Principle-based inferences in young children's categorization: Revisiting the impact of function on the naming of artifacts

Cognitive

Development,

10, 347-380

(1995)

Principle-Based Inferences in Young Children’s Categorization: Revisiting the Impact of Function on the Naming of Artifacts Deborah

G. Kemler Nelson

and 11 Swarthmore

College

Students*

Swarthmore

Three

parallel

unprincipled

studies

investigated

the

influence

of principle-based

College

inferences

similarity relations on new category learning by 3- to 6-year-old

and

children.

One of two possible functions of a single novel artifact (which differed between studies) was modeled

for and practiced

same name as the original. ance from the original,

by children,

were designed such that each could be inferred to afford only

one of the two possible functions. observed.) tion

had

principle

(Actual functions of the test objects were not directly

Patterns of categorization been

experienced,

depended

indicating

to guide their extension

learned categories

who then judged which test objects got the

Test objects, either globally similar or dissimilar in appear-

of the name.

may activate self-initiated,

dren. The conditions discussed, although

systematically

that the children Therefore,

principle-based

that prompt such categorization a role for unprincipled

on which original

func-

used a common-function categorization

into newly

reasoning in young chil-

processes in young children

are

similarity relations is not denied.

The studies presented here are concerned with how children decide on the extension of the name of a novel artifact after exposure to a single instance of the category. Do the objects’ functional affordances have an impact on

*The 11 authors, Swarthmore College students when they contributed to this research, are: Laura Almasy, Kevin Crowley, Nell Duke, Julie Anne Gardner, Virginia Kiggins, Kristina Lasher, Annie McQuilken, Michael O’Connell, Rachel Russell, David Sterner, and Elizabeth Tirk. Nell Duke and Rachel Russell collected the data for Experiment 2, and Rachel Russell, supported by a Joel Dean Summer Research Grant, undertook the primary task of coding the verbalizations from that experiment. We express our appreciation to Don Reynolds for his remarkable ability to translate our outlandish ideas into real, functioning stimulus materials. Our thanks also to the staff and children at Trinity Cooperative Day Nursery and Swarthmore Presbyterian Nursery Day School. D.G. Kemler Nelson was supported by a Lang Faculty Fellowship for part of this research. Portions of these data were reported at the Meetings of the Society for Research in Child Development, Seattle, WA. (April, 1991) Correspondence and requests for reprints should be sent to Deborah G. Kemler Nelson, Department of Psychology, Swarthmore College, Swarthmore, PA 19081-1397. Manuscript

received

July 28, 1994;

revision accepted

April 11, 1995

347

348

Deborah G. Kemler Nelson

children’s tendency to accept them as instances of the same kind? The issue is not new. Over a decade ago, several related investigations were framed by an attempt to distinguish between the implications of Nelson’s (1974) functional-core theory of word meaning and the proposal put forth by Clark (1973) that young children generalize on the basis of salient perceptual properties. In part because those studies bore an uncertain relation to the original theoretical controversy (Nelson, 1983), and in part because they largely converged on the conclusion that function was unimportant to young children (see the review by Gentner & Rattermann, 1991), this line of research trailed off over more recent years. However, we believe that there is now good reason to resurrect the question of the role that functional affordance plays in young children’s generalization of new words for artifacts. One kind of motivation comes from the recent literature on human concepts and categorization, in which the role of principle-based or knowledge-based processes has been highlighted as an important issue in understanding categorization in adults (Medin, 1989; Murphy & Medin, 1985; Rips, 1989). A second kind of motivation comes from recent emphasis on the domain-specificity and fragility of young children’s developing competence, and the suggestion that if, and only if, one looks in the right places and in the right ways (Carey, 1985a; Gelman, 1982), young children’s performance (and the psychological competence that it implies) can be shown to more closely resemble that of adults. The first of these lines of inquiry puts into a new and interesting light the question of when in development the (intended) function of an artifact becomes a principled basis for generalizing the name of that object category, and the second of these lines suggests that early invocation of such a principle is likely to be found in content domains in which children have sufficient background knowledge. There is broad agreement that, among adults at least, the function intended by the designer and/or conventionally exercised by users is “central” or “core” or “essential” to the concept that underlies categories of artifacts (e.g., Miller & Johnson-Laird, 1976). To argue for the privileged status of functional information in artifact concepts, investigators typically have assessed whether knowledge of the functional affordance or intended function of the artifact can override other kinds of potentially salient, but conflicting information about the category to which the artifact belongs. For example, Keil(1989) presented pictures and stories that described variants of common artifacts (tools such as screwdrivers and hammers) that, in one case, looked very much like the conventional artifact but lacked the same functional affordance and, in another case, lacked the conventional appearance but still functioned in the critical way. Adults (and fourth graders) opined that the former were not members of the named artifact category and that the latter were. Analogously, Rips (1989) showed that information

Principle-Based Categorization

349

about the intended function of an artifact was more critical to how adults judged its category membership than was its surface appearance (even when the latter counted more in assessments of similarity relations). At least for adults, then, the functional affordance of an artifact seems to play a very powerful role in inferences about category membership. Indeed, some investigators understand this role in still larger terms as indicating that causal accounts framed by explanatory principles or “theories” determine the nature of mature human categories (e.g., Carey, 1985b; Keil, 1989; Medin, 1989; Murphy & Medin, 1985). As related to categories of artifacts, an intention to design an object that functions effectively in a certain way would account for a number of additional properties of the object. Seen in this light, the central role of intended function within artifact categories is at least partially akin to the central role of genetic origins and other biological principles within categories of living things (Medin, 1989; Rips, 1989). From a developmental perspective, there are several reasons to speculate that the principle of intended function-or at least conventional function-as it relates to artifacts might be learned earlier than principles such as “biological origins” as it relates to animals. First, the conventional function of an artifact is directly observable. For familiar artifacts, the child has plenty of opportunity to perceive that objects that get the same name function in similar kinds of ways across a variety of people and situations (Rosch,Mervis, Gray, Johnson, & Boyes-Braem, 1976). For example, there are distinctive and easily observable behaviors that commonly occur in interaction with things called “telephones” or things called “balls.” Second, young children are highly attentive to functional information (e.g., Corrigan & Schommer, 1984; Gentner, 1978; Kolstad & Baillargeon, 1990; Mervis, 1987; Nelson, 1973). Third, for many but the most complicated artifacts, there is a close relation between perceptible structure and conventional function within a basic artifact category (Mervis, 1987; Rosch et al., 1976; Tversky & Hemenway, 1984). Thus, the perception of salient structural similarity within a category of artifacts can ordinarily serve the child as a useful indicator of or pointer to common underlying function (Medin, 1989; Smith & Heise, 1992). Finally, for at least some categories of artifacts (e.g., containers-Kolstad & Baillargeon, 1990; things that push or pull-Brown, 1990) children seem to be sensitive early on to mechanisms of physical causality that mediate between the structure of the objects and their function. However, given that structure and function tend to be so highly correlated in basic-level categories, young children could categorize artifacts much like adults, but for a different reason: Whereas functional affordance may be fundamental to mature artifact categories, approximately the same categories may be carved out by children on the basis of common shape (Baldwin, 1992, Imai, Gentner, & Uchida, 1994; Jones, Smith, & Landau, 1991; Landau, Smith, & Jones, 1988) or salient parts (Tversky & Hemenway,

