Poster group
Details of individual items:
poster
In the Visual Expectation Paradigm, young infants have demonstrated with their eye movements that they are able to abstract a rule that governs the appearance and disappearance of predictable dynamic visual events. Infants combine exogenous perceptual information with endogenous knowledge of the spatial locations of forthcoming visual targets to anticipate events and to enhance their reactions to them. The present study explores whether 13-week-old infants utilize the spatial predictability of the visual targets to program the amplitude of their saccadic eye movements. Infants viewed computer-generated shapes (4.5deg overall size, vertical movement of 4.4deg/sec) that appeared 5.7deg the right (R) or left (L) of visual center for 700ms with a 1000ms interval between each target. The image of the infant's right eye was videotaped with the corneal reflection technique. A customized eye-tracker sampled the vertical and horizontal position of the pupil and corneal spot reflection in every 16.7ms field on the videotape. In the Forward condition, 24 infants saw 30 visual targets that appeared in a spatially irregular (IR) sequence, followed by 50 targets presented in a spatially alternating (ALT) sequence. In the Reverse condition, 12 infants viewed 30 targets in the ALT sequence, followed by 50 pictures in the IR sequence. In a third, Asymmetry condition, 14 infants saw 10 pictures in an IR sequence, followed by 70 pictures in a 2-1 asymmetric (L-L-R) sequence. An incorrect anticipatory shift occurred when an infant made a saccade to the R location after the first L target was extinguished and before the second L target appeared.Forward and Reverse condition results: Mean amplitude (+SE) for reactive and anticipatory saccades are shown in Table 1. Reactive saccades during the ALT sequence were significantly larger than reactive saccades during the IR sequence, regardless of whether the ALT sequence preceded or followed the IR sequence. Anticipatory saccades initiated during the ALT sequence were significantly larger than anticipatory saccades during the IR series when the IR trials occurred before the ALT trials (Forward condition). Although the direction of the effect was the same when the ALT trials occurred first (Reverse condition), the difference was not significant. However, the effect was significant (ALT > IR, p0.04) when anticipatory saccades in the Forward and Reverse conditions were combined.Asymmetry condition results: The mean amplitude (+SE) of correct anticipatory shifts (11.61+09deg) was significant larger (p0.05) than that for incorrect anticipatory shifts (9.01+0.6deg).These findings indicate that young infants incorporate the predictable spatial location of the target in programming saccadic amplitude for both their endogenous (anticipatory) and exogenous (reactive) saccades. However, when the spatial location of visual events is unpredictable, there is a decrease in average saccadic amplitude. This situation occurred for anticipatory and reactive saccades for targets in the IR sequence and when infants made incorrect anticipatory saccades. This convergent evidence strongly suggests that, at an early age, infants use knowledge of cognitive events to exert top-down influence on the bottom-up process of saccade programming. Supported by NIMH 2 T32-MH15442-19 and NIH 1 F32 HD08439-01. Table 1. Mean amplitude and velocity (+SE) of saccades during the Irregular and Alternating sequences. AMPLITUDE Irregular AlternatingReactiveForward (n24) 11.36+0.4deg 12.04+0.4deg (p0.04)Reverse (n12) 10.35+0.5deg 12.21+0.5deg (p0.006)AnticipatoryForward (n15) 8.07+0.4deg 10.63+1.1deg (p0.04)Reverse (n12) 9.81+0.7deg 10.34+0.8deg (NS)
poster
Haith, Hazan, and Goodman (1988) demonstrated that young infants rapidly form expectations to predictable visual elements. Subsequent study has indicated that the two measures of expectation (decrease in reaction time (RT) and fixation anticipations) are not highly correlated. Wass, Lewis, and Haith (1998) reported that stimulus onset asynchrony (SOA) affected RT whereas cycle time (the time required for a sequence to repeat) affected anticipations. However, this previous study did not address which aspect of cycle time (stimulus duration, interpicture interval (IPI), or the number of events per cycle) affected anticipations. The current study was designed to disentangle the effects of SOA and IPI on expectation measures.Method Sixty-one 13 week-old infants viewed a series of pictures in the VExP. All infants saw a repeating pattern of two pictures on the left and one on the right (2-1). Infants participated in one of four conditions that had varying temporal parameters (see table 1). Results The RTs for the 2000ms SOA groups were significantly greater than for the 1600ms SOA groups F (1,51) 10.2 p < 0.01. SOA did not affect the percentage of correct or incorrect anticipations reliably. IPI duration did not affect RT or the percent correct anticipations. IPI did