Experimental Room

All experiments took place in a symmetrical, soundproof cylindrical room consisting of twelve curved panels and a ceiling-mounted, central camera and microphone. One of the panels had a spring hinge and functioned as the door but was indistinguishable from the other panels when closed. Centered within the room was an array of 45-cm-high surfaces (different for each experiment), with four white, opaque 2D disks that served as sticker hiding places.

Procedures

Before entering the testing room, the experimenter explained to the parents the procedures of the experiment and their role in it. Children were told that they would play a hiding and finding game, in which they would watch the hiding of a sticker, be picked up and turned, and then try to find the sticker. On each trial, the child and parent stood in the center of the room, while the experimenter pointed to the four white disks as possible sticker hiding places and oriented the child toward the actual goal location. The child watched the experimenter hide a sticker, and then was picked up by the parent, who rotated in place slowly 3–4 times while obstructing the child’s vision by covering or shading the child’s eyes, and while the experimenter circled the testing space and counted aloud to ten. The experimenter then stood against the wall, aligned with the predetermined facing direction for that trial, and the parent released the child and stood behind as the child was encouraged to find the sticker. If the first search was incorrect, the experimenter showed the child the correct location. Children’s first searches were recorded. In past studies, as well as the present one, children tended to head directly for a particular location following disorientation. No patterns of hesitation, looking around the room, or orienting a particular way before searching were observed.

Method

An enclosed rhombic arena consisted of four 130-cm sides that met at two 60-degree corners and two 120-degree corners. This resulted in a rhombus with not only a 2:1 ratio in the measurement of complementary angles that cross the boundary between obtuse and acute, but also a 2:1 aspect ratio (the ratio between the long and short axes of the arena) and an environment with approximately the same size as the past studies with rectangular rooms. Participants were 8 boys and 8 girls, 27–37 months old (mean = 30.2, SD = 2.7).

Results

Children searched the two geometrically correct corners (correct plus geometric twin) 73.4% of the time, choosing them significantly more than the other corners (50% chance, Cohen’s d = 1.009, t(15)=4.038,p=.001; Table 1). Children performed equally well when the object was hidden at 60-degree vs. 120-degree corner locations (78.1% vs. 68.8%, t<1). Children searched equally at the correct location and its geometric twin (34.4% vs. 39.1%, t<1), providing evidence that they were successfully disoriented by the turning procedure. There were no sex differences (t<1).

Method

Experiment 2 was identical to Experiment 1 except for the placement of the hiding containers, which stood at the middle of the sides of the rhombus, such that they were equidistant from the center of the rhombus and formed a rectangular array of hiding locations (Table 1). Participants were 7 boys and 9 girls 26–36 months old (mean = 29.4, SD = 2.6).

Results

Children concentrated their search at geometrically correct locations (70.3% geometrically correct search, Cohen’s d = 0.892, t(15)=3.569, p=.003; Figure 2). They searched equally between the correct location and its geometric twin (31.3% vs. 39.1%, t<1), showing that they were successfully disoriented by the turning procedure and used the shape of the rhombic environment to reorient themselves. There were no sex differences (t<1). Comparing across experiments, no difference was found in the use of geometry between the rhombus with the hiding locations in the corners (Experiment 1) vs. in the middle of the sides (Experiment 2) (Cohen’s d = 0.135, t<1).

Method

Each corner of the rhombus in Experiments 1 and 2 was detached from the walls, resulting in two 60-degree and two 120-degree corners, with sides 51 cm in length (Figure 3). The freestanding corners were placed equidistant from one another such that their apexes formed a square array. The corners were oriented to open towards the center of the room, as in Experiments 1 and 2. In Experiment 3 (9 girls and 7 boys, 26–35 months old, mean = 29.7, SD = 2.5), the hiding places were located directly inside the corners, where the corners could serve either as direct beacons marking a hidden object's location or as cues for reorientation. In Experiment 4 (7 girls and 9 boys, 26–37 months old, mean = 29.4, SD = 3.4), the hiding places were located midway between the corners and served only as indirect markers of the hidden object (which was situated, e.g., to the left of an acute angled corner). Table 1 depicts these arrays, in relation to the arrays of Experiments 1 and 2.

