Devcl~m.ntal Ps)\:hology1985, Vol. 21. No.6, 963-973

Copyright 1985 by the American Psychological Association. Inc.00 1 2-1649/85/S00.75

Detection of Intermodal Proprioceptive-Visual Contingency as a, Potential Basis of Self-Perception in Infancy

Lorraine E. BahrickFlorida International University

John S. WatsonUniversity of California. Berkeley

Five-month-old infants can detect the invariant relationship between their own legmotion and a video display of that motion. In three experiments they discriminatedbetween a perfectly contingent live display of their own leg motion and a noncon-tingent display of self or a peer. They showed this discrimination by preferentialfixation of the noncontingent display. This effect was evident even when the infant'sdirect view of his or her own body was occluded, eliminating video image discrim-ination on the basis of an intramodal visual comparison between the sight of self-motion and the video display of that motion. These findings suggest that the con-tingency provided by a live display of one's body motion is perceived by detectingthe invariant intermodal relationship between proprioceptive information for motionand the visual display of that motion. The detection of these relations may be fun-damental to the development of self-perception in infancy. In addition, though 3-month-olds did not show significant discrimination of the contingent and noncon-tingent displays, they did show significantly more extreme looking proportions tothe two displays than did the 5-month-olds. This may reflect the infant's progressionfrom self to social orientation.

Lewis and Brooks-Gunn (1979) found thatby the end of the first year of life, infants areable to discriminate a "live" video image ofthe self from a recorded image of the self or apeer. The authors propose that this self-rec-ognition is primarily based on the detection ofcontingent visual stimulation from the livevideo image. That is, movement of the infant'shand, for example, results in comparablemovement of the hand in the video image.Furthermore, they propose that the earlieststages of self-perception are probably based onthe infant's detection of some form of response

This research was supported by National Institute ofChild Health and Human Development Grants I ROIHDI5275 and 5 ROI HD14393. A portion of these datawere presented at the 1983 biennial meeting of the Societyfor Research in Child Development in Detroit, Michigan.We would like to thank Michele Areola. Norene Huber,Betty Kaufman. Deborah Kennedy, and Nancy Roscellifor their assistance in the collection and analysis of data.

Requests for reprints should be sent to either LorraineE. Bahrick, Department of Psychology, Florida Interna-tional University, Tamiami Campus, Miami. Florida 33199or John S. Watson, Department of Psychology, Universityof California, Berkeley, California 94720.

963

contingent stimulation, such as the contin-gency between visual and proprioceptive feed-back from body motion.

There exists considerable evidence that in-fants as young as 2 and 3 months are sensitiveto experimentally arranged contingencies be-tween behaviors, such as leg kicks, sucking, orvocalization and various forms of visual andauditory stimulation, such as lights, tones, orthe movements of an overhead mobile (Rovee& Rovee 1969; Rovee-Collier & Gekoski, 1979;Siqueland & DeLucia, 1969; Watson & Ra-mey, 1972). There is even some evidence thatthis sensitivity exists from birth (Sameroff,1971).

These kinds of contingent relations, how-ever, differ from those provided by a mirror orvideo image. When the infant turns on a lightdisplay by kicking his or her foot (e.g., Watson,1979) the onset of the light display is contingentupon the onset of the infant's leg motion.When the infant controls the motions of anoverhead mobile that is connected to his orher leg by a ribbon (e.g., Rovee & Rovee, 1969),the motions of the mobile are "conjugately"related to the motions of the infant's leg. They

964 LORRAINE E. BAHRICK AND JOHN S. WATSON

covary along dimensions of amplitude andtemporal pattern. However, the stimulationfrom such a conjugate display is not perfectlyisomorphic with the motions of the infant, likethat of a mirror or video image.

The mirror or video reflection of the selfprovides essentially perfect contingent stimu-lation. It seems a reasonable possibility thatperfect contingency may serve to specify stim-ulation arising from the self, whereas imperfectcontingency may specify stimulation arisingfrom sources other than the self.

If that were so, then one should expect tofind that the visual discrimination of self andother, based on contingency perception, wouldbe observable well within the first year of life.There have been a few studies to date that ap-pear to support this expectation. Papousek andPapousek (1974) found that 5-month-olds dis-criminated a live video image of self from arecorded image of self. Infants showed theirdiscrimination by preferential fixation of thenoncontingent, recorded display of self overthe contingent, live display of self. A subse-quent study by Field (1979) found that 3-month-olds displayed differential responsive-ness to a contingent mirror image of the selfand a noncontingent presentation of a peer.These infants looked more to the mirror imageof the self but smiled and vocalized more inthe presence of a peer. However, several pro-cedural confounds inherent in these kinds ofstudies have made interpretation of their find-ings difficult. Recording the noncontingentvideo film of self under experimental condi-tions that differ from those of the contingentvideo film of self introduces the possibilitythat the amount of general body motion dis-played by infants in the different conditionsmay differ and may then serve as a basis ofdiscrimination between the two kinds of dis-plays. Second, the use of facial images as stim-uli in these studies creates a confound of dif-ferential eye contact and eye motion in con-tingent and noncontingent displays. Themirror image provides constant eye contactand no visible eye motion for the infant,whereas the recorded image of sel f or peer pro-"ides eye contact and eye motion variability.Papousek and Papousek (1974) attempted toseparate eye contact and contingency of theimage by using video controls. Though theydid find an effect of eye contact, the overalleffect of contingency as a basis of discrimi-~

nation between live and recorded presentationsof self was most evident.

