Now You See It, Now You Don't: A Gradualistic Framework for Understanding Infants' Successes and Failures in Object Permanence Tasks

3.5-month-old infants seem to show an understanding of the concept of object permanence when tested through looking-time measures. Why, then, do infants fail to retrieve hidden objects until 8 months? Answers to this question, and to questions of infants' successes and failures in general, depend on one's conception of knowledge representations. Within a monolithic approach to object permanence, means-ends de cits provide the standard answer. However, the current experiments with 7-month-old infants indicate that the means-ends accounts are incomplete. In the rst two studies, infants were trained to pull a towel or push a button to retrieve a distant toy. Infants were then tested on trials with an opaque or transparent screen in front of the toy. Trials without toys were also included, and the di erence between Toy and No-Toy trials in number of retrieval responses was used as a measure of toy-guided retrieval. The means-ends abilities required for toy-guided retrieval in the Transparent and Opaque conditions were identical, yet toy-guided retrieval was more frequent in the Transparent condition. A third experiment eliminated the possibility that training on the retrieval of visible toys had led infants to generalize better to the Transparent condition. To explain these data, an account of the object permanence concept as a gradual strengthening of representations of occluded objects is developed in the form of a connectionist model. The simulations demonstrate how a system might come to form internal representations of occluded objects, how these representations could be graded and strengthened, and how the gradedness of representations could di erentially impact upon looking and reaching behaviors. Munakata, McClelland, Johnson, and Siegler 1 In recent years, a great deal of attention has been focused on how smart infants are. It often seems that each new nding demonstrates that infants possess some kind of knowledge at younger and younger ages, much earlier in development than previously thought. Why do infants appear so much smarter today? Research on infants' understanding of object permanence provides a rich experimental domain in which to explore this question. The object permanence concept involves the understanding that objects continue to exist independent of our percepts of them, and that objects maintain their identity through changes in location (Piaget, 1954). Piaget used the successful retrieval of a completely hidden object as one measure of the object permanence concept. Infants succeed with this task only around 8 months of age, and even then show incomplete mastery by making errors in the AB task. In this task, devised by Piaget (1954), infants watch as an object is hidden in one location (A). Those infants who are able to successfully retrieve the object may nonetheless tend not to retrieve the object when it is hidden in a new location (B), showing a perseverative error (the AB error) to reach to the original hiding location (A). More sensitive measures of the object permanence concept have been designed to evaluate how early the infant \has" this concept. Baillargeon (1993, 1987), for example, has measured looking times to \possible" and \impossible" conditions of events involving occluded objects and found that infants as young as 3.5 months look longer at the stimulus during the impossible condition. These longer looking times are taken as an indication of the infants' perception of an unusual event, and so, of some understanding of the continued existence of the occluded object. Earlier signs of competence have also been demonstrated in the AB task. Diamond (1985) has noted that infants occasionally reach to the wrong location while looking at the correct location, and often reach to the wrong location without looking in it and then immediately reach to the correct location. Recently, more systematic testing of infants' reaching versus looking behaviors on the AB task (e.g., Matthews, 1992; Lecuyer, Abgueguen, & Lemarie, 1992; Baillargeon & Graber, 1988; Baillargeon, DeVos, & Graber, 1989) has con rmed this dissociation: looking tests indicate infants' understanding of new hiding locations before reaching tests. From these types of ndings, many researchers now attribute the concept of object permanence to infants at an earlier age than did Piaget. In fact, many of Spelke's ndings using the looking-time method (Spelke, Breinlinger, Macomber, & Jacobson, 1992) appear to rely on the existence of such a concept at an even earlier age. The key question then becomes: Why do infants fail to retrieve hidden objects up until 8 months, and even then show the AB error, if they have the concept of object permanence many months earlier, perhaps even from birth? The way in which one answers this question regarding infants' failures, as well as questions about their successes { why they appear so much smarter today { depends critically on one's conception of what it means \to know" something. What does it mean to say that the infant knows that an occluded object is still there? What does it mean to have the concept of object permanence? What form does this knowledge take? These are fundamental questions, but there is relatively little theorizing to address them or to explain what infants can and cannot do. Instead, there is a common tendency to treat various developmental capacities as all-or-none (see discussions in Siegler, 1992, 1989, Siegler & Munakata, 1993; Karmilo -Smith, 1992, 1991; Flavell, 1971, 1984), and, in some cases, present from birth (Spelke et al., 1992). Often, relatively little focus