Experiment 1

Method

Participants

Thirty nine undergraduates (14 female) participated, recruited from a university psychology department undergraduate participant pool. English was their first and dominant language. Ten additional adults served as participants for the similarity ratings. They included graduate students and research assistants who had no detailed knowledge of the purpose of their ratings.

Materials

The stimuli consisted of five ‘quartets’ of novel objects (see Figure 1 ). Each quartet included one target object, made from materials such as toy legos and play dough, and three test objects. One test object matched the target object only and exactly in shape (exact-shape match), one object matched the target only and exactly in color or decorations (color-match), and one matched the target object exactly in color while also sharing some shape features. We called this latter object ‘overall match,’ since it matched the target on both color/decorations and shape. Because in a classical shape bias paradigm ( Smith, 2000 ) the two dimensions of shape and color are clearly separated and pitted against each other, in this study a shape bias refers to instances in which participants generalize novel names to perfect shape matches. The low-similarity condition consisted of the target, the exact-shape match, and the color-match. The high-similarity condition consisted of the same target and the exact-shape match, but this time the second test object was the overall match. A digital camera was used to take videos of the objects (moving horizontally in space). The stimuli were blocked, such that all color matching objects were presented in one block and all overall matching objects were presented in another block.

Open in a separate window

Figure 1

Example of stimuli used in Experiment 1

The similarity among the test objects and the target objects was not manipulated systematically because in general, it is difficult to manipulate similarity in multiple dimensions, and in particular, no metric for shape similarity exists ( Sera & Millett, 2008 ). Therefore, we used adult judgments to confirm that overall-match objects (i.e., test objects similar to the target on two dimensions: color/decorations and shape) were more similar to the target than the color-match (i.e., test objects similar to the target only in color/decorations). We asked 10 adults to rate the similarity between the color/overall match, the exact shape match, and the targets on a scale from 1 (not similar) to 7 (very similar). The mean similarity rating was 2.48 ( SD = .74) for the color-match/target objects in the low-similarity condition and 3.82 ( SD = .61) for the overall-match/target objects in the high-similarity condition. Participants rated the overall match significantly more similar to the target than the color-match, t (9) = 4.70, p < .01). Moreover, the exact shape match was rated as more similar to the target than the color-matching object, M = 3.46, SD = .47; t (9) = 3.94, p < .01, and the ratings for the exact shape match and the overall-shape match did not differ from each other ( p > .10).

Procedure

Participants were tested in a quiet room. They were told that they would see a novel object for a short time presented on a laptop followed by two different objects on the screen. Their task was to look at the first object and then choose one of the test objects that matched the instruction given by the experimenter. To control for a practice effect, two blocks of trials were employed such that half of the participants were presented with the high-similarity condition first and the other half were presented with the low-similarity condition first.

Within each block, participants viewed NoName trials followed by Name trials. During NoName trials, participants saw the target object alone on the screen for approximately one second. The two test objects (i.e., the exact-shape and the color-match/overall match objects) were then presented simultaneously, one on the left, and one on the right. The location of presentation was counterbalanced across trials. After presentation of each set of stimuli, the videos were stopped, and the participants were asked to point to the one that looked the same. Responses were recorded manually. The Name trials were similar except that each target object was presented with a novel name (i.e., “This is a zup ”). A total of five novel names was used ( zup, tiz, seb, dax , or pilk ). Then participants saw the test objects simultaneously, and were asked to identify the object having the same label (i.e., “Where is the zup ?”). The same pattern was repeated with the other four pairs of objects for both high- and low-similarity conditions.

Results

Figure 2 shows the percent of items for which participants chose the shape-match object, by trial and condition type. Participants chose the shape-match significantly more often than chance (50%) for all trials, t (38) = 24.56, p < .001. A two-way repeated measures analysis of variance (ANOVA) was performed with similarity (high, low) and trial (NoName, Name) as within-subject factors. The dependent variable was percent of items for which choice was the shape match. A significant effect of similarity appeared, F (1, 38) = 15.56, p < .001, η ² = .291, and a marginally significant effect of trial, F (1, 38) = 2.94, p = .095, η ² = .072, with no interaction.

Open in a separate window

Figure 2

Percent of shape choices in the low-similarity and high-similarity conditions, Experiment 1

Thus, participants chose the object that had the same shape as the target significantly more often in the low-similarity than in the high-similarity condition during both NoName and Name trials. Furthermore, they preferred the shape match relatively more often during Name trials than No-Name trials, primarily in the high-similarity condition as they were at ceiling in the low-similarity condition. In sum, presence of the high-similarity objects significantly diminished adults’ shape preferences.

