It is probably a good thing that infants and children begin learning about relations by attending to the objects and context in which they encounter a relation. Consider, for example, real world instances of CONTAINMENT and SUPPORT. What objects can go ``inside'' other objects, what objects can support other objects, and how one physically realizes these relations depends very much on the specific objects and their perceptible properties. Titzer, Thelen, & Smith (1998) recently demonstrated this point in a study of infants understanding of containment and support given transparent or opaque landmarks. The experimental procedure built on an earlier study by Diamond diamond:90. She showed that infants are more successful in retrieving desired toys from opaque rather than transparent containers. From one perspective, this is a perplexing result: an object in an opaque container cannot be seen and must be remembered whereas the desired object in a transparent container can be continuously seen. However, from the perspective of a concept of containment, Diamond's results are not surprising. Transparent containers present unique perceptual cues to surfaces and openings. And, indeed, the babies in Diamond's study seemed not to understand where the openings were in the transparent containers as they tried to reach through the transparent surface to retrieve the toy. Titzer et al. (1998) tested the idea that specific experience with transparent containers may be essential to successful object retrieval from such containers. In a two-month training study, they exposed eight-month-old infants to transparent containers; control infants were given opaque containers for the same period. The containers were identical in both cases except for opaqueness and varied in shape and size from small cups to buckets large enough for a baby to put over his or her head or even to sit on. The infants were given no special training; parents were simply asked to give the infants the containers to play with daily for at least 10 minutes over the two-month experiment. When the infants were brought back to the laboratory at ten months and tested in Diamond's procedure, the infants experienced with transparent containers and the control infants performed differently. Specifically, only the infants who had played with transparent boxes knew how to rapidly retrieve objects from transparent boxes. Two months of perceiving and acting on transparent containers taught these infants the unique perceptual cues relevant for containment in a transparent receptacle.
The infants generalized their learning about transparent containers to problems concerning SUPPORT. Titzer et al. (1998) tested both the infants trained with transparent containers and those trained with opaque containers on an apparatus known as the visual cliff. This is a highly studied device invented by Gibson & Walk (1964) to test infants sensitivity to depth cues. In traditional testing, the apparatus is a transparent table top that sits over a substantial drop. Infants are placed on the shallow side near the drop and their behavior is observed. And the traditional result is that ten-month-old infants avoid the cliff and the deep end, crawling away from the visual cliff to a secure position on the shallow side. Titzer et al. (1998) observed the infants in the control condition, the ones who had only played with opaque containers, also retreat from the visual cliff, apparently believing -- despite the solid transparent surface below them -- that they might fall. In contrast, infants trained with the transparent containers acted unlike infants in any other study of the visual cliff; they confidently and happily crawled right over the visual cliff. These infants had apparently learned the visual cues to transparent surfaces and knew what typically developing infants with limited experiences with transparent surfaces do not: solid transparent surfaces support just as do opaque ones. Clearly, real-word experiences and specific object properties matter in understanding SUPPORT and CONTAINMENT in the context of transparency; we suspect that this is the same for understanding of SUPPORT and CONTAINMENT in the context of opacity. The real-world use and recognition of relations requires their grounding in specific object properties
Indeed, objects properties are integral to all spatial concepts. For example, when shown a block on a box while being told ``the block is acorp the box'' English speakers interpret acorp to mean `on'. In contrast, when shown a stick on a box, English speakers interpret acorp to mean `across' [LandauLandau1996]. The shape of the trajector and the shape of the landmark matter. The relevance of object properties is also apparent in judgments of containment. Consider, for example, panels A and B in Figure 9. Most people judge the apple not to be in the bowl in panel A but to be in the bowl in panel B. An apple on top of other fruit that is contained in a vessel is IN. The relevant properties are not just spatial ones; object categories matter as well. For example, the apple is in the bowl when sitting on other apples but is not in the bowl in panel C when sitting on blocks that are in the bowl.
 |
Feist & Gentner (1998) showed in a recent study of adults' judgments of in and on that the way in which the very same object was categorized determined relational judgment. They varied the curvature of the landmark, as illustrated in Figure 10. In some conditions, the displays were unlabeled; in others the trajector was labeled as an animate fly and the landmark as a bowl, plate, or dish. All variables mattered. Inanimate trajectors were more likely to be judged in than animate ones, more curved landmarks yielded more in judgments than less curved ones, and labeling the landmark as a bowl rather than as a plate yielded more in judgments. Clearly, object labels -- not just their spatial properties -- matter when adults make relational judgments. Thus, relational development must not consist so much of stripping away all object information, but must instead consist of learning the particular kinds of object properties relevant to particular relations.
 |