350

Deborah G. Kemler Nelson

1984). Given such convergence, some investigators (e.g., Landau et al., 1988; Tversky & Hemenway, 1984)-but by no means all (e.g., Mervis & Greco, 1989; Nelson, 1983)-have reasonably doubted that functional affordance plays a major role in young children’s concept learning. In fact, there exists a rather substantial body of studies that fall in line with this view. Some of the work pertains to children’s everyday lexical concepts. For example, the same depictions of tools (e.g., screwdrivers) that were systemati~aIly categorized according to functional affordance by fourth graders and adults elicited categorization based on clusters of functionally-irrelevant characteristic features in kindergartners (Keil, 1989). Young children also failed to detect the absence of small, but functionally relevant parts in pictures of common artifacts, such as shoes (missing laces), cars (missing headlights), and scissors (missing a connecting screw) (Tversky, 1989). Moreover, preschoolers showed relatively little effect of depictions of different functional contexts on their categorization of a continuous series of line drawings as glasses, cups, or bowls (Prawat & Wilfong, 1980). Still, rather than suggesting a general tendency of children to ignore functional information in categorization, these findings may hinge more on children’s lack of specific knowledge of the artifact categories (or category distinctions) that were tested, particularly their faiiure to command the causal principles that explain why certain parts or features are critical to how these artifacts work. There is widespread agreement that when reIevant knowledge or understanding is lacking, unprincipled similarity represents a useful default option for guiding categorization. This is a generalization that applies as readily to adults as to children (e.g., Keil, 1989; Medin, 1989). In further studies of the principles behind young children’s categorization of artifacts, investigators have observed how children learn new, artificial concepts in the laboratory. In a classicstudy by Gentner (1978)children were first exposed to two novel objects, which had two different and salient functions, as well as two rather different appearances. The “‘jiggy” was a blue and yellow square box with a cartoon face mounted on the side; operating an attached lever caused the face to ehange expression. The “zimbo” was a modified gumball machine with a clear globe filled with jellybeans; operating an attached lever caused jellybeans to drop. When presented with a hybrid object that looked identical to the jiggy but which dispersed gumballs like the zimbo, 2- to 5-year-olds were likely to call it a “jig,” as if appearance mattered far more than function in generalizing the new names. In contrast, 5- to 15year-olds were more likely to use function as a basis for categorization. Surprisingly, in Gentner’s [‘1978) experiment, adults-when forced to choose--categorized more like the preschoolers. In light of the critical role that intended function has played in adults’ categorization in other studies, the fast result gives us some pause: The fact that the originai jiggy and the test object were totally identical in every detail of their complex perceptible

Principle-Based Categorization

351

may have detracted from the ecological validity of the test that Gentner arranged: It is difficult to think of a real-world case in which two complex objects are identical on their surface but are designed to perform entirely different functions. Moreover, in such a highly unusual case, by definition, function is totally uncorrelated with aspects of observable structure: The structures that supported the distinctive functions of the jiggy and zimbo were hidden from view and unknown to the subjects. Tomikawa and Dodd (1980) also designed artificial objects in order to pit function against structural appearance. Theirs were simple wooden toys, consisting of all possible pairings of three different shapes and three different functions. Preschool children not only sorted the objects according to shape rather than function, but, more importantly, they far more easily learned to name the objects in accord with their shape identities than their functional identities. The latter result emerged, however, in a somewhat peculiar procedure in which the names of the objects were introduced in a story context that treated the objects as “characters . . . (who) have various exciting adventures” rather than as the functional artifacts they were designed to be. Such a method does not seem fitting for assessing the role of functional affordance in the development of new word meanings for artifacts. Moreover, in these studies, as in Gentner’s, observable structure and function were almost entirely uncorrelated, such that structure provided very few clues to function. Indeed, there are hints that, under some conditions, the functional affordances of artifacts can play a more prominent psychological role than many of these other studies have indicated. Much of the relevant work with very young children has not dealt with category learning per se, but has focussed on other indices of the development of equivalence classes. Kolstad and Baillargeon (1990) showed that lO%month-olds, familiarized with cylinders functioning as sand-transporting containers, generalized less (i.e., dishabituated more) to an analogous event involving a perceptually similar but dysfunctional object (a cylinder apparently lacking a bottom) than to one involving a perceptually dissimilar but functional object (a different shaped container with a bottom). Thus, even before 1 year of age, the inferred functional affordances of objects as containers-based on the perceived presence of a bottom-can be so salient as to override attention to other functionally irrelevant properties. It is admittedly a leap, but not an implausible one, to the hypothesis that the functional affordance of containment will also play a prominent role when later, infants learn and generalize names for artifacts that conventionally function as containers. Such speculation is prompted by the larger hypothesis that when children have some knowledge of the mechanisms that mediate between the structure and the function of a domain of artifacts, functional affordance will play a privileged role in the concepts and categostructure