These findings would seem to support Lewisand Brooks-Gunn's (1979) assumption thatthe contingency between visual and proprio-ceptive information for self-motion could serveas a basis of self-perception in the first monthsof life. Young infants are constantly presentedwith visual and proprioceptive informationspecifying self. Each time the infant reaches,he or she can both see his or her arm reachingand feel it-reaching. There is a perfect corre-~pondence between the visual and propriocep-tive information for self-motion. The Gibsons(E.J. Gibson, 1969; J. J. Gibson, 1966, 1979)described this kind of information as an in-variant, amodal relation that specifies self.

,It seems importartt at this point to clarifyour use of the term proprioception. because ithas had a variety of connotations in the past.Gibson (1979) maintained that all senses havea proprioceptive as well as an exteroceptivefunction, and therefore he does not identifyproprioception as a separate sensory system.Rather, information about the self is obtainedthrough all the sensory systems. However, forthe purposes of this presentation, we need todistinguish among several types of informationspecifying self. Thus our use of the term pro-prioception will be reserved for that informa-tion intrinsic to behavior that arises from themuscles, joints, tendons, vestibular system, andefferent pathways that specify self-motion. Thisrestricted use of the term facilitates an impor-tant distinction to be drawn between proprio-ception in the above sense and the perceptionof self arising from sources other than thosementioned above (e.g., vision or audition).Thus the distinction between seeing one's handmove and feeling one's hand move (proprio-ception) can easily be made for our presentpurposes.

In all the studies described above concerninginfants' visual perception of self, infants haddirect visual access to the movement of theirhands, upper torso, and, to some extent, evenfacial movement. Therefore, infants' discrim-ination of a live and recorded image of theupper torso, face, and hands may have beenbased on the detection of either of two kindsof contingent relations: (a) a visual-visualcontingency or (b) a proprioceptive-visualcontingency, the theoretically more relevantalternative proposed by Lewis and Brooks-

965PROPRIOCEPTIVE- VISUAL CONTINGENCY

Gunn (1979). That is, infants may have de-tected a contingent relation between the visiblemotions of their own hands and the image ofthat motion depicted on the video screen. Or,they may have detected a relationship betweenproprioceptive information for self-motion andthe visual display of that motion depicted onthe video screen. Research to date has not dis-tinguished between' these two possibilities.Distinguishing these two interpretations re-quires empirical evidence for the capacity ofinfants to perceive cross-modal contingent re-lations between vision and proprioception.

The only evidence to date that suggests thatyoung infants detect proprioceptive-visual re-lations is that on neonatal imitation (Meltzoff& Moore, 1983a, 1983b). The young infant'sability for delayed imitation of facial gesturesis interpreted by the authors to demonstratean ability to monitor proprioceptive infor-mation from the facial muscles and joints andto match their gesture with the visually givengesture of the model.

The present research directly investigates theinfant's capacity to detect proprioceptive-vi-sual relations uniting self-motion with a visualdisplay of that motion. In order to examinethis capabililty, the usual design of mirror orvideo self-recognition studies was modified inseveral important ways. Prior confounds of dif-ferential eye contact and eye motion betweencontingent and noncontingent displays wereeliminated by presenting nonfacial stimuli.The confound of body motion variability wascontrolled by using a yoked control design.And finally, the infant's detection of proprio-ceptive-visual information for self-motion wasinvestigated directly in Experiments 2-4 byoccluding the infant's direct view of his or herown body while viewing the video images. Thiseliminated the possibility of detecting contin-gency on the basis of visual-visual relations.By incorporating these modifications, threestudies investigated 5-month-old infants' sen-sitivity to invariant intermodal relations UQit-ing proprioceptive and visual information forself-motion. A fourth study extended the in-vestigation to 3-month~0Ids.

Experi~ent I

Because of prior confounds of eye contact,eye motion, and general body motion vari-ability, the basis ofinfgnts' discrimination be-

tween the contingent image of the self and thenoncontingent image of the self or a peer is atpresent unresolved. In this study, 5-month-oldinfants were tested for discrimination betweena live video image of self and the noncontingentimage of a peer. Eye contact and eye motionvariability were eliminated by presenting theimage of the infant's body from the waist downso that only leg and foot motion were dis-played. Body motion variability was controlledacross subjects by using a yoked control design.The live image of each infant was recordedand later used as the noncontingent peer filmfor the next infant. Thus all noncontingent"peer" films were recorded while the peer wasalso viewing a contingent video image of selfalong with the noncontingent image of a peer.We were interested in whether infants, in theabsence of prior confounds, would discrimi-nate between these norrfacial stimuli on thebasis of contingency, and whether they wouldpreferentially fixate the contingent or the non-contingent display.

Method

Subjects. Twenty 5-month-old infants, 10 boys and 10girls, participated. Their mean age was 148 days (SD = 5days). Subjects were recruited through the local birth rec-ords. Two additional subjects were tested and excluded be-cause of excessive crying and fussiness.