Results

Children searched equally in the correct location and in the geometrically equivalent opposite location, both in Experiment 3 (25% and 31.3% respectively, t<1) and in Experiment 4 (26.6% and 28.1%, t<1), providing evidence they were disoriented. Combining across search at these locations, children searched at geometrically correct locations no more than they searched at geometrically incorrect locations, either in Experiment 3 (56.3% geometric searches, Cohen’s d = 0.292, t<1.2, n.s.) or in Experiment 4 (54.7% geometric searches, Cohen’s d = 0.169, t<1) (Table 1). Performance in Experiments 3 and Experiment 4 did not differ, (Cohen’s d = 0.063, t<1), showing that the placement of the hiding locations did not affect children’s performance. There were no sex differences (t<1 for each experiment).

A further analysis compared search in Experiments 3 and 4 to search in Experiments 1 and 2. A 2 (Search Location: direct vs. indirect) by 2 (Search Array: complete vs. fragmented) by 2 (Sex) ANOVA on the proportion of geometrically correct searches revealed a main effect of Search Array (F(1,56)=8.038, p=.006, partial η² = 0.126) and no other significant main effects or interactions.

Method

Nine girls and 7 boys (Exp. 5: 24–37 months old, mean = 30.6, SD = 3.9; Exp. 6: 25–37 months old, mean = 30.3, SD = 3.7) participated in each experiment. Experiment 5 used the same corners as in Experiments 3 and 4 (height 45 cm, sides 51 cm each), positioned with the hiding locations as in Experiments 1 and 2. Experiment 6 presented four spatially separated freestanding surfaces (height 45 cm, length 102 cm) arranged in a rhombic array, as if the corners of the full rhombus from the first two experiments were removed (see Figure 4).

Results

In both experiments, children searched equally at the correct corner and at its geometric twin, providing evidence that they were disoriented (Experiment 5: 21.9% vs. 23.4% searches respectively, t<1; Experiment 6: 37.5% vs. 31.3%, t<1; see Table 1). In Experiment 5, children searched at the geometrically correct locations only on 45.3% of the trials (Cohen’s d = 0.265, t<1). The placement of the hiding locations did not affect children’s performance (t<1), and there were no sex differences in accuracy (t<1). Performance using the rhombic array of corners in the present experiment was significantly worse than performance in the complete rhombus (Experiments 1 and 2), Cohen’s d = 0.845, t(46)=3.551, p=.001. In Experiment 6, in contrast, children searched in accord with geometry 68.8% of the time (Cohen’s d = 0.750, t(15)=3.000, p=.009). The placement of the hiding locations did not affect children’s performance (t=0). There were no sex differences (t<1.2, n.s.). Performance using the array of side surfaces was as high as performance using the complete rhombus in Experiments 1 and 2 (Cohen’s d = 0.131, t<1) and better than performance in Experiment 5, using the fragmented, rhombic array of corners (Cohen’s d = 0.649, t(30)=2.512, p=.018).

Method

Participants were 8 boys and 8 girls (23–36 months old, mean = 32.4, SD = 3.9). Each child participated in both Experiments 7 and 8. Half participated in Experiment 7 first and the other half participated in Experiment 8 first. The test array for Experiment 7 consisted of two surfaces 68 cm in length and two surfaces 136 cm in length (total area of surfaces same as Experiments 3–6 and 8). The area of the array for Experiment 7 could not be equated to that of the rhombus experiments because of its different global shape. To ensure that the size of the square array did not affect performance, half of the children were presented with the surfaces arranged in a square array with a smaller area than the array of Exp 1–6 (150 cm on each side), and the other half were presented with the surfaces in a square array with a larger area than in those experiments (180 cm on each side). The test array for Experiment 8 consisted of the same four surfaces used in Experiment 6 (height 45 cm, length 102 cm) in a rectangular arrangement of the same 2:1 aspect ratio tested in Experiments 1, 2, 5, and 6 (see Table 1). Finally, as in the previous experiments, half of the children in each condition were presented with the hiding locations in the empty space between the surfaces, and the other half were presented with the hiding locations at the midpoint of each of the surfaces.