352

Deborah

C. Kemler Nelson

ries they form. Of course, such knowledge will be gained at different times for different kinds of artifacts (as well as at different times for different functionally-relevant features of the same artifact). Investigations by Brown (1990) and her colleagues further suggest the role that domain-specific knowledge of structure/function relationships can play in directing the attention of very young children to the functional affordances of artifacts. Here again, the concern is not directly with categorization, but the results nevertheless suggest that functional affordance will be on the mind of even 2-year-olds under some conditions that call for generalization. In collaboration with Slattery, Brown (1990) showed that in a problem-solving situation, toddlers will selectively transfer a solution acquired with one tool to other tools that afford the same function when the problem is one that calls upon knowledge of causal principles that toddlers possess. Thus, having learned to retrieve a prized object with a rake, 2-yearolds attempted the same action with other dissimilar-looking artifacts that afforded retrieval, but not with otherwise similar-looking artifacts that lacked the rigidity or shape necessary to perform the action. The studies presented here do not take off directly from either the work of Kolstad and Baillargeon (1990) or that of Brown (1990) and her colleagues, but they are inspired by the kinds of insights that these studies promote. We have undertaken a new look at the question of whether categorization by young children might show a principled basis, specifically whether the names of new artifacts might be generalized on the basis of functional affordance. We have attempted to reopen the question under conditions in which the causal relations between the structural properties and the functional affordances of the objects are likely to be understood by the children. Moreover, in view of the substantial empirical literature that argues against the importance of functional information in young children’s categorization, we have attempted to provide a particularly rigorous assessment of the alternative. This was done by creating a novel object (to which the new name was given) that afforded two different functions (only one of which was demonstrated to an individual child). Test objects for assessing category generalization always differed from the original in a way that preserved one of the functional affordances but undermined the other. Accordingly, any tendency of the children to invoke functional affordance as a principled basis for categorization could be plainly detected in complementary patterns of generalization by two groups of children exposed to one of the two different functions. Such an experimental design has the advantage of allowing us to probe for use of the function principle without having to strictly control for (or estimate) other bases for similarity between the original object and the text exemplars. However, at the same time, we did try to systematically vary the degree to which functionally irrelevant aspects of similarity suggested a relation be-

Principle-Based

353

Categorization

tween the members of the test set and the original object.Typically,those who have doubted that young children are attuned to functional affordances in artifact categorization have suggested that they fall back instead on commonalities based on “unprincipled similarity” (Keil, 1989) or “surface similarity” (Gentner & Toupin, 1986) or “perceptual similarity” (Tomikawa & Dodd, 1980) or shape (Landau et al., 1988) or perceptually salient parts (Tversky, 1989). (In many cases, including ours, these relations tend to be redundant with one another.) Half of our test objects were designed to be rather similar overall to the original object and half were relatively dissimilar to it. Thus, it was possible to probe for the role of unprincipled similarity at the same time that we could provide a stringent test for the function principle. Note that our design does not pit functional affordance directly against similarity (e.g., Gentner, 1978; Merriman, Scott, & Marazita, 1993), but instead allows that the influence of each factor can be detected independently. Given our theoretical concerns, the most critical element in the design of the test objects was varying functional affordance in ways that were likely to be compatible with young children’s knowledge of physical causality. Accordingly, in modifying the original object to produce the test objects, we made an attempt to manipulate structural elements that were not only perceptible to the children but also bore a rather simple, transparent relation to the functional affordance that was relevant. For example, when the critical function depended on the capacity of the artifact to hold water, one kind of modification consisted of substituting a solid cylinder for the original cylindrical container. When the critical function depended on the capacity to roll, one kind of modification consisted of substituting a pyramidal shape for the cylinder. Other kinds of modifications consisted of the elimination of small parts that were functionally essential, the addition of parts that blocked function or the repositioning of parts in ways that would undermine a function that had been demonstrated. Three studies are reported in this article. Experiments la and lb were small-scale studies undertaken by undergraduate students in the first author’s Seminar in Developmental Psychology. They were parallel investigations that used different kinds of stimulus materials. Experiment 2 represented a larger effort that followed upon Experiment lb, which permitted systematic tests for age effects within the preschool range, and which resulted in a more successful attempt to elicit children’s explanations as well as their naming responses.

EXPERIMENT

la

Children were introduced to an artifact, a “bindle,” that either could be used to roll designs in playdough or to transform the color of the liquid when water was made to pass between its two chambers. Only one function was

Deborah C. Kemler Nelson

354

demonstrated for and modeled by any individual child. After seminaturalistic interaction with the object, the child was asked of a series of test objects whether it was or was not a “bindle.” Half of the test objects afforded only the Roller function and half afforded only the Color-Changer function. Method Participants. Sixteen children participated in the study, (M = 5; 5, r = 4; 6-6; 10). All were attending a suburban preschool or kindergarten. Eight children were assigned to the Roller group and 8 to the Color-Changer group. Two additional children experienced the procedure, but did not contribute data for the analyses because of response biases: one rejected all the test objects as members of the category, and one accepted all of them. Materials. The original object was a transparent, two-chambered plexiglass cylinder with a handle that could be used to (a) roll designs in playdough, and (b) transform the color of liquid (when the liquid was passed between its chambers). A black-and-white photograph of the object is provided in Figure 1. To afford rolling designs, a plastic-covered, wire spiral was glued around the outside of the cylinder, which made impressions in the playdough when the cylinder was rolled over it. To afford liquid colorchanging, a small plunger at the end of the cylinder could be depressed to

Figure 1.

A black-and-white

photograph

of the original bindle.

Principle-Based

355

Categorization

force colored water to pass from the top chamber to the lower chamber. Colored powder (a mixture of baking soda and food coloring) in the lower chamber then effected a change in the color of the liquid that passed into the chamber. A small pipe, similar to the one enveloping the plunger, allowed air to pass between the chambers, and plugs on both ends permitted the experimenter to fill and empty the chambers of the cylinder. Of the 8 test objects, 4 were designed to look globally similar to the original, and 4 were designed to look more dissimilar. Within each set, 2 could function as Rollers, but not as Color-Changers, and 2 could function as Color-Changers, but not Rollers. For example, for one Similar, +ColorChanger/-Roller test object (hereafter called a “Color-Changer”), the handle was attached directly to the plexiglass cylinder such that the object could not function as a Roller, but easily afforded the color-changing function. For one Dissimilar, +Roller/-Color-Changer test object (hereafter called a “Roller”), the plexiglass cylinder was replaced with a solid wooden part of exactly the same shape, but which could not be filled with water to perform the Color-Changer function. Further details on the characteristics of the test objects are given in Table 1. Ratings of 8 college students confirmed that the four test objects designated as Similar “looked more similar” to the original object than the four designated as Dissimilar. Without any information about the objects’ funcTable 1. Descriptions of the Test Objects Used in Experiment la. Similar and Dissimilar Objects are Designated as S and D Respectively. Rollers and Color-Changers are Designated as R and C Respettiveiy. Mean Similarity Rankings of Adults are Also Shown.