Apparalu.\" and procedures. Infants were seated in aninfant seat in a booth enclosed by patterned curtains ontwo sides, and they faced two television screens approxi-mately 30 inches (76 cm) away. The television screens werepositioned side by side, about 18 inches (46 cm) apart(from the inside edge of one screen to the other), and acolor video camera was centered between them, directlyin line with the infant's legs. The video camera was turnedupside down and stood at a height near the infant's headlevel and at an angle approximating the infant's view ofhis or her own legs when looking down at them. Becauseof this, the infant's legs appeared on the video screen withthe feet extending in the upper portion of the screen, justas when the infant looks down at his or her feet they extt"'dupward into the infant's visual field. By inverting the cam-era, the right-left inversions typically produced by videofilms were altered. In this case, although the feet extendedupward, the motion of the infant's right leg, for example,was displayed as motion of the leg on the right side of thescreen, in this respect resembling a mirror image.

Figure I is a photograph of an infant viewing the display,and it illustrates our apparatus. All infants were fitted withbright yellow booties and were placed in an infant seat thatwas covered by a bright blue cloth. This was done to max-imize visibility of the feet and to equate for color andbrightness across subjects.

Infants viewed two color video screens side by side for4 min. One screen displayed a live transmission of theinfant's legs and feet (contingent display) and the other

966 LORRAINE E. BAHRICK AND JOHN S. WATSON

Figure J. Photograph of the apparatus and stimulus display.

showed the legs and feet of a peer (noncontingent display).The yoked control design controlled for amount of leg mo-tion and visual characteristics of the noncontingent displayacross subjects. A trained observer, blind to the lateral po-sitions of the displays, monitored the subject's visual fix-ations to the two screens. She held a set of buttons con-nected to a Rustrak strip chart recorder and pressed onebutton while the infant fixated the right-hand screen andanother button while the infant fixated the left-hand screen.A second trained observer also monitored fixations for aproportion of the subjects. Interobserver reliability wascalculated for the primary and secondary observers on 34out of 89 subjects across the four experiments reportedhere, and was .99. This was a correlation between the pro-portions of total looking time infants spent fixating thenoncontingent displays across the two observers.

Lateral positions of the contingent and noncontingentdisplays were counterbalanced across subjects. Ten infantssaw the live display on the right and the peer display onthe left, and 10 received the opposite arrangement. Thetwo televisions were connected to cable splitters and aswitch that allowed us to present either image to the left-or right-hand screens. Either the infant's parent or a secondexperimenter operated the switch that turned on both dis-plays and determined to which side the live and peer dis-plays would be presented.

Results and Discussion

Looking proportions were derived by divid-ing the time infants spent looking to the non-contingent peer image by the time they spentlooking to both films, across the 4-min period.Subjects spent an average of .65 of their totallooking time viewing the noncontingent imageof the peer. This proportion is significantly dif-ferent from the chance value of .50 according

to a t test, t(19) = 3.30, p < .005. At the in-dividual subject level, 16 out of 20 subjectsspent a majority of their total looking timeviewing the noncontingent peer film. Therewere no significant side preferences, l( 18) =.637, p > .05, nor were there any significantsex differences in looking to the two displays,t(18) = 0.147, P > .05. Looking proportionswere also derived f()r each infant for each ofthe four I-min periods separately to test fortrends over time in looking to the noncontin-gent films. A one-way repeated-measuresanalysis of variance (ANOVA) revealed no sig-nificant differences in looking across the fourI-min periods, F(3, 57) = 1.4, p > .05. In fact,looking proportions to the noncontingent peerfilm during the first minute (.66) were com-parable to those of the last minute (.68).

Five-month-old infants can differentiate be-tween a contingent display of their own legmotion and a noncontingent display of a peer'slegs. They showed this discrimination bylooking preferentially to the noncontingentdisplay. This discrimination was evident withnonfacial stimuli and under conditions inwhich prior confounds of eye contact, eye mo-tion, and body motion variability were elim-inated. These findings support the interpre-tation that 5-month-old infants discriminatebetween video displays of self and peer on thebasis of the contingent relations between theirown body motion and that displayed in the

967PROPRIOCEPTIVE-VISUAL CONTINGENCY

live video image. How might infants have de-tected this contingency? What kind of infor-mation might they have used as a basis of dis-crimination? There are at least two possibili-ties. Infants might have (a) made a visualcomparison between the motions of their ownlegs and those pictured on the video display,and/or they may have (b) used proprioceptiveinformation from their muscles and joints anddetected an intermodal relation between thisproprioceptive information from self-motionand the visual information displayed on thevideo screen. Experiment 2 was designed todistinguish between these two alternatives.

Experiment 2

Little direct evidence exists bearing onwhether young infants make use of proprio-ceptive information for determining how theirbody is moving through space-that is, infor-mation from the muscles, joints, and vestibularsystem. Some findings, however, are consistentwith this interpretation, including those of vi-sually guided reaching (e.g., Bower, Broughton,& Moore, 1970; Hofsten, 1979, 1983; Mc-Donnel, 1975) and contingency learning in in-fancy (e.g., Rovee & Rovee, 1969; Watson &Ramey, 1972), though they could as well beexplained on the basis of vision alone. Anotherarea of research bearing on this question isthat concerning the infant's use of visual in-formation for monitoring posture and motorbehavior. These studies of "visual propriocep-tion" (in the Gibsonian sense; Butterworth &Cicchetti, 1978; Butterworth & Hicks, 1977;Lee & Aronson, 1974) found that the postureof infants who could sit or stand unsupportedwas disrupted by visual information for bodysway, even in the absence of vestibular or pro-prioceptive information for body sway. Also,research by Jouen (1984) indicated that theinfant's head righting response subsequent toa body tilt varies according to the orientationof visual stimulation. This research suggestsan early link between vision and propriocep-tion, and it establishes the function of visionfor guiding motor behavior in infancy. Al-though these findings suggest that infants madeuse of both visual and proprioceptive infor-mation for posture control, they may also beinterpreted as evidence for the primacy of vi-sion in guiding motor behavior.