Results

In both experiments, children searched equally at the correct corner and at its geometric twin, providing evidence that they were disoriented (Experiment 7: 18.8% vs. 26.7 % searches respectively, t=1.05, n.s.; Experiment 8: 34.4% vs. 34.4%, t=0). In Experiment 7, children searched at the geometrically correct locations only on 45.3% of the trials (Cohen’s d = 0.191, t<1, see Figure 5). There were no differences in performance dependent on the placement of the hiding locations, the size of the square array, whether the test conducted before or after Experiment 8, or the sex of the child (all t’s<1). In Experiment 8, in contrast, children searched in accord with geometry 68.8% of the time (Cohen’s d = 0.634, t(15)=2.535, p=0.023, see Figure 5). The placement of the hiding locations did not affect children’s performance (t<1), there was no difference between children who participated in the length condition (Exp. 7) before or after Experiment 8 (t<1), and there were no sex differences (t<1). Performance using the rectangular array of side surfaces in Experiment 8 was as high as performance using the complete rhombic array of equal-length surfaces in Experiment 6 (t=0) and better than performance using length differences in Experiment 7 (Cohen’s d = 0.518, t(15)=2.39, p=0.030).

Summary of Findings

Across 8 experiments, 2-year-old children used the distance and directional relations between the surfaces of a rhombic or rectangular enclosure to guide their reorientation. In Experiments 1 and 2, children confined their search to geometrically correct locations in the rhombus, both when they searched directly at a corner of the array and when they searched to the left or right of a distinctive corner, providing evidence that they did not use the corners as local landmarks. Moreover, children searched the incorrect but geometrically appropriate location as often as the correct location, providing evidence that they did not succeed by maintaining their sense of orientation over the turning procedure. Thus, some geometric shape property of the rhombus guided children's reorientation in these initial experiments. This finding does not reveal by itself which geometric relationships children used, however, because the geometrically correct locations were distinguished by distance, length and angle.

Experiments 3–6 revealed that in a rhombic environment, children's reorientation is guided by distance and direction but not length or angle. Children failed to reorient by the distinctive corner angles of the rhombus when they were placed in an equidistant square array (Experiments 3 and 4) or in a rhombic array (Experiment 5). In Experiment 6, in contrast, children successfully reoriented by the four equal-length walls of a room lacking any corner angles, when they were arranged so that two surfaces were closer to the child on their left side and two were closer on their right side. Children's contrasting performance with corners vs. walls cannot be explained by the geometrical configuration of the hiding containers or the array as a whole, because of the contrasting findings of Experiments 5 and 6, in which children were presented with the same rhombic configuration but performed very differently. Children evidently represent their position as a particular distance and direction from one or more extended surfaces in the array.

Experiments 7 and 8 confirmed the findings of Experiment 6 and uncovered an even greater limitation to the geometric representations underlying children's reorientation: Not only do disoriented children fail to use angle differences when corners are presented at equal distances (as in Experiments 3–4), they also fail to use length differences when surfaces are presented at equal distances (Experiment 7). In contrast, both the children in Experiment 6 and those in Experiment 8 reliably used the relative distances and directions of extended surfaces to guide their reorientation in a rectangular space, just as they do in a rhombic space.

The present findings reveal sharp limits to the geometric properties of landmark objects that disoriented children used for beacon-guided navigation. Children failed to use both distinctive angle sizes (Experiment 3–5) and distinctive surface lengths (Experiment 7) as direct features of landmarks specifying the location of a hidden object. Disoriented children's failure to use angle and length to distinguish between local landmarks contrasts with their early sensitivity to angle and length for visual form discrimination. It also contrasts with past navigation research, in which disoriented children have returned to hidden locations successfully when they were marked by beacon objects that were distinguishable by color and shape (Lee et al., 2006) or by shape only (Nardini et al., 2009), although these past studies tested older children (4 and 5 years of age, respectively) and properties such as size and surface area were not equated across the experiments. Nevertheless, the failure to use available local landmarks (and heavy reliance on environmental geometry) is often seen in tests of reorientation in both nonhuman animals and toddlers (Cheng, 1986; Hermer & Spelke, 1994; Wang, et al., 1999), suggesting that there may be something particular about being in a state of disorientation that, in some instances, discourages the use of local landmarks. Landmark use also may have been discouraged, in the present studies, because the potential landmarks were surfaces (corner or wall fragments) rather than recognizable objects. Finally, while the present experiments show children’s failure to spontaneously use angle and length for navigation across only four test trials, children may be capable of learning to use these cues to directly identify beacons or to reorient over many reinforced trials.