Object

Characteristics Distinguishing the Test Object From the Original Bindle

Similarity Ranking

SRI Piugs removed and access to plunger impossible because the cylinder is SR2 SC1

SC,

W

=%

‘W

entirely sealed. Small holes present in the plexiglass cylinder such that it cannot hold liquid. Handie attached to the cylinder itself such that the cylinder will not rotate when rolled. Circular, plexiglass ends of the cylinders are enlarged so that the wire spiral will not touch the surface when the cylinder is rolled. An unpainted, solid wooden cylinder, which cannot serve as a liquid container, replaces the plexiglass cylinder. Three unconnected, transparent, disc-like segments, each filied with confetti, are mounted on the axle attached to the handle in place of the single twochambered cylinder. A pyramid-shaped, two-chambered body constructed of transparent plexiglass in place of the plexiglass cylinder. The plexiglass cylinder, turned on its end, stands on a set of fin-like triangular projections that are wooden and green and attached at the cylinder’s bottom.

1.75 1.75 3.88 3.38 5.25

5.00

8.00 7.00

356

Deborah G. Kemler Nelson

tions, these adults were asked to rank the test objects’ similarity in appearance to the original. Their average rankings for each of the objects are shown in Table 1. Rollers tended to be judged more similar to the original than Color-Changers, but-far more important-every single rater ranked both Similar objects of each type as more similar to the original than both Dissimilar objects of the same type. Indeed, 6 of the 8 judges rated every Similar object as more similar to the original than every one of the Dissimilar objects. Overall, the objects designated as Similar received a mean ranking of 2.69, as compared to a mean ranking of 6.31 for those designated as Dissimilar, and each of the Similar objects had a mean ranking lower than each of the Dissimilar objects. Hence, our manipulation of unprincipled similarity appears to have been highly successful.

Procedure Each child was exposed either to the Roller function or to the Color-Changer function of the original object-never both. The object was introduced as a “Bindle-something that my friend asked me to show to children at your school.” The experimenter modeled and described the bindle’s function while repeatedly using its name. The child was told, “Let me show you what I can do with this bindle.” However, the functional information was never given as an explicit definition of the novel word. To demonstrate the function in the Roller condition, the experimenter held the object by the handle and rolled out designs in a sheet of playdough. In the Color-Changer condition, the object was set into a stand, and colored water was poured into the upper chamber; then, a plunger was pulled, allowing the water to flow into the bottom chamber, where it mixed with food coloring and changed color. Following the experimenter’s demonstration, the child was encouraged to try the bindle out, if necessary with the experimenter’s help. During this phase, the experimenter continued to make frequent reference to the object as a “bindle,” while commenting on what the child was doing. After the child’s seminaturalistic interaction with the object, the experimenter removed it from view, and then, one at a time, brought out each of the test objects. The child was told, “My friend also sent these other things. I think that some of them are bindles and some of them are not bindles. I’d like you to tell me what you think.“For each object, the experimenter asked, “Do you think that this is a bindle or that it’s not a bindle?” Each time, the child was allowed to inspect the object, but was firmly discouraged from trying it out before responding. At the time of the test, the playdough was not in sight and there was no liquid or powder in the chambers. For each child, the first sub-block of four test objects consisted of 2 Similar and 2 Dissimilar items; in each case, one of the two afforded the Roller function and one afforded the Color-Changer function. The original object itself was queried on the fifth test trial, followed by a brief opportu-

Principle-Based Categorization

357

nity to use the object again. Then, the remaining 4 test objects were presented in series. Across subjects, each test object occurred equahy often in the first and second sub-block. The order of objects within each sub-block was randomly determined for each subject, except that, due to an administrative error, the two similar objects were presented prior to the two dissimilar objects. Following the 9 test trials, the experimenter re-presented each of the 8 test objects in the same order. After reminding the subject of his or her previous categorization decision, the experimenter asked the child to explain it. Analyses of these explanations are not included in the current report because relatively few of them were elicited. However, in the very few cases (6 responses out of 128) in which the child changed his or her categorization decision for an object during this phase, the final decision was used in the analyses of the data. Results and Discussion

All children judged that the original object was a member of the kind; i.e., was a “bindle,” when it was queried during the test phase. With regard to categorization of the test objects, Figure 2 shows the proportion of responses that constituted acceptances of the objects as a “bindle.” These proportions are computed separately for each exposure group (Roller or Color-Changer), according to whether the objects were globally Similar or Dissimilar to the original object, and according to whether the objects could be inferred to function as the original object had. Each proportion is based on 8 children, each contributing two responses. Because the same physical objects served as the stimuli for the two exposure groups, but the demonstrated function of the original object differed, the +Function test objects for one group served as the -Function test objects for the other group. As Figure 2 depicts, the children’s categorization decisions were affected by the objects’ functional affordances. Overall, +Function objects were accepted 72% of the time as members of the kind, whereas -Function objects were accepted 39% of the time. Eleven of 16 children were more likely to accept +Function than -Function objects, and no subject showed the opposite pattern. Moreover, 7 of the 8 distinct physical objects were more likely to be accepted as category members when they were +Function than when they were -Function. The exception was one of the Similar objects, the Roller with small holes. Indeed, as can be seen in Figure 2, the effect of functional affordance tended to be far more marked for the Dissimilar objects. It is noteworthy that not a single Dissimilar object was ever included in the category when the original functional affordance was eliminated, but these same objects were accepted over 50% of the time when the functional affordance was preserved. The uniform tendency to reject Dis-

Deborah C. Kemler Nelson

358

fn

5 8

0.8

E 8

0.6

: E

0.4

$ 8

0.2

n

+ FUNCTION 81 - FUNCTION

P 0.0 Roller Group

Color-changer Group

Similar Objects

Roller Group Dissimilar

Color-changer Group Objects

Figure 2. Generalization patterns in Experiment la. For each exposure group, the proportion of acceptances of test objects as bindles, according to whether they were globally similar or dissimilar from the standard, and according to whether they could be inferred to function as the standard had. The +Function test objects for one group served as the -Function test objects for the other group. The absence of -Function bars above the dissimilar objects indicates zero acceptances in these cases.

similar-dysfunctional objects is also the primary basis for the effect of object similarity: In contrast to the total lack of acceptances of Dissimilar-dysfunctional objects, Similar-dysfunctional objects were accepted 78% of the time. Thus, in Experiment la, there seems to be very clear evidence that preschool children’s categorization of new artifact kinds is partially based on principled reasoning about whether the object can function as the named object does. Additionally, there is a suggestion that such reasoning comes into play particularly when the new object is not too similar in appearance to the original. However, the latter effect must be interpreted with a good deal of caution, given the methodology of the study. It seems unlikely, but not impossible, that the unintended procedure of presenting Similar objects on average earlier in the test phase than Dissimilar objects was alone responsible for the trend. A more compelling caution arises from a different source-namely, the possibility that some of the changes embodied in the Similar test objects were so subtle that the children failed to notice them at all. We arranged a follow-up session, within 2 weeks of the first, in which we asked 12 children who had accepted all the Similar objects (5 in the Roller group and 7 in the Color-Changer group), to indicate the difference between each of the Similar objects and the original bindle, while both were in full view. They were able to do so correctly (by pointing or verbalizing) only 58% of the time.