The strongest evidence to date suggestingthat infants make use of proprioceptive infor-mation is that demonstrating imitation behav-ior in neonates (Meltzoff & Moore, 1983a,1983b). Because these infants had minimal vi-sual access to their own facial expressions,these findings are difficult to account for onthe basis of vision alone. It seems likely, there-fore, that even very young infants make use ofproprioceptive information from the muscles,joints, skin, and/or efferent pathways to controland determine the nature (e.g., direction, in-tensity and form) of their physical motion.

Experiment 2 directly tests this possibility.Infants were again shown a live video displayof their own legs side by side with those of apeer. In this replication of Experiment I, how-ever, the infant's direct view of his own bodywas occluded. By occluding infants' views oftheir body, they no longer have access to visualinformation for their leg motion and must relyon proprioception for detecting the contin-gency between their leg motion and the videodisplay of it. We were interested in whetherinfants would visually discriminate betweenthe contingent and noncontingent displayswithout visual access to their own leg motion,demonstrating that they could rely on pro-prioceptive information for specifying self-motion.

Method

Twenty 5-month-old infants. 10 boys and 10 girls. par-ticipated. Their mean age was 150 days (SD = 5 days).Two additional subjects were tested and excluded from thestudy for excessive crying and fussiness. The apparatusand procedures were identical to those used in ExperimentI, except that the infant's direct view of his or her ownbody from the waist down was occluded. This was accom-plished by mounting a wooden tray in front of the infantseat and securing the bottom of a large bib to the tray andthe top of the bib around the infant's neck. The infant'sarms were placed above the bib so that they could moveabout freely and, as in Experiment I, were not visible inthe video display. The infant's legs and torso extended un-derneath the tray and bib so that no direct visual accessto that portion of their body was possible. As in ExperimentI, subjects viewed a contingent video display of their ownlegs side by side with a noncontingent display of a peer'slegs for 4 min. In this study, 12 infants saw the display ofthe peer's legs on the right and their own legs on the left,whereas 8 received the q>posite arrangement. This unequaldistribution occurred by chance as a result of our procedurefor controlling observer bias. The infant's parent manip-ulated the switch controlling the lateral positions of thevideo displays.

968

LORRAINE E. BAHRICK AND JOHN S. WATSON

fI

Result,s and Discussion

Results of the study replicated those of Ex-periment I. Subjects spent .66 of the totallooking time viewing the noncontingent peerimage. This proportion is significantly differentfrom the chance value of .50 according to a ttest, t(19) = 4.39, P < .001. At the individualsubject level, 17 of the 20 subjects spent a ma-jority of their total fixation time watching thedisplay of the peer. There was no significantside preference, t( 18) = 0.298, p > .05, norwere there any sex differences in looking to thenoncontingent display, t( 18) = 2.03, p > .05.An ANOVA was performed on looking prefer-ences for each l'11inute of the 4-min viewingperiod to determine whether there were anytrends over time in looking to the peer display.Results indicated no differences over the fourI-min periods, F(3, 57) = 2.38, p > .05.

Five-month-olds show visual discriminationofa contingent display of their own leg motionand a noncontingent display of a peer's legmotion, even when their view of their ownbody is occluded. Their performance was sim-ilar to that of Experiment I where they hadvisual access to their own bodies. These resultssuggest that young infants need not rely onvisual information for determining how theirbodies are moving about in space. They neednot detect the contingency of the live videodisplay by making a visual comparison of theirleg motion and that displayed on the screen.Rather, they may be able to use proprioceptiveinformation from the muscles and joints todetermine the nature of their physical motion.As a consequence, infants could have discrim-inated between the contingent and noncontin-gent images by detecting the inter modal rela-tionship between the nonvisual proprioceptiveinformation for motion and the visual displayof that motion.

However, before this conclusion can be ac-cepted, another interpretation of the resultsmust be considered. It is possible that infantsdid not detect a contingent relation at all.Rather, they may have recognized features oftheir own legs and feet and discriminated be-tween their own image and that of the peer onthis basis. Though this interpretation seemedrather unlikely for such young infants, it nev-ertheless remained a possibility. Despite thefact that individual features of the infant's legs

and feet were partially controlled by fitting allinfants with yellow booties, there was somedistinctive visual information. The subject'sclothes were visible above the booties andmight have been recognized by the infant.Also, the shape of the infant's legs and idio-syncratic patterns of leg motion might con-ceivably have served to distinguish amongsubjects. Although studies of self-recognition(Lewis & Brooks-Gunn, 1979) report that in-fants do not recognize features of their ownbody until-approximately 11/2 years of age, thisinterpretation could not be conclusively ruledout in the present studies. The possibility thatinfants recognized features of their own bodywas thus tested in Experiment 3 by eliminatingthe distinguishing features altogether.