Experiment 2

In Experiment 1 we manipulated color-matching and overall-matching test objects in low-similarity and high-similarity conditions while keeping the target and the exact-shape match the same. In Experiment 2, we tested 24-month-olds, using different stimuli in high- and low-similarity sets to minimize memory load. Moreover, we used the Intermodal Preferential Looking Paradigm (IPL) to reduce task demands. An advantage of the IPL paradigm is that the language abilities of children can be tested at early points in development without using a forced- choice task.

Method

Participants

Ten typically developing children (six female) participated. They were recruited from a university database of children. They ranged in age from 23 to 25 months ( M = 24.6, SD = 0.61). Ten adults served as participants for the similarity ratings. They included graduate students and research assistants who had no detailed knowledge of the purpose of the ratings; none had participated in Experiment 1.

Materials

Six novel target objects were constructed from simple toys such as wooden blocks and legos; each was 3–5 inches wide and 3–8 inches long. Three of the object sets were designated as low-similarity items, because one of the test objects matched the target only in shape (exact-shape match), whereas the other test object was identical to the target only in color; it shared no other physical similarities (color match). The other three object sets were designated as high-similarity items, because, as with the previous items, one of the test objects matched the target only in shape (exact-shape match); however, this time the other test object was a match to the target not only in color, but also shared some shape properties with the target (overall match) (see Figure 3 ).

Open in a separate window

Figure 3

Example of stimuli used in Experiment 2

To confirm that the overall matching objects in the high-similarity condition were ‘more similar’ to the targets than the color matching objects in the low-similarity condition, 10 adults rated the similarity between the overall/color match, the exact-shape match, and their targets on a scale from 1 (not similar) to 7 (very similar). They rated the overall matching objects in the high-similarity condition ( M = 2.83, SD = .57) as significantly more similar to their targets than the color matching objects in the low-similarity condition ( M = 1.73, SD = .68), t (9) = 4.10, p < .01. However, in both conditions, the exact shape matches were rated as more similar to the target than the overall-matching and the color-matching objects (low-similarity condition M = 4.83, SD = .50; high-similarity condition M = 4.00, SD = .49). Moreover, participants rated the exact shape matches as more similar to the target in the low-similarity condition than the exact shape matches in the high similarity condition, t (9) = 4.03, p < .01.

Each of the six target objects was video recorded moving back and forth horizontally; 4-second clips were then extracted and arranged in a configuration. The stimuli were presented in two consecutive blocks of trials. The first block constituted the NoName condition, in which no novel names were introduced, and the second block constituted the Name condition, in which the target object was presented with a novel name. Each block was composed of four trials for each set of target and test objects. The first two trials served as familiarization trials, in which the target object appeared first on one side and then on the other side of the screen. Trial 3 presented the test objects; the audio was neutral and non-directing. Trial 4 served as the test trial, in which the test objects were presented again, now with the test audio. The expectation is that children will look longer at the stimulus that matches the audio ( Hirsh-Pasek & Golinkoff, 1996 ). Each trial lasted approximately 4 seconds, with 3 seconds of inter-stimulus interval (when a centering light was presented), for a total of 25 seconds per block and 5 minutes for the entire video. The side of the matching screen was counterbalanced within participants, and alternated in a Right/Left–Left/Right pattern.

Procedure

Children were tested individually in their homes as part of a larger, longitudinal study of language development (Swensen, Kelley, Fein & Naigles, 2007). They were seated either on the floor or on their parents’ lap approximately three feet in front a 3.5 x 5 foot screen. The stimuli were projected from an Apple Powerbook onto the screen via an LCD projector. The audio stimuli were presented via a central speaker that was located behind the screen, outside the view of the child. A digital camera located in front of the screen recorded the child’s eye movements, which were coded off-line. The MCDI Toddler Form (CDI; Fenson, Dale, Reznick, Thal, Bates, Hartung, Pethick & Reilly, 1991 ) had been mailed to parents ahead of time and was collected at the end of the session. Only the “Total Words Produced” part of the CDI was tabulated and analyzed for this study.

Coding

Research assistants coded the films after each visit frame-by-frame, recording children’s fixations to the right or left videos, to the center, or away. The stimulus audio was not available to the coders. The children’s visual fixations were tabulated and analyzed by percent of looking to each video for each trial. On each trial, visual fixations were registered after the child looked at the center lights for more than 0.3 seconds. Trials where a child had not looked at the center lights for at least 0.3 seconds, and/or did not look at the either screen once the videos were presented, were excluded. The empty cells for these trials (comprising 17.5% of trials overall) were replaced by the means for that item. ¹ Data from two children were coded by a second person. The average correlation between the two observers was .98, indicating high inter-rater reliability.