Printipfe-BasedCategorization

3.59

Particularly since their naming responses had been made without the same benefit of a simultaneous comparison, these children often may have failed to notice the critical differences when they were categorizing. In addition, the follow-up evidence alerted us to the probability that the ways functions were undermined in the Similar objects were less well understood by the children than the simpler ways functions were undermined in the Dissimilar objects, In retrospect, it is clear that in the design of the Dissimilar objects, we took advantage of the greater leeway of the dissimilarity relationship and found ways of undermining function more straightforward than for the Similar objects Compare, for example, for the Roller condition, the impact of changing the cylinder to a pyramid (Dissimilar object) as opposed to attaching the central handle directly to the cylinder (Similar object). When we pointed out the critical structural differences within the Similar objects to children in the follow-up session, it had some impact on their categorization (19 out of 24 +Function acceptances as opposed to 12 out of 24 -Function acceptances), but the fact that the effect was not larger is also striking in light of the robust function effect originally observed for the Dissimilar objects. It appears that the simplicity of the structure/function relation was less optimal for the Similar objects than the Dissimilar objects. This confound was eliminated in the design of the materials for Experiment lb.

Experiment lb was partially designed to address these concerns, Not only was the order bias eliminated, but the Similar and Dissimilar objects were matched with each other for the particular ways in which functional affordances were undermined. An entirely new set of stimulus materials was developed for the study. Thus, Experiment lb provided a good opportunity to test the generality of the main finding of Experiment la, namely that preschoolers’ category learning is sometimes guided by inferences about functional affordances. The original artifact used in the experiment was a T-shaped metal object with attached brushes and wires which could be used as a painter to paint four parallel lines or could be used as a musical instrument when three tunable, wire strings were plucked, Similar-Looking and dissimilar-looking test objects either afforded the Painter function or the Musical Instrument function, but not both. Whichever the function disabled in the test object, it was eliminated in one of two well-defined ways: (a) either by substituting a dysfunctional part for a small, critical part or (b) by repositioning a smalf, critical part in a location on the object that rendered it nonfunctional. Each of these transformations was instantiated in a Similar and Dissimilar Painter and in a Similar and Dissimilar Instrument. Hence, the kind of causal

360

Deborah

G. Kemler Nelson

reasoning necessary to infer the functional affordance of the test object was balanced across both object type and object similarity. Experiment lb included a group of college students, as well as a sample of preschoolers. This enabled a direct test of the claim that functional affordance is crucial to adults’ categorization of artifacts using the same materials and the same procedural context that were used with the children. In light of Gentner’s (1978) failure to obtain such a pattern in her jiggy/zimbo study, such a direct test was deemed to be particularly useful. Method Participants. Sixteen children participated in the study (M = 4;8, r = 3;9-5;5). Thus, they were somewhat younger than the children of Experiment la. All were attending a suburban preschool, and none had participated in the previous experiment. Eight children were assigned to the Painter group, and 8 to the Instrument group. In addition, 8 adults were administered the same procedure as the children; 4 of them were in the Painter condition, and 4 were in the Instrument condition. Materials. The original object could be operated as both a parallel-line Painter and a stringed musical Instrument. A diagram in Figure 3 shows its essential features. The skeleton of the object was constructed of two rigid, metal bars joined in a T-shape, such that one could serve as a handle and the other as a crossbar. Plastic guards covered the end of the handle and one end of the crossbar. Four paintbrush tips were aligned at the top of the crossbar and could be dipped simultaneously into a row of paints, and then used to paint four parallel lines on paper. Three wire strings were connected between the crossbar and the handle in a fan-like configuration. The tension of the three strings (and hence, the pitch the strings produced when strummed) could be adjusted independently by means of three metal tuners located between the paintbrushes on the crossbar. A description of the 8 test objects is given in Table 2. Diagrams of 2 of them are also shown in Figure 3. The 8 test objects were designed such that each object afforded one and only one of the possible functions of the original. For one pair of +Painter/-Instrument objects (hereafter called Painters), the metal strings were removed and replaced by parts that could not be used to produce music. For the other such pair, the strings were embedded in one of the objects’ handles such as to be virtually impossible to pluck. For one pair of +Instrument/-Painter objects (hereafter called Instruments), the paintbrushes were replaced by elements that did not permit painting. For the other such pair, the brushes were placed in locations that made it impossible to paint with them. Within all pairs, degree of

361

Prifl~jpl~-~as~ Categorization

A.

The originaf stennet

Deborah G. Kemler Nelson

362

Table 2. Descriptions of the Test Objects Used in Experiments lb and 2. Simitar and Dissimilar Objects are Designated as S and D Respectiveiy. Instruments and Painters are Designated as I and P Respectively. Objects Subscripted With a 1 Were Made Dysfunctional by Repositioning a Critical Part; Those Subscripted With a 2 Were Made Dysfunctional by Replacing a Critical Part. Mean Similarity Ran~ngs of Adults are Also Shown.

Object

Characteristics Distinguishing the Test Object From the Original Stennet

Similarity Ranking

Brushes mounted within the crook of the T-frame so that they are inaccessible. Strings inlaid within grooves in the handle so that they are inaccessible. Strings of small, colored beads replace the brushes. A fan-shaped piece of gauze replaces the fan of strings. A solid, star-shaped, orange, wooden base on top of which strings and tuners are mounted. Brushes mounted within the crooks of the star-shaped base so as to be inaccessible. A blue, wooden, rounded handle with the brushes mounted into metal fork-like, projections and aligned at the top of the handle. Strings are embedded in the wooden handle so as to be inaccessible. A green, metallic, open rectangular frame on which the strings and tuners are mounted. Red pegs with blue caps replace the brushes and are mounted near the tuners at the top of the frame. A red, lattice-like, wooden rectangular body with brushes mounted in alignment at one end. Decorative yellow cord is threaded through an open part of the lattice to replace the strings.

