Object Individuation and Physical Reasoning in Infancy: An Integrative Account

Abstract

Much of the research on object individuation in infancy has used a task in which two different objects emerge in alternation from behind a large screen, which is then removed to reveal either one or two objects. In their seminal work, Xu and Carey (1996) found that it is typically not until the end of the first year that infants detect a violation when a single object is revealed. Since then, a large number of investigations have modified the standard task in various ways and found that young infants succeed with some but not with other modifications, yielding a complex and unwieldy picture. In this article, we argue that this confusing picture can be better understood by bringing to bear insights from a related subfield of infancy research, physical reasoning. By considering how infants reason about object information within and across physical events, we can make sense of apparently inconsistent findings from different object-individuation tasks. In turn, object-individuation findings deepen our understanding of how physical reasoning develops in infancy. Integrating the insights from physical-reasoning and object-individuation investigations thus enriches both subfields and brings about a clearer account of how infants represent objects and events.

ACKNOWLEDGMENTS

The preparation of this article was supported by a grant from the National Institute of Child Health and Human Development to Renée Baillargeon (HD-21104). We would like to thank Luca Bonatti, Andrei Cimpian, Jerry DeJong, Cindy Fisher, Jane Hedges, Amy Needham, Dan Swingley, and Teresa Wilcox for helpful comments and discussions.

Notes

¹ To give another example of how EBL contributes to the formation of event categories, consider how infants identify the category of support events, at about 4 to 5 months of age (e.g., J. Li, Baillargeon, & Needham, 2012). To start, infants notice that when a first inert object is released in contact with a second inert object, the first inert object sometimes remains stable and sometimes falls. Infants cannot predict or interpret these contrastive outcomes: the same physical representation (“first inert object released in contact with second inert object”) leads to different outcomes (“first inert object remains stable,” “first inert object falls”), suggesting that important information is missing from the representations. Infants then search for the conditions associated with these contrastive outcomes and eventually detect that inert objects remain stable when released on top of other inert objects, but not against or under them. Next, infants engage in a causal analysis of this regularity: their knowledge of gravity (e.g., Needham & Baillargeon, 1993) and solidity (e.g., Baillargeon, Spelke, & Wasserman, 1985) suggests that when object-A is placed on top of object-B, object-B can block the free fall of object-A; however, when object-A is placed against or under object-B, there is then nothing to block the free fall of object-A. This explanation leads to the creation of a new support category, a kind of causal interaction in which objects play the role of “support” and “supportee.” From this point on, when infants see an event in which an object is released in contact with another object, the PR system represents more precisely the spatial arrangement of the objects (i.e., on top of, against, or under); when necessary conditions are met, the PR system categorizes the event as a support event and assigns the roles of support and supportee to the appropriate objects in the event.

² What happens when the spatiotemporal and the categorical information in the structural layer of an event's physical representation point to inconsistent numbers of objects in the event? In particular, what happens if an object disappears behind one screen and then reappears from behind another screen, without appearing in the gap between them? In this situation, the spatiotemporal information suggests that two objects are present (since no single object could follow such a discontinuous path), whereas the categorical information suggests that a single object is present (since the same object is seen to disappear and reappear). Beginning at about 100 days of age, infants resolve such conflicts by assuming that two identical objects are involved in the event (e.g., Aguiar & Baillargeon, 2002; Spelke, Kestenbaum, Simons, & Wein, 1995a). Similarly, if a cover is lowered over one object, and a second cover is then lifted to reveal the object, infants aged 100 days and over assume that an identical object was already hidden under the second cover (e.g., Wu, Li, & Baillargeon, 2012).

³ How does the PR system gather information about a variable? Researchers have suggested that when infants attend to objects, such as the three objects at the start of our event in Figure 2A, (1) the object-tracking (OT) system assigns an index to each object (e.g., Leslie, Xu, Tremoulet, & Scholl, 1998; Pylyshyn, 1989), and (2) the object-representation (OR) system opens a temporary file for each object, listing both individual (e.g., color) and relational (e.g., relative height) features (e.g., Huttenlocher, Duffy, & Levine, 2002; Kahneman, Treisman, & Gibbs, 1992; Rose, Gottfried, Melloy-Carminar, & Bridger, 1982). When the duck and block move behind the screen, the PR system also becomes involved: the objects are now engaged in a physical interaction, and the PR system's main purpose is that of predicting how such interactions will unfold. One possibility currently being investigated is that, when the PR system requires information about a variable (e.g., color), it taps the OR system for this information (e.g., Wang & Baillargeon, 2008b; Wang & Mitroff, 2009). In the first year of life, object representations in the PR system often contain only a small subset of the featural information in the object files of the OR system; as mentioned earlier, PR representations are initially sparse and become gradually richer as infants identify relevant variables (for further discussion of the links between the OR and PR systems, see Baillargeon et al., 2011; Wang & Baillargeon, 2008b).

⁴ There appear to be significant developments during the first year of life in infants' ability to bind object information; several factors appear to affect this ability, including the number of objects in an event and the number of events shown side by side (e.g., Kibbe & Leslie, in press; Oakes, Ross-Sheehy, & Luck, 2006; Rose, Feldman, & Jankowski, 2001; Wilcox & Schweinle, 2002). For example, Wilcox and Schweinle (2002) found that in a very simple object-individuation task with two distinct objects, A and B, 7.5-month-olds expected to see A and B when the screen was removed and were surprised to see only A, A and A, or A and C, suggesting that they had formed precise representations of the two objects. In contrast, 5.5-month-olds were surprised to see only A, but they were not surprised to see either A and B or A and A, suggesting that they correctly expected two objects, but lacked precise expectations about their featural properties. Due to space limitations, we do not discuss the development of binding in this article.