however strongly impact the percent of incorrect anticipations. The 1175 ms IPI groups had significantly more incorrect shifts than did the 900 ms IPI groups F (1,55) 22.98 p < 0.01. IPI also significantly impacted the pattern of anticipations to each of the three locations (home 1, home 2, and target). The infants in the 900 ms IPI groups anticipated the most to the home 1 location, the least to the home 2 location (incorrect anticipations), and intermediately to the target location. The 1175 ms IPI groups demonstrated a different pattern in which the greatest percentage of anticipations occurred to the home 2 location and equal but lesser percentages to the home 1 and target locations. Analysis of the anticipation latency also yielded some interesting results. The infants in the 900ms IPI groups were more likely to produce anticipations near the time of stimulus onset for the home 1 and target locations than they were for the home 2 location (an incorrect anticipation). Infants in the 1175ms IPI groups were only able to cluster their anticipations close to the time of onset of the home 1 location. Discussion These findings suggest that the SOA value affects RT whereas the IPI duration affects anticipations. It also appears that at a 900 ms IPI babies are able to estimate when the pictures will appear in both locations, and they are able to keep in mind the sequence of the pictures. However, at an 1175ms IPI babies are only able to estimate the time of onset of the picture in the more frequent location. The high percentage of incorrect shifting (home 2) for the babies in the longer IPI conditions indicates that there is a relatively brief time limit on how long infants can maintain the entire sequence of pictures in memory.Funded by NIMH 535242
poster
Mareschal, Plunkett and Harris (1999) have suggested that early indevelopment infants cannot access object information processed down thedorsal and ventral cortical streams simultaneously. We test this hypothesisdirectly by using a preferential looking paradigm to exploring 4-, 6- and9-month-olds responses to a violation of either feature information,location information, or the conjunction of feature and location informationin occluded faces. In familiarisation trials infants saw 2 square occluders positioned side byin the middle of the monitor with a 1.5-cm gap separating them. A faceappeared from behind one of the randomly selected occluders. The facetranslated until it reached the edge of the monitor. It remained stationaryfor 5 seconds then returned back behind the occluder from which it hadcome. As soon as it had disappeared, a second face appeared from behind thesecond occluded and translated to the opposite edge of the monitor. After 5seconds this face also returned behind its occluder. No face ever appearedin the gap separating the two occluders. Test trials began in the same wayas the familiarisation trials. However, after the second face had returnedbehind its occluder, the occluders moved upwards to reveal one of fourthings. In the baseline condition, the occluders moved up to reveal the sametwo the original two faces, each behind the appropriate occlude. In thesurface feature condition (SF) the occluders revealed one familiar facebehind an occluder and one novel faces behind the other occluder. In thespatial temporal condition (ST) the occluders revealed no face behind oneoccluder and two faces behind the second occluder. Finally, in theconjunction condition the occluders revealed the original two faces, butthey had been swapped around such that each one was behind the inappropriateoccluder. The faces were randomly selected from a pool of 10 female faceswith similar silhouettes and presentation order was determined by areplicated Latin-square design.The results collected so far are based on 15 4-month-olds, 20 6-month-oldsand 6 9-month-olds. The 4-month-olds looked significantly longer in the SFcondition than the baseline condition (t(14)3D3.27), p<.004) and show nosignificant differences in their looking times during the other testconditions. In contrast, the 6-month-olds show no significant differences intheir looking times in any test conditions. However, further investigationsuggests that there are two groups of 6-month-olds. One group (n3D11) lookedsignificantly longer in the ST condition than the baseline condition(t(10)3D3.532, p<.002) with no significant differences in looking time in theother test conditions. The other group (n3D9) looked significantly longer inthe SF (t(8)3D2.881), p<.009) and the conjunction condition (t(8)3D2.934),p<.008) than the baseline condition with no significant difference betweenlooking in the baseline and ST conditions. Finally, initial results with9-month-olds suggests that that these infants look equally in the SF,conjunction and baseline condition and less in the ST condition. Theseresults are consistent with the suggestion that young infants only accessinformation in one cortical stream when responding to violations of hiddenobject information.