2.38 1.50 3.00 3.12

7.38

5.62

6.12

6.88

similarity was manip~ated by modifying or not modifying functionally irrelevant properties of the original object. So, for example, one Similar object looked just like the original except that short strings of beads replaced the paintbrushes. For the Dissimilar object in that pair, not only were the brushes replaced, but also the object was entirely different in shape and color: The strings and tuners were mounted on a green rectangular frame. As a check on the intended similarity relations, an additional group of 8 college students, none of whom participated in the main experiment, was asked to rank order the 8 test objects in their similarity of appearance to the original. The rankings were elicited with all stimuli in view, including the original, and without any mention of functional affordance (actual or intended). The mean similarity ranking for each of the test objects is shown in Table 2. Individually, every adult judged each of the 4 test objects designated as Similar as resembling the original artifact more than each of the 4 Dissimilar test objects. Accordingly, the mean ranking for ail Similar objects was 2.50 and for all Dissimilar objects was 6.50. The intended manipulation of unprincipled similarity relations seems to have been totally successful. There seem to be no other systematic effects in the similarity rankings save for a hint that an object created by repositioning a critical part tended

Principle-Based Categorization

363

to be perceived as more similar to the original than its comparable test object created by replacing a critical part with another one (the DI objects being the exception). However, the magnitude of this difference-if reliable at all-is very small compared to the magnitude of the intended distinction between Similar and Dissimilar objects. Each child (and adult) was exposed either to the Painter Procedure. function or to the Instrument function of the original object-never both. The object was introduced as a “stennet.” The experimenter modeled and described the stennet’s function while repeatedly using its name. The subject was told, “Let me show you what I can do with this stennet.” As in Experiment la, the functional information was not given as a definition of the name. To demonstrate the function in the Painter condition, the experimenter brought out a stand loaded with a row of four jars of different colored paint, and a large piece of blank paper. The experimenter then dipped the stennet into the paints and made a design on the paper. Subsequently, the participant was encouraged to paint his or her own designs and the experimenter continued to make comments that included reference to the object as a Ystennet.” To demonstrate the function in the Instrument condition, the experimenter brought out a guitar pick and plucked out a simple melody. The child was subsequently encouraged to produce melodies on his or her own, and the experimenter continued to refer to the “stennet” in casual remarks about the child’s actions. The seminaturalistic interaction in this phase lasted for approximately 3 min, and was followed immediately by the test phase. As in the earlier experiment, the test phase consisted of two cycles through the objects. In the first cycle, the subject was asked whether each of the 8 test objects as well as the original artifact “is a stennet or is not a stennet.” Responses were elicited without giving the subject the opportunity to try out the function of the object; as previously, any conclusion about its functional affordance had to be spontaneously inferred by the participant. Neither paints and paper nor a guitar pick were present during the test phase. On the first four test trials, each subject was questioned about two Similar objects, one that afforded painting and one that afforded music-making, and two Dissimilar objects analogously balanced. The set of four was chosen with the restriction that Similar and Dissimilar objects in which the same function was disabled were disabled in different ways. Across subjects, each possible set occurred equally often in the first block of trials. The order of the four objects within the set was determined randomly for each subject. On the fifth trial, the original object was queried, followed by a brief opportunity to try it out again, and the remaining objects, in random order, were presented on trials 6 to 9.

Deborah G. Kemler Nelson

364

In the second cycle, the 8 test objects were presented in the same order that they had first been shown. The children were reminded of their original categorization responses, and asked to explain them.These verbalizations are not reported here, since they were few in number, but on the rare occasions that the children changed their naming responses at this point (2 times of a possible 128), the final response was used in the analyses of categorization. Results and Discussion

All of those who participated judged that the original object was a member of the category; i.e., was a “stennet,” when it was queried during the test phase. Preschoolers’ Responses to the Test Objects. With regard to categorization of the test objects, Figure 4 shows the proportion of responses that constituted acceptances of the objects as a “stennet.” These proportions are shown separately for each exposure group (Painter or Instrument), according to whether the objects were globally Similar or Dissimilar to the original object, and according to whether the objects could be inferred to function as the original object had. Each proportion is based on 8 children, each contributing two responses. Recalf that since the same objects served as the stimuli for the two exposure groups, but the function of the original object differed, the -t-Function test objects for one group were the -Function test objects for the other group.

6 ‘_ r

0.4

n + FUNCTION q - FUNCTION

0.2 Q

Painter Group

Instrument Group

Similar Objects

Painter Group

Instrument Group

Dissimilar

Objects

Figure 4. Generalization patterns in Experiment lb. For each exposure group, the proportion of acceptances of test objects as stennets, according to whether they were globally similar or dissimilar from the standard, and according to whether they could be inferred to function as the standard had. The -f-Function test objects for one group served as the -Function test objects for the other group.

365

Principle-Based Categorization

As Figure 4 depicts, the children’s categorization decisions were mediated by inferences about the objects’ functional affordances. Overall, i-Function objects were accepted 68% of the time as members of the kind, whereas -Function objects were accepted only 28% of the time. Twelve of the 16 children accepted more -+-Function than -Function objects, and only one showed the reverse pattern. Moreover, every one of the 8 physical objects was accepted more often by the group for whom the original funo tion was afforded than the group for which it was not. The similarity of the test object to the original also seemed to have some effect. Overall, 55% of the Similar objects were accepted, compared to 42% of the Dissimilar objects. Unlike Experiment la, there was no hint that the effect of functional affordance was stronger for the Dissimilar than for the Similar objects. Thus, it is probable that the appearance of an interaction between similarity and functional affordance in the first experiment was due to confounded differences in the means by which functional affordance was manipulated. When functional affordance is eliminated in the same way in similar-looking and dissimilar-looking test objects, it seems to have an equal impact on the child’s decision-making about category membership-although global similarity makes a contribution as an independent factor. Adults’ Responses to the Test Objects. Adults relied almost purely on inferences of functional affordance in naming the test objects. For 5 of the 8 college students, every one of their 8 responses was in accord with the functional affordances arranged in the materials and unprincipled similarity relations had no effect whatsoever. The three remaining participants each gave more functionally retevant than functionally irrelevant responses, but together they accepted four dysfunctional objects-three Similar and one Dissimilar, and one subject rejected a Dissimilar functional item, Although there is the slightest of hints here that unprincipled similarity may be having an occasional effect, the preponderance of the data (59 of 64 responses) suggests that functional affordance is absolutely fundamental to the ways that adults classify these artifacts. This conclusion is also consistent with most of the existing literature (Keil, 1989; Miller & Johnson-Laird, 1976; Rips, 1989), and strengthens the motivation for our studies of children.