Navigation

The present research sheds light on the geometric information that children represent for purposes of navigation. It suggests that past claims that children, from an early age, reorient by analyzing the shape of the surrounding layout (e.g., Cheng & Newcombe, 2005; Hermer & Spelke, 1996; Lee & Spelke, 2010b) must be qualified. Children's reorientation does not depend on either of the two primary properties characterizing the shape of any form: the relative lengths of its parts and the angles at which those parts meet. Instead, reorientation depends on the distance and directional relations among the extended surfaces in the layout: either the distances and directions of surfaces with respect to one another, or their distances and directions with respect to the child's position at the center of the array. This specificity observed in children’s ability to distinguish between locations provides evidence against image-matching theories that would predict successful navigation by such salient differences in the retinal images of those angles and lengths (Cheng, 2008).

The present findings bear on recent findings concerning children's reorientation in environments of various shapes. Newcombe, Ratliff, Shallcross, & Twyman (2009) showed that 3-year-old children successfully find hidden toys in an octagonal arena consisting of an array of alternating walls of two different lengths. Because this arena had no informative aspect ratio, the authors interpreted those findings as evidence that children's reorientation depends on representations of surface length and direction. While the present study was conducted with slightly younger children, Experiment 5 in the present series complements the finding that an informative aspect ratio is not necessary for reorientation, by showing that it also is not sufficient: children failed to reorient within a rhomboid arena with an informative aspect ratio but no extended surfaces. Nevertheless, the present findings suggest an alternative interpretation of the performance of children in the octagonal environment used by Newcombe et al. In an octagonal enclosure, alternating walls that differ in length also differ in distance from the child's position at the center of the display: shorter walls are more distant than longer walls. Thus, children may have reoriented by anchoring directional information to surface distance rather than length. This possibility could be tested through research with fragmented octagonal arrays.

The present experiments also bear on disoriented 1.5–3-year-old children’s performance in rooms that vary in symmetry. In one experiment by Lew et al. (2010), children were disoriented in either a rectangular room or a room consisting of four walls of unequal lengths/distances, meeting at unequal angles. In further experiments, children were disoriented in a circular enclosure with 3–4 large objects at its borders providing distinctive geometric information in either a symmetrical (isosceles or rectangular) or asymmetrical arrangement. Children reoriented successfully only in the symmetric environments. This finding appears to provide the strongest evidence that children reorient by environmental shape, contrary to the present findings, because symmetry is a global shape property. Nevertheless, two features of the environments used by Lew et al. complicate the interpretation of their findings. First, the asymmetrical quadrilateral room varied more subtly in surface distance relationships than did the symmetrical rectangular room. Second, the number of distinctive distance relationships was higher in the asymmetrical environments than in the symmetrical ones. Thus, children's search failures in these studies may be explained either by the subtlety of the distance relations or by the multiplicity of different distance relations, presented in Lew et al's studies. Because the present experiments only test symmetrical arrays, this explanation could be tested by further research comparing children's reorientation in fragmented arenas whose surfaces are placed at distances from the center of the room that are equated, in number and magnitude, across arrays that vary in symmetry. Based on the present findings, we predict that children will reorient successfully in asymmetrical as well as symmetrical environments when the array presents surfaces at a small number of distinctive distances, and that they will fail to reorient in symmetrical as well as asymmetrical environments when surface distances are either subtle or too numerous.

The results of the present study broadly accord with research on navigation using behavioral and neurophysiological methods with animals, using behavioral and neuroimaging methods with human adults, and using the computational tools of computer science to design autonomous, mobile robots. We briefly discuss each of these broader implications.

First, these findings not only inform spatial cognition in humans but also in nonhuman animals and their neural representations of space. Behavioral search patterns of rats in a circular water maze with external landmarks show that disoriented rats rely on the distance between the hidden platform and the walls of the apparatus over the distance relationships to an array of external landmarks, even when the walls are made of transparent material designed to reduce the salience of the walls and improve the visibility of the landmarks (Maurer, & Derivaz, 2000). Furthermore, the importance of distance and direction is consistent with the findings that a variety of animals represent parts of the environment with respect to its axes of elongation (Gallistel & Cheng, 2005; Kelly, Chiandetti, & Vallortigara, 2011; c.f., Pearce, Good, Jones, & McGregor, 2004; Tommasi & Polli, 2004), although our findings suggest that principal axis theories also should be qualified. Children’s successful reorientation with an array of surfaces at varying distances, and their failure to use the same aspect ratio when reorienting within an array of corners, suggest that environmental axes used for navigation are computed not with respect to any environmental input but specifically from the extended surface layout. Nevertheless, the present studies do not address whether these geometric representations of the environment are local or global, and they do not reveal whether the same limits on reorientation by aspect ratio apply to non-human animals. Further research with animals using arrays that tease apart aspect ratio, axes of symmetry, and geometric properties (i.e., angle, length, distance) is needed to determine whether similar geometric representations underlie navigation across various species of animals, as some behavioral experiments (Maurer & Derivaz, 2000) and experiments using neurophysiological methods (O'Keefe & Burgess, 1996; Lever et al., 2002) suggest.