poster
Hood (1995) examined young children's ability to predict a location of a ball which was dropped down through a opaque curved tube, and suggested that gravity rules are dominant in understanding objects' movement for 2-3 years old children. On the other hand, in the field of infant research, it has been reported that concept of gravity is not core in human knowledge of physical world, and that young infants can not reason about the behavior of hidden falling objects (Spelke, 1992). These findings imply that the gravity rules should become dominant at some age point before 2 years old. When is it? To answer this question, we conducted an experiment similar to Hood (1995) with infants.MethodParticipants: Sixty-five infants ranging in age from 4 months 2 days to 13 months 15 days (mean 8 months 1 days) participated.Apparatus: On a white back wall (314 mm x 232 mm) of an apparatus, two blue tubes whose diameter was 53 mm were attached vertically; one was long and curved, another was short and straight. The lower end of the curved tube was swerved 103 mm horizontally from the upper end. The short tube located just underneath the upper end of the curved tube. As a falling object, a red ball whose diameter was 30 mm was used.Event: Participants saw two events. In a possible event, the ball was dropped into the upper end of the curved tube and came out of its lower end. The ball seemed to go through the curved tube. In an impossible event, the ball was dropped into the upper end of the curved tube and came out of the lower end of the short tube. The ball seemed to fall straight down, violating the continuity rule.Procedure: Each infant saw the possible and impossible events in turn for three times. Infant's looking times after the ball came out was recorded.Results and Discussion The difference in the mean looking times for the possible and impossible events was calculated for each infant. While infants below 5.5 months tended to look at the two events equally, infants above 5.5 months tended to look longer at the possible than at the impossible events. This was supported by a result of 2 (Age) x 2 (Event) x 3 (Presentation) ANOVA (F(1,63)7.53, p<.01) . This result means that infants above 5.5 months anticipated the ball to fall straight down even though the entrance and exit were never connected. It suggests that, like 2-3 years old children, infants above 5.5 months are governed by naive theories of gravity.Reference Hood, B. M. (1995). Gravity rules for 2- to 4-year olds? (1995). Cognitive Development, 10, 577-598. Spelke, E. S., Breinlinger, K., Macomber, J., and Jacobson, K. (1992). Origins of knowledge. Psychological Review, 99(4), 605-632.
poster
In a previous study investigating young children's spatialreasoning abilities, 2-, 2.5-, and 3-year-old children searched for a ballthat rolled behind an occluder with four doors. The ball was stoppedbehind the occluder by a partially-visible barrier wall that could beplaced at any of the four doors. Children could find the ball by openingthe correct door. In our original study, 3-year-olds were the only groupto perform consistently above chance. The present study investigated whether reducing the working memorydemands of the task improved performance in the younger children. Memorydemands were reduced by increasing the saliency of the partially-visiblewall used in the first study. Twenty-one 2.5-year-old children werestudied (10 males, 11 females). The primary finding was that the addedvisual saliency of the wall resulted in improved performance, but only forthe females. The proportion of trials where the first reach was correctincreased to almost 50% (chance performance .25), and three femalesresponded perfectly over 12 trials considering first and second reachattempts. The males still performed near chance. Because successful performance could be due to associating theball with the wall without understanding why the ball stopped, wepresented 4 additional trials where two walls were placed behind theoccluder. If the children understood the physics of the situation, theywould know that the ball should stop at the first wall and theirperformance should be similar on one- and two-wall trials. If thechildren rely on associating the sight of the wall with a nearby door whensearching for the ball on one-wall trials, performance should besignificantly disrupted on two wall trials. The results show that thefemales had difficulty on the two-wall trials. The females meanproportion correct on the first reach decreased to .39 on two-wall trials. The results of the studies suggest that the females used thevisible wall as an associative cue, but they did not fully understandwhere the ball was located and why it was located there. Using the wallas an associative cue allowed the females to perform well when one wallwas located behind the doors. However, when there were two walls, theirstrategy of opening the doors near the wall did not work.