Results

The results appear in Figure 4 . A two-way repeated measures ANOVA was conducted with similarity (high, low) and trial (NoName, Name) as within-subject factors. A significant similarity by trial interaction appeared, F (1, 9) = 31.22, p < .001, η ² = .77, and no other effects. Planned comparisons revealed that, with the low-similarity items, children preferred the exact-shape match significantly more during the Name trials than the NoName trials, t (9) = 3.73, p < .01. They preferred the exact-shape match significantly more than the color match during Name trials, t (9) = 4.50, p < .01, but showed no preference during NoName trials ( Figure 4 ). In contrast, for the high-similarity items, there was a significant preference for the exact-shape match during NoName trials, t (9) = 3.85, p < .01, which then significantly shifted to the ‘overall’ match for Name trials, t (9) = 4.76, p < .01 ( Figure 4 ).

Open in a separate window

Figure 4

Percent of looking to the shape match in the low-similarity and high-similarity conditions, Experiment 2

Children’s overall noun and count noun vocabularies, as measured by the CDI, were compared with their shape bias performance, as measured by their preference for the exact-shape match during the NoName trials subtracted from their preference for the exact-shape match during the Name trials across all items; no significant correlations were found.

In sum, children demonstrated a preference for the exact-shape match in the low-similarity condition and a preference for the overall match in the high-similarity condition in the presence of novel names (Name trials). The exact-shape match preferences replicate the usual shape bias effect found with simple objects; the ‘overall’ match preferences support Tversky’s model that an increased similarity among objects will lead to increased task difficulty, which, in turn, will result in a diminished preference for the exact-shape match.

General Discussion

In Experiment 1, we investigated whether adults preferred shape similarity vs. overall similarity in a novel-name extension task; we found that adults were more directed to shape with low-similarity items than with high-similarity items, regardless of the presence of a novel name. In Experiment 2, we examined the same phenomenon, this time with 2-year-olds. Like adults, the children showed a stronger shape bias with low-similarity than high-similarity items; however, for the children this effect held only during Name trials. Experiment 2, however, included different sets of low- and high-similarity items to increase the contrast between the two conditions; hence, similarity between target and exemplars was not systematically manipulated.

What was it about the items in the high-similarity set that caused children’s preference for the overall-match in Experiment 2 in the presence of names, and the decline in adults’ overall preference for the exact-shape match in Experiment 1? We suggest that the preference for the overall match in the Name trials occurred because the overall matching items in the high-similarity condition were similar to the target in both color and somewhat in shape. The increased similarity between the target and the overall-matching object in color and a less-then-perfect shape dimension reduced the salient quality of the object which had a perfect similarity to the target in shape, resulting in a preference for the overall match during the Name trials. For example, the overall-matching objects for items 2 and 3 in Experiment 1 were in fact very similar to the target objects in shape (see Figure 1 ), making transferring the name of a novel object to just one same-shape item more difficult. Moreover, in many previous studies, researchers employed simple and smooth target objects with novel shapes and only a single color, such as an inverted U-shape, coin, compact disk, or banana ( Smith, 2000 ; Imai, Gentner, & Uchida, 1994 ). When similarity is increased among objects, the contrast among them is reduced; therefore, task difficulty increases because the dimensional distinction among objects is blurred. Thus, the shape bias mechanism seems to operate at least partially as a function of the perceptual features of objects, especially in young children.

An interesting result of this study was that adults’ performance was different from children’s. Children showed a shape bias (looked longer at the shape match during the Name trial compared to the NoName trial) with objects in the low-similarity condition, whereas their preference was for the overall match as opposed to the exact-shape match in the high-similarity condition. Adults, in contrast, did not show such an interaction even though a high similarity among objects hindered their performance in both NoName and Name trials. Little word-driven shape preference was observed in adults, since the preference for shape with the items in the low-similarity condition was at ceiling in both trials (98 %), whereas novel names facilitated the children’s preference for shape with low-similarity items.

