Symposium
Chair: Teresa Wilcox
Discussant: Karen Wynn
Over the last several years, there has been considerable debate about the kind ofinformation that infants can use to individuate objects in occlusion events. Much of this researchhas focused on infants' ability to use featural information. There is now substantive evidence thateven young infants can use featural similarities and differences to draw conclusions about thenumber of objects present in an occlusion event. However, there are interesting developmentalchanges in this ability. For example, infants demonstrate sensitivity to form features, such asshape and size, before they demonstrate sensitivity to surface features, such as pattern and color. In addition, whereas infants are quite competent at using featural information to individuateobjects (i.e., to draw conclusions about the number of objects involved in an event), they often failto subsequently identify objects by feature (i.e., to code and represent individuals by feature). These findings, along with many other recent results, indicate that object individuation is acomplex process that we have yet to fully appreciate. The purpose of this symposium is to bring together recent investigations of infants' abilityto use featural information to reason about object identity. An examination of infants' successesand failures, on a variety of tasks, and the developmental trends that are revealed, can be used tobuild a more complete and unified picture of infants' ability to individuate, represent, and identifyobjects within the context of occlusion events. The research presented in this symposium isparticularly exciting because it goes beyond simply telling us whether infants can individuateobjects or not. The present research sheds light on the processes that infants' engage in whenthey attempt to build and use object representations, the nature and content of theserepresentations, and possible mechanisms for developmental change.
Details of individual items:
paper
Recent research indicates that infants as young as 4.5 months of age can use featural informationto individuate objects in occlusion events (Wilcox & Baillargeon, 1998). However, the type of featuralinformation to which infants are sensitive changes with age. For example, by 4.5 months infants use shapeand size, but it is not until 7.5 months that infants use pattern and 11.5 months that they use color, toreason about the number of objects involved in an event (Wilcox, in press). It has been suggested thatthese results reflect infants' bias to attend to form over surface features when reasoning about occlusionevents. What has been left open to interpretation, however, is whether young infants fail to use surfacefeatures because they have not yet learned that pattern and color can be used to individuate objects orwhether they realize that pattern and color can be used to individuate, but fail to do so spontaneously withinthe context of occlusion events. A series of experiments has recently been conducted to test between thistwo possibilities. One set of experiments investigated 7.5- and 9.5-month-olds' sensitivity to color information afterfirst viewing events that highlighted the functional value of attending to color differences. For example, inone experiment infants first saw an event in which one object of a pair (i.e., a green can) pounded a nail;next, they saw the other object of the pair (i.e., a red can) pour salt. These events were then repeated with adifferent set of objects (i.e., a green cup pounded and a red cup poured). After viewing the pound/pourevents, infants' ability to use color to individuate objects in an occlusion event was tested. The resultsindicated that the 9.5-month-olds used the color difference to individuate the objects in the test event. Additional experiments replicated these findings with other color-function pairings, and also revealed that if(a) the actions the objects performed had no functional significance or (b) the infants saw the sameobject pair rather than two different object pairs during the pound/pour events, the infants failed to usethe color difference in the subsequent test event. Finally, 7.5-month-olds also demonstrated success atusing color to individuate objects, but only after viewing three pairs of pound/pour events involving threedistinct object sets. Another set of experiments investigated the development of young infants' sensitivity patterninformation using a similar procedure. In these experiments, 5.5- and 4.5-month-olds saw the eventsdescribed above with one exception: the objects in the pound/pour and test events differed in pattern(i.e., stripes and dots) rather than color. The results indicated that viewing the pound/pour eventsincreased the 5.5-month-olds' sensitivity to pattern differences in the individuation task. However, like the7.5-month-olds in the color experiments, the 5.5-month-olds needed to see three pairs of pound/pour eventswith three distinct sets of objects: seeing two pairs of pound/pour events, or three pairs of pound/pourevents with only two object sets, did not lead to improved performance. Finally, the 4.5-month-oldssucceeded only when they were allowed to view both objects of the pair simultaneously during thepound/pour events. Together, these results suggest that young infants recognize that color and pattern differences cansignal the presence of distinct objects, but that their ability to use this knowledge is limited to moresupportive contexts. In addition, the formation and use of object categories, that include both function andfeature information, play an important role in this process. These results will be discussed within thecontext of current models of object representation and physical reasoning in infancy.