poster
What do infants know about complex support events involving more than2 objects? We examined what information 13.5-month-old infants usedto determine whether a stack of 3 boxes was adequately orinadequately supported.Previous work (Baillargeon, 1995) demonstrated that from about 12.5months of age infants can use the proportional distribution of anobject to determine if it is adequately supported. At that ageinfants look longer at an asymmetrical L-shaped box 'supported' byits smaller (Figure 1a) as opposed to its larger end (Figure 1b).The current series of experiments examines whether slightly olderinfants are able to use the proportional distribution informationwhen multiple boxes are being supported.In a familiarization trial, infants were shown a box that would serveas the support (bottom box) and two boxes to be supported (middle andtop boxes). A gloved hand moved the top box independently of themiddle box to show that the two boxes were separate (Figure 2a).Test trials alternated between a full support (Figure 2b) and apartial support display (Figure 2c). If infants can make use ofproportional distribution information integrating multiple objects(as adults do), the partial support display should be surprisingcompared to the full support display. The infants tended to lookequally at the two test displays.Perhaps the infants found the test displays complicated and didn'tknow what to expect. To address this, different infants were shownthe same test displays, but were familiarized with a gloved handgrasping and moving the top box with the middle box attached to it,so they moved as a single object (Figure 3). If infants believed thetop and middle boxes to be a single object they should use theirknowledge of proportional distribution and look longer at the partialsupport display (which was conceptually the same as the L-box displaydescribed above). Infants looked reliably longer at the partial thanat the full support display. So if infants thought of the top andmiddle boxes as a single object they used proportional distributioninformation to reason about support, but if they thought of the twoboxes as separate objects they did not seem to use that information.Did infants fail to use the proportional distribution informationwith 3 boxes because they were attending to only one of the boxesbeing supported? Two news groups of infants were familiarized withthe top and middle boxes as separate objects. Test events alternatedbetween a full support display (as before) and one of two partialsupport displays. In the middle-violation display the bottom box wassupported by only 25% of its bottom surface which should violate theexpectations of infants at this age (Baillargeon, 1995). The top boxwas fully supported on the middle box (Figure 4a). In thetop-violation condition the top box was supported by only 25% of itsbottom surface while the middle box was supported 50% (Figure 4b).Neither the infants in the bottom-violation nor those in thetop-violation conditions looked longer at the partial supportdisplay.In sum, though infants realized that a single asymmetrical objectshould fall when its more massive end was not supported, they did notuse this knowledge in reasoning about a stack of 3 boxes. Infantslooked longer at the partial support display when the display wascomposed of 2 boxes, but did not look longer at the same test displaywhen they had been shown that it was composed of 3 boxes.Furthermore, infants did not reason about the 3 box display in apiecemeal fashion. Infants failed to detect single box violations(that they can detect in simple 2 box displays) when the displaycontained 3 boxes. Further experiments are planned to determine whyinfants succeed in reasoning about stacks of 2 but not 3 boxes.Figure 1. Infants expected the L-shaped box to fall in (a), but notin (b).a. b.Figure 2. During familiarization, the top box was moved separatelyfrom the middle box. The bottom box remained sitting to the left(a). In the full support test display (b), the top and the middleboxes were each supported 100% and as a unit they were adequatelysupported. In the partial support display (c), the top (100%) andthe middle boxes (50%) were each adequately supported, but as a unitthey were not.a. b. c. Figure 3. During familiarization, the top and middle boxes weremoved as a unit. The test displays were the same as in the previousexperiment (2b-c above).Figure 4. In the middle-violation display (a), the middle box wasnot adequately supported (25%) though the top box was (100%). In thetop violation display (b), the top box was not adequately supported(25%) though the middle box was (50%). Each of these partial supportdisplays was compared to a full support display (as in 2b).a. b.