Experiment 3

In this study, features that distinguishedamong infants' legs ~nd feet were eliminatedby presenting the contingent video image ofself along with a noncontingent, recorded im-age of self, rather than the image ofa peer. Thenon contingent, recorded film was made justprior to testing, and infants wore the sameclothes in both video displays. No visual char-acteristics differentiated the live from the re-corded video displays. Only the contingent re-lation between the infant's leg motion and thelive image of that motion could be used as abasis of discrimination. This contingencycould be detected only through use of the pro-prioceptive information because the occludingbib was again used.

MethodTwenty' 5-month-olds were tested, II boys and 9 girls.

Their mean age was 143 days (SD = 10 days). The appa-ratus and procedures were the same as those used in Ex-periment 2. including the use of the occluding bib, exceptthat a 4-min period preceded the usual testing session inwhich the infant's legs were recorded. During this period,the infant was seated in the infant seat facing the televisionscreens and wore the occluding bib. Subjects were showna 4-min live video display of their own legs in two orien-tations (180° vs. 90°). This was done so that subjects wouldnot become bored or fussy during the recording period aswell as to roughly approximate the viewing conditions ofthe test session in the event that the infants' general amountof leg motion was influenced by viewing displays of infantlegs. After the subject's legs were recorded, the infant wastaken out of the seat and was entertained in the testingroom for approximately a 5-min period. Next, the testsession began and was identical to that of Experiment 2

969

PROPRIOCEPTIVE- VISUAL CONTINGENCY

Table IExperiments J. 2. and 3: Proportion of Total Looking Time Spent Fixating the Noncontingent Display

Experiment 3: Noncontingentdisplay of self

Experiment 2: Noncontingentpeer display

Experiment I: Noncontingentpeer displayStatistic

0.690.1964.34**

0.660.1634.39**

PSDt

0.650.2073.30.

.p < .005. ..p < .001.

rl

except that the recorded video display of self was shownalong with the live display of self. Half the subjects viewedthe noncontingent recorded film on the right and the liveone on the left. whereas half received the opposite arrange-ment. An observer. blind to the lateral positions of thedisplays. monitored the subject's fixations.

Res/ilts and Discussion

Results of the study, along with those of Ex-periments I and 2, are depicted in Table I.These results paralleled those of Experiments1 and 2 exactly. The proportion of total fixationtime spent viewing the noncontingent, re-corded film was .69. This is significantly dif-ferent from the chance value of .50 accordingto a 1 test, 1(19) = 4.34, p < .001. Eighteen ofthe 20 subjects spent a majority of their totalfixation time viewing the noncontingent dis-play. There were no significant side prefer-ences, I( 18) = 1.56, p > .05, or sex differences,I( 18) = 0.988, p > .05, in looking to the non-contingent display. An ANOVA revealed no sig-nificant differences among the proportions oflooking time to the noncontingent displayacross the four I-min periods, F(3, 57) = 0.09,p> .05.

Infants spent a majority of their total fixa-tion time viewing the recorded, noncontingentdisplay, just as they had in Experiments 1 and2. The elimination of feat ural differences be-tween the contingent and noncontingent dis-plays in this study had no decremental effecton viewing time for the recorded film. Theseresults cast doubt on the possibility that infantsrelied exclusively on recognition of visual fea-tures as a basis of discriminating between thedisplays of self and peer leg motion in Exper-iment 2. Rather, this study provides direct ev-idence for an infant's detection of inter modalcontingency between proprioceptive and visualinformation for motion.

Several alternate interpretations of this study

.are also possible, though taken together withthe results of the other studies, they do notprovide serious problems. First, though pre-cautions were taken to elicit similar amountsof leg motion in the recorded and live displaysby placing infants in similar viewing conditionsduring both, amount of leg motion may nev-ertheless have varied systematically across thetwo displays. Infants may have become habit-uated to the testing situation more rapidly be-cause of the additional 4-min recording periodand thus displayed lecss leg motion on the livescreen during testing. On the other hand, theymay have become less wary of the experimentalsituation, and by the testing session felt morefree to explore and move about. Casual obser-vation, however, revecaled no systematic differ-ences in amount of leg motion displayed onthe live and recorded screens. Another possibleinterpretation is that infants may have habit-uated to the live displays during the recordingsession and thus looked more to the recordeddisplay during testing than they might other-wise. However, this interpretation requires thatinfants nevertheless discriminate between thetwo kinds of displays during the test sessionon the basis of contingency. Taken togetherwith Experiments 1 and 2, these results providestrong evidence for the interpretation that in-fants detect invariant relations between pro-prioceptive and visual information for motion.They also are consistent with Lewis andBrooks-Gunn's (1979) proposal that early self.perception may be based on the detection ofcontingent relations between vision and pro-prioception.

Another interesting question remained un-answered, however. Why did infants show theirdiscrimination of the contingent and noncon-tingent displays in all three experiments bylooking preferentially to the noncontingentone? One possibility is that infants found the

970

LORRAINE E. BAHRICK AND JOHN S. WATSON

live image aversive because the image of theirleg motion was displaced in space and picturedon a television screen. On the other hand, theinformation provided by the contingent displaywas completely redundant with the proprio-ceptive information available to the infant.Thus the live display, without eye contact pro-vided by facial stimuli, may seem boring andlack novelty for the infant. If infants did indeedfind the contingent image redundant and bor-ing, it seemed likely that prior to 5 months ofage, there ought to be an age when infants be-gin learning about the nature of contingency,and find these relations interesting to observe.During this earlier period, they might spend agreater amount of time viewing the contingentdisplay of self and testing out the contingentproperties of the image. Thus a fourth exper-iment was conducted to determine whetheryounger infants would spend more time look-ing to a live display of their own leg motionthan to a noncontingent peer display.

