Infant A-not-B errors: A case for conceptual dynamics

Infants often erroneously search for an object at a previously found location (A) despite seeing it hidden at a new location (B). Smith et al (1999) deny such A-notB errors arise from internal object-related conceptual states. Instead, they arise from the dynamics of goalrelated perception and action. We show that their model does not account for cup orientation, and suggest augmenting dynamics with conceptual/relational structure to account for this e ect. Infant A-not-B errors Object occlusion tasks are thought to probe the conceptual mind. The canonical form of one such task, called A-not-B, consists of two hiding locations and a retrievable object (e.g., toy). On the rst trial, the infant sees the toy hidden at one of the two occluding locations (A). The infant is allowed to retrieve the toy. On the second trial, the same toy is seen hidden at the other location (B). Typically, on the second trial, the infant will search for the toy where it was successfully retrieved on the rst trial. This behaviour is called A-not-B, or perseverative error. Many factors are known to a ect performance, for example: age; delay between occlusion and retrieval; spatial separation of locations; and their visual distinctiveness. While theoretical accounts are almost equally proli c, they generally assume some form of internal object representation upon which the infant acts in forming a response (see Smith, Thelen, Titzer, & McLin, 1999, for a recent review). A systematic program varying task parameters, then, should nally illuminate the most basic component of human cognition, concepts. The dynamical systems model That infants are accessing concepts is not, however, universally accepted. Smith et al. (1999) argue that infant errors are the result of a complex interaction between their desire to retrieve the object, and a dynamically changing environment. Support for their dynamic systems model comes from three sorts of experimental observations: (1) The likelihood of a reach in one direction was an increasing/decreasing function of the number of previous reaches in the same/other direction. (2) The direction of reach was \pulled" by a visual distractor. (3) Greater amounts of experience with occluders resulted in fewer errors (Smith et al., 1999). They concluded that behaviour is far too uid, too easily altered by extraneous events to be the result of some enduring structure corresponding to an \object concept" in infants, or even adults. Instead, behaviour is just an interaction between goal-directed perception, action and their prior states. No explanation for cup orientation Denial of concepts challenges one of the most generally held assumptions of cognitive science. Yet, empirical support for concepts is itself di cult to nd unequivocally. While concepts are supposed to capture certain invariant environmental properties, concepts must arise from speci c environmental states to be e ective. This confound is potentially resolvable by looking for generalizations that span experience in a way that necessitates internal structure. For example, generalization over relations between certain perceptual states is not possible with \standard function approximation" style connectionism without some form of structured representation (Phillips, 2000). In a similar vein, empirical evidence recalled here is not accounted for without supposing some form of internal concept representation. Freeman, Lloyd, and Sinha (1980) provided evidence that occluder functionality (cup used as a container) affects performance. Infants were tested on an A-not-B task with three conditions, where: (1) at opaque screens baseline; (2) upright; and (3) inverted cups were used as occluders. In the cup conditions, the object was placed behind the cup. The critical result was that infant errors were signi cantly less than baseline errors in the upright condition, but signi cantly more than baseline in the inverted condition. A series of control experiments ruled out accounts based purely on perceptual features, such as the location of the cavity; and motor requirements such as reaching in versus reaching behind the occluder. Therefore, something other than the perceptual or motor aspects of the task must a ect performance. In a dynamic systems model, performance is also affected by experience. Smith et al. (1999) reported that infants given additional playtime with transparent containers performed signi cantly better than otherwise normal infants. However, experience alone cannot explain the orientation e ect. A simple one factor account (occluder) citing di erential experience with cups and screens cannot explain why errors with cups were both signi cantly greater and fewer than screens. A two factor account (occluder, orientation) citing greater experience with upright cups than screens than inverted cups accounts for the signi cant di erence in error in the transposed movement condition, but it also predicts a di erence in the static movement condition, where none was found. A three factor explanation (occluder, orientation, movement) could account for the above pro le of errors, but only by anchoring to several unlikely conditions on the infant's world: First, cups are typically used as containers, whereas screens are used as partitions. But experience driven error requires there having been more cases of cups used as partitioners/occluders than screens, even though their primary role is as containers. Second, when used in their secondary role, as partitions, orientation is not signi cant. But, errors require there having been more experience with upright than inverted cups as partitioners/occluders. Last, if role reduces to experience, fewer errors should occur when the object was hidden under/in the cup than behind it, but no di erence was found. Unable to guarantee these conditions means their dynamic systems model cannot sustain a realistic explanation of infant errors due to cup orientation.