paper
Xu and Carey (1996) showed that by 12 months of age infants can track 2sets of objects, that is, they can keep track of a duck and a truck. Keeping thetwo objects distinct in memory implies representing two sets of one each,rather than a single set of two objects (e.g., two things as opposed to oneduck and one truck). Although it is not possible to conclude directly from Xu& Carey that their 12-month-olds expected a duck and a truck rather thansimply two things, Hall & Leslie (1995) provided evidence supporting astronger interpretation. Hall & Leslie familiarized 12 month olds to a diskand a triangle viewed one at a time. When the screen was removed, babieswere shown either a disk and a triangle(expected) or two disks (unexpected)or two triangles (unexpected). Infants looked longer on the unexpected trialsshowing they had stored in memory at least 'two things different in some way'rather than 'thing and thing.' Stronger evidence yet comes from a study byLeslie, Tremoulet, Krauss, & Mathur (1998) which examined featureconjunction of size and shape. Twelve-month-olds were familiarized to asmall disk and a large triangle removed from behind a screen and shownsimultaneously. Following familiarization, infants were shown the objectsonce more, then the screen was removed. Infants looked longer at aconjunction-switched pair of objects (large disk and small triangle) than atthe control stimulus following simultaneous (but not following sequential)presentation. This means that infants are able to track in memory two sets ofobjects simultaneously, at least so long as each set comprises a single object. We report here a series of looking time experiments testing12-month-old's ability to track sets totaling three or four objects. Infants arerequired to make judgements about the identity of the objects involved, andnot simply the number of objects. For example, infants familiarized with twotoy frogs and one toy fish, are tested with one frog and two fish. Results showthat 12-month-olds have at least two slots available in visual workingmemory but perhaps not more. These studies attempt to bootstrap thequestions of (a) what is the size of infant visual working memory and (b)what kinds of representations do infants construct in working memory.Among the latter questions we address the nature of the object representationand how the numerosity of sets of objects is represented.ReferencesHall, D.G., & Leslie, A.M. (1995). The role of shape and color in infant'stracking of object identity. Society for Research in Child Development,Biennial Conference, Indianapolis, IN, March 1995.Leslie, A.M., Tremoulet, P., Krauss, K.L., & Mathur, N. (1998). Objectidentification by feature conjunction in 12-month-old infants. Abstract.Infant Behavior & Development, 21, 531.Xu, F., & Carey, S. (1996). Infants' metaphysics: The case of numericalidentity. Cognitive Psychology, 30, 111 153.
paper
New and exciting questions are being asked about the natureand contents of infants' representations. Research by Wilcox andBaillargeon has shown thatinfants as young as 5 months of age cantrack identity information. Xu et al. (1996) have shown thatbetween the ages of 10 and 12 months there are developmentalchanges in tracking object identity, influenced by the ability tolabel objects. In this symposium, I would like to present dataon 6.5-month-old infants and their ability (and the limitationsof this ability) to track identity. We used a looking paradigm to test whether 6.5-month-oldinfants could map knowledge about an occlusion event onto acontainment event(occluder-container event). Recent evidencesuggested that infants view occlusion and containment events asbelonging to distinct event categories and reason and learnseparately about these two categories (Hespos & Baillargeon,1999). The infants in the first experiment were assigned to aball-box or a box-box condition (see Figure 1). At the start ofeach test trial in the ball-box condition, a gloved hand lifted aKoosh ball above the right edge of a large screen, and thenreturned it behind the screen. Next, the hand lifted a box abovethe left edge of the screen and again returned it behind thescreen. This entire sequence was repeated three times. Next, thescreen was lowered to the apparatus floor to reveal a rectangularcontainer. The hand lifted the box above the right and then theleft edge of the container's front wall; this sequence wasrepeated until the infant looked away and the trial ended. Theinfants in the box-box condition received similar test trials,except that the ball was replaced by a second, identical box. The performance of the infants in the occluder-containerevent was compared to that of a second group of infants (seeFigure 2) who saw similar ball-box and box-box conditions, withone exception: the container was replaced with a screen(occluder-occluder event). When the large screen was lowered, asecond, smaller screen was revealed that was identical inappearance to the front wall of the container in the occluder-container event. In both the occluder-container and the occluder-occluder events, observers monitored how long the infant lookedafter the large screen was lowered. The reasoning was that the infants in the occluder-containerevent, who saw events from two distinct event categories shouldhave difficulty retrieving their representation of the occlusionevent and mapping it onto their representation of the containmentevent. That is, they should fail to notice that, two distinctobjects (the ball and box) were involved in the occlusion event,and only one object (the box) was involved in the containmentevent. Therefore, no difference should be found between thelooking times of the infants in the ball-box and box-boxconditions for the occluder-container event. In contrast, theinfants who saw the occluder-occluder event, saw events from asingle event category. They should monitor the occlusion event asit continues to unfold, first with the large and then the smallscreen, and they should detect the change in the ball-boxcondition -- the box appearing over the right edge of theoccluder where the ball had previously appeared. The ball-boxinfants would thus look reliably longer than the ball-ballinfants after the large screen was lowered. The results confirmed these predictions: in the occluder-container event, the ball-box and box-box infants tended to lookequally during the test trials; however, in the occluder-occluderevent, the ball-box infants looked reliably longer than did thebox-box infants. Further support for this finding was found in a thirdexperiment where the infants saw events described above exceptthe trajectory was simplified (see Figure 3). The infants in theball-box condition looked reliably longer than did those in thebox-box condition in both events. These data suggest that theinfants in the ball-box condition succeeded at mapping theocclusion onto the containment event when the object's trajectorywas simplified. These findings are discussed in relation to otherrecent research on tracking object identity.