Experiment 4

In this study, we chose to replicate Experi-ment 2 with 3-month-old infants. Research oncontingency learning has estimated that thiscapability appears by the age of 3 months orearlier (e.g., Rovee-Collier & Gekoski, 1979;Watson & Ramey, 1972). Also, using facialstimuli, Field (1979) found that 3-month-oldsshowed greater visual fixation to their mirrorimage but more reaching and vocalizing in thepresence of a peer. We thought that infants ofthis age might also discriminate our nonfacialstimuli, and because at this age they may belearning about the nature of contingency, theymay spend a greater amount of time fixatingthe contingent image of self than the noncon-tingent image of the peer.

MethodTwenty-nine 3-month-olds were'tested, 16 girls and 13

boys. Their mean age was 99 days (SD = 5 days). Twoadditional subjects were excluded from the study becauseof excessive fussiness. The apparatus and procedures wereidentical to those used in Experiment 2. Subjects viewedthe live display of their own legs side by side with those ofthe peer for a 4-min period and wore the occluding bib.

Results and Discussion

Overall, subjects looked to the noncontin-gent display of the peer .47 of the time. This

proportion was not significantly different frortichance.

There was no significant side preferenc~,/(27) = 0.772, p > .05, nor were there anysignificant sex differences in looking to thenoncontingent display, /(27) = 1.57, p > .05,An ANOVA revealed no significant trends ovetthe four I-min periods in looking to the non-contingent peer image, F(3, 84) = 1.98, P >.05; however, there was a nonsignificant ten.dency for infants tp spend more time watchingthe noneontingent film toward the end of th~4-min session (.46 in Min I vs. .54 in Min 4).

There was, however, some suggestion ofbimodality in the distribution of lookingproportions for the 3-month-old subjects, asshown in Figure 2. Looking proportions fot

.only three subjects fell near the mean, between.4 and .6, whereas proportions for 14 subjectsfell below .4, and 12 fell above .6. This diS"-tribution is reflected in the larger standard d~-viation for 3-month-olds (.281) than for 5.month-olds in Experiment 2 (.163). A non.parametric test, the Moses test of extremereactions (Siegel, 1956) was conducted to d~termine whether these two samples were drawnfrom populations whose distributions differedsignificantly, one ~xhibiting a more centralizeddistribution, and the other a more dispersedone. Results of the test indicated that the 3-month-olds did indeed show looking propor~tions that were significantly more extreme thaQthose of the 5.month-olds in Experiment 2(p < .005). These results suggest that some 3-month-olds predominately watched the peerdisplay, whereas others predominately watchedthe contingent display of self. Although it ispossible that this reflects only an alternatingpattern of side preference, it may be that 3..month-olds are in the process of making atransition from greater interest in self and thecontingency provided by a live video image ofself, to greater interest in other and the poten.tial for social interaction.

Although these results, taken together withthose of the 5-month.olds, may reflect a de-velopmental trend in infants' ability to detectperfect contingent relations provided by avideo image of the self, it is alS9 possible thatthey reflect a trend in the development ofpref-erence for this kind of perfect contingency. In-fants' preference for invariant relations spec-ifying self may decline with age as they becomemore socially oriented. This interpretation is~

971PROPRIOCEPTIVE-VISUAL CONTINGENCY

Frequency

8

,81.1.00

.00-.20 21-.40 41-.60 .61-.80

Proportion of total looking time to the non-contingent display

Figure 2. Experiment 4: Distribution of looking proportions for the noncontingent display.

consistent with the finding that 3-month-oldsshow significantly more extreme looking pro-portions than do 5-month-olds.

General Discussion

Five-month-old infants show differential vi-sual fixation to a contingent video image oftheir legs moving and a noncontingent imageof a peer's legs moving. They showed this dis-crimination under conditions in which poten-tial confounds of eye contact, eye motion, andgeneral body motion variability were elimi-nated. Eye contact and eye motion were elim-inated by using non facial stimuli, that is, im-ages of an infant's legs moving. Potential dis-crimination on the basis of general bodymotion was ruled out by using a yoked controldesign. Each subject's live image was recordedand later served as the noncontingent imagefor the next infant, thus controlling for amountof general body motion across subjects.

Purely visual characteristics that might havedistinguished the contingent and non con tin-gent video displays were also controlled. InExperiments I, 2, and 4, they were partiallycontrolled by fitting all infants with yellowbooties. In Experiment 3, they were eliminatedaltogether by presenting a live image along witha recorded image of self for discrimination.

Even in the absence of distinguishing visualcharacteristics, infants showed a robust dis-crimination of the contingent and noncontin-gent displays.