poster
Recent research suggests that infants view occlusion and containment asdistinct event categories, and reason and learn separately about these twocategories (e.g., Hespos & Baillargeon, 1999). In one experiment, forexample, 4.5-month-old infants were found not to be surprised when a tallobject was fully lowered inside a short container. However, infants weresurprised (as indicated by reliably longer looking) when the object wasfully lowered (a) behind a short container or (b) behind a short occluder(the back and bottom of the container were removed to form a roundedoccluder). These results suggested that, by 4.5 months of age, infants canreason about height in occlusion but not in containment events. Subsequentexperiments showed that it is not until about 7.5 months of age that infantssucceed at reasoning about height in containment events (e.g., Hespos,1998).Why do infants view occlusion and containment as separate event categories?Although in some cases occlusion and containment outcomes are similar (as inthe experiment just described), in many other cases they are not. Forexample, an object that has been lowered inside a container will move withit when lifted, whereas an object that has been lowered behind an occluderwill not. If it is true that such differences (i.e., the consequences ofoccluders and containers' displacements) are implicated in the formation ofinfants' event categories, then one might expect infants to treat events inwhich objects become contained or covered as again belonging to differentcategories. Objects that are covered, like objects that become occluded,typically remain stationary when covers are lifted.The present experiment compared 8- to 9-month-old infants' reasoning aboutthe height of objects and containers or covers. The infants were assignedto a container or a cover condition. The infants in the container conditionsaw two test events. At the start of each event, a tall object stood nextto a container (to facilitate height comparisons). Next, a hand lifted theobject and lowered it inside the container until it disappeared from view.In one event, the container was taller than the object (tall event). In theother event, however, the container was considerably shorter than theobject, so that it should have been impossible for the object to becomefully hidden inside the container (short event). The infants in the covercondition saw similar events, with two exceptions: first, the containerswere turned upside down to form covers; and second, the hand lowered thetall and short covers over the object, rather than lowering the objectinside the containers.The infants in the container condition looked reliably longer at the shortthan at the tall test event, but the infants in the cover condition tendedto look equally at the two test events. These results suggest that just asthere is a dE9calage in infants' reasoning about height with occluders andcontainers (e.g., Hespos, 1998; Hespos & Baillargeon, 1999), there is also adE9calage in infants' reasoning about height with containers and covers.These findings are important for two reasons. First, they provide furtherevidence that infants 'sort' physical events into narrow event categories,and reason and learn separately about each category. Second, the presentresults support the hypothesis that attention to the consequences ofobjects' displacements plays a significant role in infants' decisions aboutwhat events belong to the same and what events to different categories.