paper
Recent work has revealed that very young infants have impressive abilities to individuate small setsof objects that undergo temporary occlusion (e.g., Xu & Carey, 1992; Wynn, 1992). However, the sameinfants often fail to utilise surface feature cues such as colour to individuate objects (Simon, Hespos,Rochat, 1995) relying only on spatial (Leslie, et al., 1996; Wilcox, in press) or temporal (Xu & Carey,1996; Spelke, 1995) cues to individuate objects. Wilcox (in press) has explored the developing ability touse distinct cues to individuate. She found that 4.5-month-olds could use shape and size information. 7.5-month-olds could use texture information and 11.5-month-olds can use color information to individuateobjects. Taken together these findings constitute a well-established set of behavioural phenomena. What isstill missing is some account of how development occurs, preferably an account that can relate behaviouraldevelopment to an underlying substrate. In this talk I will describe a connectionist (artificial neural network) model of the development ofindividuation behaviours in infancy. This model builds on the Mareschal, Plunkett, and Harris (1999)model of object-directed behaviours in infancy. The model incorporates the dual route processing accountof cortical object processing (Milner & Goodale, 1995). One route is closely tied to the motor system andprocesses spatial temporal object properties necessary for recognition (e.g. colour and texture). Spatialtemporal and surface feature information are integrated into a single object representation as and whenrequired by a particular response task. The need to integrate information relating to the same object but stored in different registers (i.e. inthe different pathways) leads to the binding problem. If information about 2 or more objects needs to bestored (e.g., blue square moving up and red circle moving left) the system needs a way of keeping track ofwhich surface features correspond to which spatial temporal features (the binding problem). In the case ofdual route cortical processing, one route will have the information 'blue' and 'red' while the other routewill have the information 'up' and 'left'. A system that solves the binding problem will also need torepresent that 'blue' goes with 'up' and 'red' goes with 'left'. We suggest that infants' inability to usesurface feature to individuate hidden objects early in development reflects an inability to maintain bindingacross cortically separable representation during occlusions. To solve the binding problem in a neurologically plausible way, the Mareschal et al. Model wasaugmented with a temporal synchronisation mechanism. Neurophysiological evidence (Singer & Gray,1995; Sougne, 1998) suggests that he brain solves the binding problem by synchronising the firing rates ofneurones coding information about the same object. As long a 'blue' and 'up' fire in synchrony and 'red'and 'left' fire in synchrony it is possible to retrieve the appropriate feature trajectory combination. Thebasis of the developmental model is to assume that cross-cortical binding mechanisms are in place frombirth. Although the mechanisms are in place the information content within each route develops withexperience. Early in development no accurate bindings are possible because there is no reliable informationencoded down either pathway. Later in development the representation are mature enough to permitbindings as long as the objects remain visible, but fail when the objects are occluded. Finally, therepresentations become mature enough to persist during a temporary occlusion, thus allowing thepersistence of bindings during occlusion. The model makes similar errors to infants when faced with two or three occluded objects. Although novel features are processed, a change in bindings (i.e., a change in which surface feature goeswith which object) does not lead to a surprise response. The model illustrates how a mechanisms firstproposed to accounts for other object permanence behaviours can be extend to account for individuationbehaviours. It constitutes a first step to providing a general and neurally plausible account of thedevelopment of infant object directed behaviours.