And finally, Experiments 2, 3, and 4 intro-duced a control procedure to determine thenature of the contingent relations detected byinfants. Did infants detect intramodal relationsbetween the sight of their moving legs and thevideo display of that motion, or did they detectintermodal relations between proprioceptivestimulation from their moving legs and thevideo display of that motion? To distinguishbetween these two alternatives, the infants' di-rect visual access to their own bodies was oc-cluded, eliminating the possibility of an intra-modal visual-visual comparison. Discrimi-nation of the contingent and noncontingentdisplays by 5-month-olds was evident evenwhen infants wore an occluding bib. Therefore,these infants must have detected contingencyby perceiving the invariance between the pro-prioceptive stimulation from their own motionand the visual stimulation from the live videodisplay of that motion. This early detection ofproprioceptive-visual invariants is consistentwith Lewis and Brooks-Gunn's (1979) pro-posal that early self-perception may initiallydevelop through the detection of these kindsof intermodal relations. It is also consistent

972 LORRAINE E. BAHRICK AND JOHN S. WATSON

with Meltzoffand Moore's (1983a, 1983b) in-terpretation that neonatal imitation is basedon the detection of proprioceptive-visual re-lations.

There is a perfect correspondence betweenthe visual and proprioceptive stimulation fromone's own body motion. When the infantwatches his or her moving body, the visual andproprioceptive information for that motion areunited by a common temporal relationship,trajectory or direction of motion, and intensityof motion. Changes within each of these di-mensions are invariant over vision and pro-prioception. For example, the infant can bothfeel and see the increasing intensity of his orher leg kicking. In the present studies, infantsmay have detected the contingency betweenproprioceptive information for motion and thevisual display of that motion on the basis ofany or all of these kinds of relations.

Research on intermodal perception in in-fancy has found that infants by 5 months ofage are capable of detecting these invariant re-lations across other modalities. Sensitivity hasbeen shown across vision and audition to var-ious kinds of temporal invariants (e.g., Allen,Walker, Symonds, & Marcell, 1977; Spelke,1979; Walker, 1982), including the temporalmicrostructure specifying object substance(Bahrick, 1983). Infants also detect audio-vi-sual invariants between speech sounds and themouth shape that produces them (Kuhl &Meltzoff, 1984). Across the visual and tactilemodalities infants have shown sensitivity toinvariants of shape, texture, and substance(e.g., Gibson & Walker, 1984; Meltzoff & Bor-ton, 1979), as well as for trajectory or directionof motion (Aitken, 1983). Given that infantscan detect these relations across the visual andacoustic, or visual and tactile modalities, itmight well seem reasonable that they wouldalso be capable of detecting these invariants

,across vision and proprioception. The presentresearch supports this inference.

There are two possible interpretations of thepresent results demonstrating discriminationof the contingent and noncontingent displayson the basis of proprioceptive-visual invariantdetection by 5-month-old infants. One inter,:pretation is that although infants detectedinvariant proprioceptive-visual informationspecifying the self, they did not, in fact, per-ceive this information to specify the self.Rather, they simply responded to the invariant~

without detecting its affordance. The utterna.tive interpretation, consistent with Gibson's(1969) view, is that detection of this proprio-ceptive-visual invariant entails perception ofthe self from the beginning. The present resultscannot distinguish between these two alter-natives. Obtaining a definitive answer to thisquestion provides a challenge for future re-search. Nevertheless, the present results indi-cate that 5-month-old infants possess percep-tual capabilities that are fundamental to theperception of self.

Another finding of these studies that de-serves further attention is the fact that infantsshowed their discrirriination of the contingentand noncontingent displays by preferentialfixation of the noncontingent display. Thisreplicates Papousek and Papousek (1974), whopresented facial stimuli to 5-month-olds. Thispreference for noncontingent stimulationspecifying other over perfectly contingentstimulation specifying self may be interpretedas part of the infant's development ofincreas-ing responsiveness to social stimulation or toexternal stimulation in general. In the outsideenvironment and social world, none of our ac-tions produce a perfectly contingent and con-comitant set of transformations like thosespecifying self-motion. When the infant batsa mobile, the mobile moves, but in a patternand rhythm somewhat different from that in-fant's own actions. When the infant knocksover a toy, it falls and has a motion of its own.And when the infant smiles, the mother re-sponds in her way, by vocalizing or approach-ing or engaging in game-like interactions.These kinds of contingent relations are not in-variant with the infant's behavior. They areimperfect.

Watson (1972, 1979) proposed that the hu-man infant's concept of a social object is ini-tially formed through experience with this kindof imperfect contingency. This hypothesisprovides a theoretical context for the presentresults. Stimulation that is perfectly contingentor invariant with the infant's behavior mayspecify self, whereas stimulation that is im-perfectly related to the infant's behavior mayspecify other, quite early in infancy.

Experiment 4 suggests the possibility thatthe relative preference for the noncontingentdisplay of other over the perfectly contingentdisplay of self may be reversed prior to the ageof 3 months. Although 3-month-olds showed

973PROPRIOCEPTIVE-VISUAL CONTINGENCY

no significant preference for the image of selfover peer, they did show significantly more ex-treme looking proportions than did the 5-month-olds. Their looking proportions werebimodally distributed, whereas those of the 5-month-olds were more centrally distributed.These findings, along with those of Field(1979), that 3-month-olds look more to theirmirror image than to a peer, suggest the pos-sibility that the age of 3 months may representa transition period in which attention shiftsfrom perfect contingency specifying self to im-perfect contingency specifying other.