poster
Recent research (e.g., Kaufman, 1997) suggests that infants aged 5 monthsand older distinguish between inert and self-moving inanimate objects andhold different expectations for objects from these two categories, at leastfor certain physical events. In one experiment, for example, infants wereassigned to an inert or a self-moving condition. The infants in the inertcondition saw two test events. In one (near-wall event), a hand hit a smallbox. The hand then travelled to the right, hit a wall partition that filledthe right end of the apparatus, and then (as though bouncing back) returnedto its starting position. In the other event (far-wall event), the boxreversed its trajectory in the same exact location but this timespontaneously, because the wall partition was much narrower so that the boxnever reached it. The infants in the self-moving condition saw the sametest events as the infants in the inert condition, except that the boxinitiated its own motion; the hand lay flat on the apparatus floorthroughout the events. The infants in the inert condition looked reliablylonger at the far- than at the near-wall event, whereas those in theself-moving condition looked about equally, and equally low, at the twoevents. These and control results suggested that the infants (a)categorized the box as inert or self-moving based on how its motion wasinitiated and (b) were surprised to see the inert but not the self-movingbox reverse its trajectory.The present research built on these initial results as well as prior resultsthat even young infants expect an inert object to be displaced when hit(e.g., Kotovsky, 1992). The experiment asked whether 6-month-old infantsalso expect a self-moving object to be displaced when hit, or whether theyadmit that self-moving objects may not move when hit (perhaps because theyare capable of resisting other objects' impact; e.g., Leslie, 1995).The infants were assigned to an inert or a self-moving condition. Theinfants in the inert condition sat in front of a large screen. Duringfamiliarization, a red box emerged from the right edge of the screen,travelled to the right until it hit a wall partition, and then (as thoughbouncing back) returned behind the screen. The infants in the self-movingcondition saw a similar event except that the box reversed spontaneously:the wall partition was much smaller so that the box never reached it. Afterfamiliarization, the screen was removed, and the infants in the twoconditions saw the same test event: a hand hit the box, which remainedstationary.The infants in the two conditions tended to look equally during thefamiliarization but not the test trials: during these trials, the infants inthe inert condition looked reliably longer than did those in the self-movingcondition. These and control data suggest that the infants (a) categorizedthe box as inert or self-moving based on how its reversal was effected and(b) expected the inert but not the self-moving box to move when hit.These results confirm the conclusion that infants hold somewhat differentexpectations for inert and self-moving inanimate objects. Together, theevidence that infants view self-moving objects as capable of initiatingtheir own motion, reversing their path, and resisting other objects' impact,raises fascinating questions about how infants conceptualize self-movingobjects (e.g., as possessing internal forces). Additional experiments areplanned to decide among several possible conceptualizations.
poster
Human languages show considerable variability in the kinds of spatialrelationships that they capture, and human children begin to learndifferent systems of spatial terms quite readily (e.g., Choi & Bowerman,1991; McDonough, Choi, & Bowerman, 1999). What are the conceptualunderpinnings of this ability? Do young children have a small stock ofspatial concepts that subsequently are shaped by exposure to the terms oftheir language, or do they have a large array of spatial concepts fromwhich they select those that are appropriate to their language? To begin todistinguish these possibilities, we investigated the origins anddevelopment of a spatial distinction that is marked in some languages(e.g., Korean) but not in the ambient language of our participants(English): the distinction between tight-fitting and loose-fittingcontainment.In the first experiment, we tested 5-month-old U.S. infants' categorizationof tight- vs. loose-fitting containment relations using ahabituation-dishabituation paradigm. First, infants saw a narrowcylindrical object lowered into a series of loose-fitting, medium-sizedcontainers on a series of trials until their looking time declined. Next,the infants were presented with 6 test trials in which the same cylindricalobject was lowered, in alternation, into a 33% wider container (also aloose fit) and into 33% narrower container (tight fit). If infants make alanguage-independent categorical distinction between a tight and a loosefit in a containment event, then they should look significantly longer atthe tight-fit trials. Our results confirmed this prediction with 13 out of15 infants looking longer at the tight-fit test event (p < .05).Because the results from the first experiment could be due to an inherentbias to look longer at tight-fitting events, a second experiment comparedlooking times to the same test events after habituation to a tight-fittingevent in which a medium-sized object was lowered into a medium-sizedcontainer. In our preliminary findings, 7 out of 8 infants looked longer atthe loose fitting test events, providing evidence that infants categorizedthe containment events as tight- or loose-fitting and mapped thecategorical distinction seen during habituation trials onto the events thatthey saw during test trials.These findings support the view that the distinction between tight andloose fitting containment develops prior to and independently of language.More generally, children's ability to learn different sets of overlappingspatial categories may depend more on their ability to select appropriatelyfrom a large set of preexisting concepts than from an ability to shapeinitial concepts to fit the terms and expressions they hear.