The difference between adults’ and young children’s performance suggests differences in changes in the weights of the Tverskian equation (i.e., S(A, B) = θ f (A ∩ B)−α f (A − B)−β f (B − A)) for adults and young children when the similarity relations between objects are altered. Tversky’s contrast model assumes that similarity is a function of the weighted difference of common and distinctive features shared by the object stimuli ( Tversky, 1977 ). In this model, the weights of the objects in a similarity function can change according to context, such as level of attention, saliency of object features, and task difficulty. In a shape bias task, the exact-shape match is weighted more heavily due to the high saliency of the shape feature. However, this relationship can be changed if the second test object is similar to the target not only in color but also in shape. Moreover, the weights can be altered further in favor of the overall-match by increasing task difficulty, thus resulting in a preference for the overall match. Therefore, the preference for the overall match by children in the presence of names, and the adults’ overall decreased performance, were both a consequence of this altered relationship in a task that utilized similarity judgments. For example, for the adult participants, the weight for the shape was reduced when pitted against an overall match; however, this reduction was not enough to cause adults to choose the overall match in the novel-name extension task. For children, in contrast, the weights in Tversky’s equation changed in favor of the overall match, and, hence, the young children in this study seemed to be more affected by overall similarity than adults.

A possible reason why adults are more shape oriented than young children either with complex objects or with objects that have high similarities between one another is that adults may treat shape feature as a category, whereas, for young children, shape may simply be a cue for similarity. Therefore, young children may be more susceptible to slight changes in the shared properties of the test objects because they might think they are relevant for similarity ( Son et al., 2008 ). Children detect multiple correspondences among objects by following Tversky’s approach; that is, the more overlapping features two objects have, the more similar they are to each other. In adults, on the other hand, because of the importance of shape as a strong indicator of category membership, the weight associated with shape does not decline as dramatically as in young children even after the similarity relationships among objects have been changed. In fact, it has been shown that adults represent shape more abstractly than young children (Smith, 2003), and when young children are trained with abstract shapes that are like caricatures of real-life objects, their performance increases in novel name extension tasks ( Son et al., 2008 ). Other studies have also shown that adults attend to specific object features that may indicate category membership, whereas young children rely on “global similarity” and tend to overemphasize “the way things look” ( DeLoache, 1989 ; Paik & Mix, 2006 ; Smith, 1984 ).

That children prefer shape over other dimensions in a shape bias task only with objects whose similarity to each other is kept at minimum may also suggest a processing limitation in novel name extension tasks in young children. Adults’ selective attention mechanism is tuned to picking out the most category-relevant information (in this study the shape feature), whereas young children are at a disadvantage because their immature attention mechanisms make them prone to attending to too much information or to the wrong information ( Son et al., 2008 ). Therefore, simpler objects help young children pay attention to the most salient feature (i.e., shape) thereby constraining the number of features that they are attending to.

Previous studies have shown that, in similarity tasks, children perform better with simple objects than with objects that are richly detailed or have many features. These findings may, in fact, indicate constraints in attention and memory skills. For example, Rattermann, Gentner, & DeLoache (1990) tested 3- and 4-year-olds and adults with two sets of objects in a similarity task. One set contained simple objects that differed only in size, and the other set included richly-detailed objects. Compared to adults, children in both groups performed better with simple objects. Similarly, Son et al. (2008) found that 2-year-olds generalize better when presented with objects that contain simpler features or fewer features than objects having many features.

Our results suggest a developmental change in the perception of shape and how context (i.e., similarity relationships between objects) affects shape perception in young children and adults. If 2-year-olds and adults differ in their performance in novel-name extension tasks, a significant question arises as to around what age young children’s performance becomes adult like. Although further research is needed in this area, some studies have shown that children under five years of age classify on the basis of overall similarity rather than on a single attribute such as color or size ( Smith, 1984 ). Smith and Evans (1989) gave 2-, 3-, and 4-year-olds three sets of objects and asked them to classify similar or identical objects into two groups. In the first set, all the objects within a group were similar to each other and the between-group dissimilarity was high. In the second set, the objects in one group were identical to each other, and again, the between-group dissimilarity was high. In the third set, one group of objects to be classified included identical objects, but the objects in the other group were also similar to the objects in the first group; thus, between-group dissimilarity was low. Children classified all identical and similar objects together, which, according to Smith and Evans (1989) , shows that young children were attempting to classify objects by ‘overall similarity’ by ignoring the dimensions that separated identical objects from similar ones. Similarly, in our study, the decrease in performance with high-similarity items in young children could be due to an attention bias to overall similarity and ignoring of category specific information (i.e., shape).

The present studies have shown that when similarity among objects in a shape-bias task is increased, both 2-year-olds’ and adults’ preference for shape decreases. Furthermore, young children seem to be more vulnerable to a change in similarity relations among objects. When objects are named, children’s preference for the exact-shape match declines dramatically; instead, they prefer the object that has an overall resemblance to the target object. Our findings indicate that adults may treat shape as a category, which explains why adults still preferred the shape with or without the presence of names, whereas young children are more driven by an overall similarity among objects. Further research is needed to delineate what other object properties can also influence similarity judgments in a novel-name extension task, and how this may changes with development.

Acknowledgments

Footnotes

References