The results of these studies taken togetherdemonstrate that information provided by afive video image of one's body motion can beperceived through the detection of invariantjntermodal relations between visually andproprioceptively specified motion. Five-month-olds prefer to view a noncontingentdisplay of self or peer over a perfectly contin-gent display of their own motion. The detectionpf proprioceptive-visual invariants may befundamental to the infant's perception of selfand may underlie the development of visualself-recognition in infancy.

ReferencesAitken. S. (1983). Infant response to amodal information

on approach. Paper presented at the biennial meetingof the Society for Research in Child Development, De-troit.

Allen, T. W., Walker, K., Symonds, L., & Marcell, M.(1977). Intrasensory and intersensory perception oftemporal sequences during infancy. Developmental Psy-chology, 13.225-229.

Bahrick, L. E. (1983). Infants' perception of substance andtemporal synchrony in multi modal events. Infant Be-havior and Development, 6, 429-451.

Power, T. G. R., Broughton, J. M., & Moore, M. K. (1970).Demonstration of intention in the reaching behavior ofneonate humans. Nature, 228. 679-680.

Butterworth, G., & Cicchetti, D. (1978). Visual calibrationof posture in normal and motor retarded Down's syn-drome infants. Perception, 7. 513-525.

Butterworth, G., & Hicks, L. (1977). Visual proprioceptionand postural stability in infancy. Perception, 6, 255-262.

field, T. (1979). Differential behavioral and cardiac re-sponses of 3-month-old infants to a mirror and peer.Infant Behavior and Development. 2. 179-184.

Gibson, E. J. (1969). Principles of perceptual learning anddevelopment. New York: Appleton-Century-Crofts.

Gibson, E. J., & Walker, A. S. (1984). Development ofknowledge of visual and tactual affordances of substance.Child Development, 55, 453-460.

Gibson, J. J. (1966). The senses considered as perceptualsystems. Boston: Houghton-Mimin.

Gibson, J. J. (1979). The ecological approach to visual per-ception. Boston: Houghton-Mimin.

Hofsten, C. von. (1979). Development of visually guided

reaching: The approach phase. Journal of HumanMovement Studies. 5. 160-178.

Hofsten, C. von. (1983). Catching skills in infancy. Journalof Experimental Psychology: Human Perception and

Performance. 9. 75-85.Jouen, F. (1984). Visual-vestibular interactions in infancy.

Infant Behavior and Development. 7. 135-145.Kuhl, P. K., & Meltzoff, A. N. (1984). The intermodal

representation of speech in infants. Infant Behavior andDevelopment. 7.361-381.

Lee, D. N., & Aronson, E. (1974). Visual proprioceptivecontrol of standing in human infants. Perception andPsychophysics. 15. 529-532.

Lewis, M., & Brooks-Gunn, J. (1979). Social cognitionand the acquisition of self New York: Plenum Press.

McDonnel, P. M. (1975). The development of visuallyguided reaching. Perception and Psychophysics. 18. 181-185.

Meltzoff, A. N., & Borton, R. W. (1979). Intermodalmatching by human neonates. Nature. 282. 403-404.

Meltzoff, A. N., & Moore, M. K. (1983a). The origins ofimitation in infancy: Paradigm, phenomena, and the-ories. In L. Lipsitt & C. Rovee-C-ollier (Eds.), Advancesin infancy research (Vol. 2, pp. 265-30 I). Norwood, NJ:Ablex.

Meltzoff, A. N., & Moore, M. K. (1983b). Newborn infantsimitate adult facial gestures. Child Development. 54.702-709.

Papousek, H., & Papousek, M. (1974). Mirror-image andself recognition in young human infants: A new methodof experimental analysis. Developmental Ps)lChobiology.7, 149-157.

Rovee-Collier, C., & Gekoski, M. J. (1979). The economicsof infancy: A review of conjugate reinforcement. InH. W. Reese & L. P. Lipsitt (Eds.), Advances in childdevelopment and behavior (Vol. 13 pp. 195-255). NewYork: Academic Press.

Rovee, C. K., & Rovee, D. T. (1969). Conjugate reinforce-ment of infant exploratory behavior. Journal of Exper-imental Child Psychology, 8. 33-39.

Sameroff, A. J. (1971). Can conditioned responses be es-tablished in the newborn infant? Developmental Psy-chology, 5. 1-12.

Siegel, S. (1956). Nonparametric statisticsfor the behavioralsciences. New York: McGraw-Hill.

Siqueland, E. R., & DeLucia, C. A. (1969). Visual rein-forcement of nonnutritive sucking in human infants.Science. 165. 1144-1146.

Spelke, E. S. (1979). Perceiving bimodally specified eventsin infancy. Developmental Psychology. 15.626-636.

Walker, A. S. (1982). Intermodal perception of expressivebehaviors by human infants. Journal of ExperimentalChild Psychology, 33. 514-535.

Watson, J. S. (1972). Smiling, cooing, and "the game,"Merrill-Palmer Quarterly, 18. 323-339.

Watson, J. S. (1979). Perception of contingency as a de-terminant of social responsiveness. In E. B. Thoman(Ed.), The origins of social responsiveness (pp. 33-64).New York: Erlbaum.

Watson, J. S., & Ramey, C. T. (1972). Reactions to responsecontingent stimulation in early infancy. Merrill-PalmerQuarterly, 18.219-227.

Received May 23, 1984Revision received March II, 1985.