The developmental evidence indicates that there is not some magical point in development at which children become able to use relations. Rather, relational development appears to progress domain by domain -- with children understanding relations in domains in which they are knowledgeable and reasoning poorly in domains in which they are relative novices (see Gentner & Rattermann, 1994, for a review). Thus, one sees in development the same developmental trend over and over in different domains -- first children center on objects, then as they know more about the specific relational domain, they attend to relations presented in known contexts and with known objects, and ultimately, they attend to and/or reason about the relation across diverse kinds of objects and settings. That is, they progress from a more similarity-based to a more abstract understanding of relations within each domain.
This trend, for example, is evident in four- to seven-month-old babies' attention to the relations of OVER and UNDER. In one study, Quinn (1994) used a familiarization paradigm. This paradigm makes use of infants' increased attention to novelty. During familiarization trials, the infant is presented repeatedly with stimuli from one category and then on test trials is presented with novel stimuli that are either in that category or not. The reasoning is this: if infants perceive the within-category test stimulus to be like (in the same category as) the familiarization set, then they should find it boring in post-familiarization. In contrast, if infants see the out-of-category test stimulus as unlike (not in the same category as) the familiarization set, they should show increased attention to this stimulus because it will be perceived as novel. Figure 7 (top) shows one familiarization and test set used by Quinn. The infants were repeatedly shown the a and b familiarization sets -- sets which depict the relation of OVER. On test they were presented four novel test stimuli, two of which also depicted the relation of OVER and two the relation of UNDER.
In a subsequent study, Quinn and colleagues (1996) varied the shapes of the components of the stimulus displays from familiarization to test. This variation among the objects involved in the relational display disrupted four-month-olds' performance. When the objects changed, all the test displays apparently looked new and therefore were attended to. In contrast, older infants, seven-month-olds, looked less at the test displays presenting the same relational configuration and more at the test displays presenting the different configurations even when the component objects in the displays varied. Thus, seven-month olds but not younger infants were able to generalize over different kinds of objects involved in the relation. The developmental trend is from object-centered conservative generalizations to ones apparently based on a more abstract representation of the relation.
Gentner & Rattermann gentner+rattermann review a number of other studies of infants' attention to relations that make the same point: attention to relations is at first highly dependent on the objects involved and becomes less so with development. They also review numerous studies of relational concepts and reasoning in much older children that again show the same trend from more object-based relations to more abstract ones. In an unpublished study by Ratterman, Gentner, and DeLoache cited in Gentner & Ratterman gentner+rattermann three- and four-year-olds were presented with the following task: The experimenter and child were each given three objects as illustrated in Figure 8. The experimenter selected one from her set as the ``winner'' and the child's task was to select the corresponding object. Young children had considerable difficulty choosing relationally and tended to choose the object from their set that matched the experimenters in actual size or in other object properties. However, in a subsequent experiment, it was shown that young children could respond relationally if the task was presented in a domain that they understood -- specifically, if the objects in each set were presented as ``the daddy one,'' ``the mommy one,'' and ``the baby one.'' This result, along with that of a study by Kotovski & Gentner kotovski in which infants were progressively trained to make more and more abstract relational inferences, demonstrate that the developmental change is driven more by experience than by maturation or age.