The finding that surface distances guide reorientation is in close agreement with studies of the activity of individual neurons in the hippocampus and surrounding cortex of navigating rats. As we have noted, hippocampal place cells encode an animal’s position by representing location relative to extended wall-like surfaces in the environment (O’Keefe & Burgess, 1996; Lever et al., 2002; Solstad et al., 2008). A model based on place cell responses in rats proposed that place cells are attuned to barrier-like surfaces at a particular distance and direction, with sharper coding at closer distances (Hartley, Burgess, Lever, Cacucci, & O’Keefe, 2000). This boundary vector cell (BVC) model can predict the responses of place cells that remain constant over changes in the curvature of the encoded surface – from flat, to curved, to bent – consistent with a lack of sensitivity to angle. In recent research, cells that fit the BVC model appear to provide input to place cells regarding the surrounding surface layout (Lever, Burton, Jeewajee, O’Keefe & Burgess, 2009). Furthermore, although place cell firing is modulated by extended surfaces, it is not modulated by objects or discrete landmarks in the middle of the space (Cressant, Muller, & Poucet, 1997).

Our findings also accord with the performance of human adults in virtual navigation tasks. As we noted in the introduction, adults, who are given the task of returning a previously viewed object to its former position, will use information both from landmark objects and from extended surfaces to encode and retrieve that position. Adults' encoding of object positions with respect to landmarks, however, depends on attention, whereas their encoding of object positions with respect to an extended surface does not (Doeller & Burgess, 2008). Moreover, the latter encoding, but not the former, is associated with activity in the hippocampus, homologous to that shown in the neurophysiological studies just discussed (Doeller, King, & Burgess, 2008). The automaticity of the encoding of one's distance and direction from nearby extended surfaces, observed in these studies, could explain why information about surface distance and direction is available to children after disorientation. Full disorientation is a rare event, and children likely do not prepare for it in advance. Because surface distance and direction are automatically encoded, however, these relationships will be available to children, after disorientation, to allow them to reestablish their own orientation and position within the array.

Finally, our findings are consistent with those of research in computer science that aims to design navigation systems by which robots can represent both the layout through which they navigate and their own place within that layout. A large class of systems accomplishes both these tasks at once, through a process of Simultaneous Localization and Mapping (SLAM: see Thrun, 2002, for a review and discussion). This system is of special interest in the present context, because reorientation will be possible only if some process of mapping the environment has taken place. Studies in robotics therefore provide an interesting perspective on research on reorientation in humans and animals.

Effective robot navigation systems, including SLAM systems, must overcome a number of difficult problems (see Thrun, 2002), three of which are relevant to the present discussion: the problem of correspondence (navigable layouts often contain similar objects and surface features at different positions, leading to confusion of a new location with a familiar, featurally similar location), the problem of change (navigable layouts often contain moveable objects, whose displacement alters scene features it ways that can lead to failures of recognition of scenes that were previously encountered), and the problem of computational complexity (every individual scene in a natural environment contains a large amount of visual information, and every motion of the robot produces a new, similarly information-rich vista; if all this information were stored, the computational resources needed for navigation would be prohibitively large).

Investigators of robot navigation have long recognized that a focus on extended planar or near-planar surfaces can solve the third of these problems (Egerton, Callaghan, & Chernett, 2000, Thrun, 2002). Navigation by representations of extended surfaces is extremely economical, because the position and orientation of a planar surface can be represented by just three points; smoothly curved surfaces require only a few more points for their curvature to be specified (Gee, Chekhlov, Calway, & Mayol-Cuevas, 2008). In contrast, navigation by representations that include landmark objects require more information, because local landmarks often are misrecognized without highly detailed local representations (Egerton, Callaghan, & Chernett, 2000).

