johnson, J. (1975). Detection of parts in

patterns and images. Memory & Cognition, 3, 569-575.

Reisberg, D. (1983). General mental resources and per-ceptual judgments. Journal of Experimental Psychology:

Human Perception and Performance, 9, 966-979.Reisberg, D., & O'Shaughnessy, M. (1984). Diverting

subjects' concentration slows figural reversals. Percep-tion, 13, 461-468.

Rock, I. (1973). Orientation and form. New York: Aca-demic Press.

Rock, I. (1983). The logic of perception. Cambridge, MA:MIT Press.

Segal, S., & Fusella, V. (1970). Influence of imagedpictures and sounds on detection of visual and auditorysignals. Journal of Experimental Psychology, 83, 458-464.

Shepard, R., & Cooper, L. (1983). Mental images andtheir transformations. Cambridge, MA: MIT Press.

Slee, J. (1980). Individual differences in visual imageryability and the retrieval of visual appearances. Journal

of Mental Imagery, 4, 93-113.

Received March 20, 1984Revision received December 19, 1984 •