poster
The physical world consists of many different kinds of materials, andtheir movements present visual tracking systems with varying degrees ofcomplexity. Consequently, constructing knowledge about the full rangeof physical types poses a hard problem for the developing infant. Thecurrent studies examined some information processing constraints oninfants' abilities to track hidden items of various physical types. Thestimuli were a rigid, cohesive object; a non-rigid, cohesive object; anda portion of non-rigid and non-cohesive substance. It was hypothesizedthat stimulus complexity varies as a function of similarity to rigid,cohesive objects. Three experiments provided 8-month-old infants withvisual and haptic exposure to the stimuli and then examined the changesin infants' ability to track occluded items as a function of stimuluscomplexity and task difficulty.Each infant was allowed to play with one of the three different kinds ofstimuli. Then, the infant saw a series of trials in which pairs of thatitem were hidden, one at a time behind a screen. The screen was thenremoved to reveal a pair of items or a single item. The infant'slooking time to this series of events was recorded. In previousstudies, infants reportedly tracked the number of objects or the amountof hidden material. For example, when two dolls were hidden, infantslooked longer when one doll was revealed than when two dolls wererevealed. That study was purported to be a test of numericalcognition. However, the procedure as applied in these studies dependedon infants' ability to track amounts of hidden material, regardless ofwhether or not those amounts were represented as being constituted ofsome specific number of items.Infants' abilities to track and represent hidden items may depend ontask characteristics as well as stimulus characteristics. Tracking thelocations of a pair of hidden items is easier with two screens than withone. To vary the information processing demands of the task, wemanipulated the extent to which there were spatial cues to the locationsof the hidden stimuli. In some events, items were hidden behind theleft and right sides of a single screen. In other events, the itemswere hidden behind two separated screens.In Experiment 1, 8-month-olds were shown rigid cohesive objects, orportions of non-rigid, non-cohesive sand hidden behind a single screen.Looking times at consistent and inconsistent outcomes were compared.Infants looked longer when the amount of objects was inconsistent withthe hiding event, but not when the amount of sand was inconsistent withthe hiding event. We concluded that infants were better able to trackand represent the hidden objects than hidden portions of the sand. Themovements of the non-rigid, non-cohesive stimulus may have been morecomplex than the object movements, but task demands introduced anadditional layer of complexity that may have masked infants'competence. In Experiment 2, we reduced task complexity by hidingobjects or sand behind two screens instead of one. The findings ofExperiment 1 were replicated, making it less likely that task demandsper se contributed to the failure in the sand condition of Experiment1. We next turned to stimulus complexity. The stimuli in Experiments 1& 2 were distinguished by cohesiveness and rigidity. Thus, eitherproperty may have contributed to the pattern of results in thoseexperiments. In Experiment 3, we presented non-rigid but cohesivestimuli (a flexible object) hidden behind one screen or behind twoscreens. Infants in Experiment 3 looked longer at the inconsistentoutcome in the two-screen case but not in the more difficult one-screencase.It is concluded that infants are capable of tracking and representingoccluded rigid and non-rigid stimuli. Others have reported that8-month-old infants will even track non-cohesive groups consisting ofrigid objects, provided they have enough advance exposure to thosegroups. Therefore, we conclude that rigidity and cohesivenesssimultaneously contribute to complexity, and increase the relativedifficulty of tracking moving portions of substance. We introduce amodified theory of infants' representation of physical events in orderto accommodate non-rigid and non-cohesive entities.
poster
Using looking-time methods, Baillargeon (1986) found that young infantsare surprised by occluded tracking events that appear to be physicallyimpossible. In order to generate fine-grained predictions of infants' eyemovements during these events, a computational model of infant tracking ispresented. The model was first trained to follow a target object in thefamiliarization event studied by Baillargeon (1986; Baillargeon & DeVos,1991), and then tested on the possible and impossible events from the samestudy. There were no differences in the model's tracking behavior duringthe familiarization and possible test events. During the impossibleevent, however, tracking was disrupted as a new pattern of eye movementswas generated. These findings (1) highlight potential differences ininfants' eye movements during possible and impossible events, and (2)suggest the need for microanalytic measures of attention in objectknowledge studies.