EX~ERI~E~

2

The findings reported so far were generated within two small-scale class projects conducted by groups of undergraduate students. Caution dictates that the critical results be replicated on a larger sample and under more carefully controhed conditions before they are accepted as fully convincing, Accordingly, Experiment 2 provided a tightly controfled and large-scale

Deborah

366

G. Kemler Nelson

replication of Experiment lb. In addition, Experiment 2 was designed to take a systematic look at whether use of the function principle in categorization tends to increase over the preschool period. Groups of 3-year-olds, 4-year-olds, and kindergartners (5 years or older) were tested in the same procedure with the same materials. Many of the 3-year-olds were younger than any of the children who had participated in the previous studies. Method Participants. Three groups of 15 children contributed data to the study. None of the children had participated in the earlier experiments. The 3-yearolds (M = 3;7, r = 3;3-4;0) and 4-year-olds (M = 4;7, r = 4;2-5;3) were enrolled in suburban preschools, and the kindergartners (M = $9, r = 5;3-6;5) attended after-school programs at the same locations. Twelve other children participated in the study, but their data were eliminated because of (a) a response bias to accept all test objects (3 subjects), (b) a response bias to reject all test objects (5 subjects), (c) a procedural irregularity (2 subjects), or (d) a failure to cooperate in completing the session (2 subjects). Of these children, five were in the 3-year-old group, four were in the 4-year-old group, and three were kindergartners. Within each age group, 8 participants were assigned to the Painter condition and 8 to the Instrument condition. Materials.

The stennet

materials used in Experiment

lb were used

again in this study. Procedure. The procedure was essentially the same as that used in the earlier experiments However, instead of allowing the children to interact again with the original object when it appeared on the fifth trial of the test phase, this second interaction was delayed until after the experimenter had completed an entire cycle through the test objects. Then, the children were allowed to try out the original stennet again, followed by the second phase of the test, when they were reminded of their earlier categorizations of the test objects and asked to justify them. As before, on the few occasions (22 of a possible 384) that the children changed their categorization responses on the second run, these final responses were the ones that were scored. Results and Discussion Categorizutions. Without exception, the children accepted the original object as a “stennet” during the test phase. The children’s categorizations of the test objects are shown in Figure 5. For each age group, separately for the Painter and Instrument conditions,

n

+FUNCTlON 81 -FUNCTION

Pslnter Group

Instrument Group

Slmllar Ob]ects

a

Painter Group

Instrument Group

Dbslmllar

Objects

1.0

% E

0.6

H m

0.6

z 0.4 S 7 E 2 a

H

+FUNCTlON - FUNCTION

H

+FUNCTlON - FUNCTION

q 0.2

00 Painter Group

Instrument Group

Similar Objects

s

1.0

: h

0.8

g 0

0.6

Painter Group

Instrument Group

Dlsslmllar Objects

ii .g 5 g

0.4

q 0.2

e n. 0.0 Painter Group

Instrument Group

Similar Objects

Palnter Group

Instrument Group

Dlsslmllar ObJects

Figure 5. Generalization patterns in Experiment 2 produced by 3-, 4-, and 5 to 6-year olds. For each exposure group, the proportion of acceptances of test objects as stennets, according to whether they were globally similar or dissimilar from the standard, and according to whether they could be inferred to function as the standard had. The +Function test objects for one group served as the -Function test objects for the other group. 367

368

Deborah C. Kemler Nelson

the figure shows the proportion of objects of each type that were accepted as stennets. Recall that, as previously, the same objects that served as +Function responses for children in the Painter group served as -Function responses for children in the Instrument group. Accordingly, an impressive and replicated feature of the findings is that in every age group, +Function categorizations as stennets considerably outnumbered -Function categorizations. Overall, 3 year olds accepted 72% of the test objects that afforded the function of the original, but only 23% of those that did not; analogously, 4-year-olds accepted 70% of the functional objects and only 36% that were not functional; and kindergartners accepted 92% of the functional objects, but only 26% of the dysfunctional ones. Of the 16 children in each age group, fifteen 3-year-olds, eleven 4-year-olds, and all sixteen kindergartners accepted more +Function than -Function objects. Overall, only three subjects showed a difference in the opposite direction. Unprincipled similarity relations also seemed to make some difference to the children. Collapsing over functional affordance, the 3-year-olds categorized as a stennet 58% of the Similar test objects as opposed to 38% of the Dissimilar objects. For the 4-year-olds, the comparable figures are 73% and 33%; and for the kindergartners, 67% and 52%. Ten of sixteen 3-yearolds, eleven of sixteen 4-year-olds, and nine of sixteen kindergartners accepted more similar than dissimilar objects. Differences in the opposite direction occurred in three cases among the 3-year-olds, 0 cases among the 4-year-olds, and two cases among the kindergartners. At each age level, there appears to be no interaction between functional affordance and similarity, but simple effects of both. These impressions were tested in an analysis of variance in which Age (3,4, or 5 years), Condition (stennet as Painter or Instrument), Functional Affordance of the test object (+Function or -Function), and Similarity relation to the original object (Similar or Dissimilar) were entered as variables. The only reliable effects to emerge were a main effect of Functional Affordance, F(1,42) = 120.20, p < .OOl, a main effect of Similarity, F(1, 42) = 29.15, p < .OOl, and an interaction between Age and Functional Affordance, F(2,42) = 4.02,~ < .05. The interaction between Age and Similarity was also marginally significant, F(2,42) = 2.64,~ < .lO. Post-hoc analyses were carried out using Tukey’s tests with Cicchetti’s (1972) suggested correction. The mean number of acceptances of +Function and -Function objects for each age group (with a possible maximum of 4.0) is shown in Table 3. The critical difference between means (p < .05) is .892. Thus, in every age group, there were more acceptances of functional than dysfunctional objects. In no pair-wise comparison was there a reliable age difference at the conventional level of statistical significance when functional affordance is constant. The overall interaction between Age and Functional Affordance seems to be reflecting an only marginally greater

Principle-Based

369

Categorization

Table 3. Mean Number of Acceptances of Test Objects as Stennets (Maximum = 4) According to Age of Child and Functional Affordance of Object Functional Affordance Aee 3 4 5

+ Function

- Function

2.87 2.82 3.68

.94 1.44 1.06

tendency for the kindergartners to classify +Function objects as members of the category than either the 3- or 4-year-olds. The mean number of acceptances of Similar and Dissimilar objects for each age group (with a potential maximum of 4.0 again) is shown in Table 4. The critical difference (p < .05) between means is .936. Apparently, only the 4-year-olds showed a statistically significant difference in their responses to Similar and Dissimilar objects in a pair-wise comparison. However, in light of the facts that the overall interaction between Age and Similarity is only marginally reliable and the main effect of Similarity is highly reliable, it is doubtful that much significance should be attached to this possible age qualification. The more secure conclusion is that in their categorizations the children as a whole took some account of unprincipled similarity-in addition, of course, to functional affordance. The larger data set collected in this experiment permitted a closer inspection of the responses elicited by each of the specific objects. The proportion of positive stennet categorizations for each one of the test objects is shown in Table 5, partitioned by age and condition. Each of the separate age group proportions is based on 8 responses. There is a consistent pattern to be discerned. A greater tendency to categorize a particular object as a stennet if it maintained the functional affordance than if it eliminated the function held for 7 of the 8 objects among the 3-year-olds, 6 of the 8 objects among the 4-year-olds, and all 8 objects among the kindergartners. All three exceptions (two of which were ties) involved those two objects that could