Drawing on research on animals, Gallistel (1990) has observed that a focus on extended planar or near-planar surfaces also can address the first two problems. Concerning the correspondence problem, the extended surfaces in natural layouts, and in large, artificial layouts, often have similar colors and textures, and they often stand under or behind objects that can have similar shapes, but they rarely appear in symmetrical arrangements. A navigation system that focuses on the distance relations among these surfaces therefore will make fewer correspondence errors than one that focuses on surface properties like color and texture, or on object shape properties like angle and length. Concerning the problem of change, the extended surfaces in natural layouts are less likely to move than are objects. A navigation system that focuses on extended surfaces rather than objects therefore will be less likely to fail to recognize familiar scenes as the objects within them are displaced.

In summary, the present findings accord with those of a rich array of fields from psychology to biology to applied mathematics and computer science. These findings do not accord, however, with the findings of research in what would seem to be a closely related area of developmental psychology: research on the development of form perception.

Visual form analysis

Although children failed to use angular or length relations in the present studies, a large literature on visual form and object perception shows that even younger children and infants are sensitive to this information in other task contexts. A long history of research using various tasks including word learning and object manipulation provides evidence that young children perceive the length and angular relations that specify the shapes of objects (DeLoache, Kolstad, & Anderson, 1991; Gentner, 1978; Landau, Smith & Jones, 1988; Samuelson & Smith, 2005; Smith, 2009). Even infants recognize common objects by their shapes in both 2D and 3D arrays (Pierroutsakos & Deloache, 2003). When various geometric properties were presented to infants using 2D visual forms, they displayed high sensitivity to angle and length from their first months of life (Schwartz & Day, 1979; Slater, Mattock, Brown, & Bremner, 1991).

In contrast, infants and children show much lower sensitivity to the directional (or sense) information that distinguishes a form or object from its mirror image (Lourenco, Huttenlocher, & Fabian, 2005). In studies of adults, visual form analysis and object recognition are invariant both over mirror image reflection and over changes in scale (Biederman & Cooper, 1992; Malach et al., 1995): they therefore generalize over relationships of distance and direction. Both scale- and directional invariance characterize visual form perception throughout development (Izard & Spelke, 2009). Children's evident lack of sensitivity to these properties in the present studies therefore does not reflect either a general failure to perceive length or angular relationships, or a general success in perception of distance and directional relationships, in visual arrays. Instead, navigation and object recognition evidently depend on different geometric properties and relations (Landau & Jackendoff, 1993).

Why do children fail to represent the shapes of the arrays in the present studies, when they succeed at representing the shapes of forms and objects? First, children may readily extract the shapes of small forms, but less readily extract the shapes of larger (room-sized) ones. Second, children may readily perceive the shapes of forms that are convex (like an object) but not the shapes of forms that are concave (like a hole in a surface or the space enclosed within a room). Third, children may extract shape with equal readiness from arrays large and small, and from both convex objects and the holes or spaces they enclose, but they may fail to use this shape information to guide their navigation. Future research focusing on tasks of shape recognition is needed to test these possibilities (for a start, see Shutts, Ornkloo, von Hofsten, Keen, & Spelke, 2009).

Regardless of the reasons for children's failure to reorient by length or angle information, the contrast between the findings of present experiments and those of the large literature on children's visual form perception suggests that distinct systems of geometrical analysis serve to guide children's analysis of the shapes of objects and visual forms, on one hand, and the shapes of 3D navigable layouts, on the other (Spelke, Lee, & Izard, 2010). Much remains to be learned concerning the conditions under which each of these systems is activated. The convergence of the present findings with those of studies of oriented navigation (Hartley et al., 2004; Doeller & Burgess, 2008) suggests that the system for analyzing distance and directional relationships to extended surfaces is activated not only after disorientation but during oriented navigation as well. It is not clear, however, whether the differing systems are activated by the differing tasks that children are given (finding vs. recognizing objects) or the differing displays they encounter (large 3D enclosing surfaces vs. small objects or 2D forms).

We thank the Laboratory for Developmental Studies at Harvard University, Annie Douglas and Danielle Hinchey, for help in running these experiments. This research was funded by NIH-grant HD 23103 to ESS and Postdoctoral Research Fellowship to SAL by the Center for Mind/Brain Sciences at University of Trento.