Table 4. Mean Number of Acceptances of Test Objects as Stennets (Maximum = 4) According to Age of Child and Similarity of Object to the Original Stennet. Similarity Relation Aee

Similar

3 4 5

2.30 2.94 2.68

Dissimilar 1.50 1.32 2.06

Deborah C. Kemler Nelson

370

Table 5. Proportion of Acceptances of Each Test Object as a Stennet According to Age of the Child and Functional Affordance of the Object. Similar and Dissimilar Objects are Designated as S and D Respectively. Instruments and Painters are Designated as I and P Respectively. Objects Subscripted With a 1 Were Made Dysfunctional by Repositioning a Critical Part; Those Subscripted With a 2 Were Made Dysfunctional by Replacing a Critical Part. 3-Year-Olds

4-Year-Olds

5-Year-Olds

All

+F

-F

+F

.75 .88

.38 .38

.88 1.00

.88 .75

1.00 1.00

.75 .50

.88 .96

.67 .54

sp2

1.00 .75

.25 .25

.75 1.00

.25 .38

1.00 1.00

.OO .12

.92 .92

.17 .25

DI, I’P,

.38 .62

.38 .12

.50 .50

.50 .12

.88 .88

.62 .12

.59 .67

.50 .12

“I,

.75 .62

.12 .OO

SO .50

.OO .Oo

.88 .75

.OO .OO

.71 .62

.04 .c0

Object

SIl -1

SI,

DP2

-F

+F

-F

+F

-F

function as instruments but not as painters, and in which the functional affordance of painting was blocked by placing the brushes in a dysfunctional location (such that the brushes could not be dipped in the row of paints or used to apply paint to the paper). The questionable role of functional affordance in categorizing these particular objects was present to a lesser degree even in the oldest group of children: For the two Instruments whose function was eliminated by embedding brushes, a majority of kindergartners still classified them as stennets. It seems plausible that the children were less certain than we (and the adult subjects in Experiment lb) that embedding the brushes eliminated the painting function of these objects. Analyses of the children’s verbalizations, reported below, will shed some light on this possibility. These two exceptions aside, there is a remarkably uniform pattern in the breakdown in Table 5, supporting the major conclusion that inferences about functional affordance play a central role in young children’s categorization of artifacts. Explanations. A very large majority of children’s explanations for their categorizations, both positive and negative, included mentions of particular object parts and/or mentions of the functional affordances of the objects. An initial inspection of these responses showed that across participants, mention of brushes, brush substitutes (when the brushes were replaced), strings, and string substitutes (when the strings were replaced) accounted for most of the parts that were mentioned. These, of course, are just the parts of the objects that are critically and differentially relevant to the two functions of the original object. (Those who participated did also

Principle-Based

Categorization

371

mention handles and string tuners, which also may be deemed relevant to the objects’ functions, but they mentioned these parts at a much lesser rate.) Of the functions mentioned, virtually all of them referred to either the musical instrument function or the painting function. Because our major question pertained to whether participants’ explanations, like their categorizations, indicated special attention to object function, we coded all their responses for whether or not they included: (a) reference to any of the critical functional parts; i.e., brushes, brush substitutes, strings, or string substitutes, and (b) reference to the instrumental or painting function itself. For the former, we accepted any response that alluded to the part; i.e., a comment on its presence or its absence, or a mention that the part was in an unusual location or missing at a usual location. For the latter, we accepted attempts to perform or mime a function in response to the probe to explain the categorization, as well as actual verbal references to the object’s capacity to function or not function in a particular way. For each object, then, a child was scored as having referred or not to (a) brushes or brush substitutes, (b) strings or string substitutes, (c) the painting function, and (d) the musical instrument function. These four categories were not mutually exclusive, since a single subject could mention any or all of these with regard to any given test object. In a reliability check performed on the data of 10 participants, there proved to be no disagreements between the primary coder and a secondary coder with regard to scoring these four categories. To summarize the scores, we computed separately for each exposure condition (Painter or Instrument) within each age group, the number of mentions of each of these kinds of parts and each of these kinds of functions. These numbers are shown in Table 6 and are designated as pertaining to “relevant” or “irrelevant” attributes according to their relation to the participants’ exposure condition. Thus, mentions of strings or string substitutes are deemed “relevant” for the Instrument condition and “irrelevant” for the Painting condition, and vice versa for mentions of brushes or brush substitutes. Analogously, references to the musical instrument function are deemed “relevant” for the Instrument condition and “irrelevant” for the Painting condition, and vice versa for references to the painting function. The pattern in Table 6 is very clear. For all groups, mentions of relevant parts and functions far outnumbered mentions of irrelevant parts and functions, respectively. With regard to parts, there is some indication that the degree of discrepancy between the relevant and irrelevant frequencies was somewhat greater, the older the subject, but this age difference did not seem to hold for references to function. Still, even the 3-year-olds showed almost twice as many mentions of relevant than irrelevant parts. Further inspection of the data revealed that, unlike the older children, 3-year-olds differentially mentioned relevant and irrelevant parts only for

372

Deborah G. Kemler Nelson

Table 6. The Number of References to Relevant and Irrelevant Parts and to Relevant and Irrelevant Functions for Each Condition for Each Age Group. The Text Provides More Detail About Scoring Criteria. Parts

Functions

Relevant

Irrelevant

Relevant

Irrelevant

22 25 47

14 10 24

35 14 49

2 7 9

35 46 81

23 9 32

32 14 46

2 3 5

45 52 97

9 3 12

31 9 40

1 8 9

3-year-olds Instrument Painting All

4-year-olds Instrument Painting All

5year-olds Instrument Painting All

the Similar objects (32 relevant vs. 10 irrelevant)-not the Dissimilar ones (15 relevant vs. 14 irrelevant). The 4- and 5-year-old children’s greater mention of relevant parts held for both Similar objects (49 relevant vs. 16 irrelevant for the 4’s; 47 relevant vs. 8 irrelevant for the 5’s) and Dissimilar objects (32 relevant versus 16 irrelevant for the 4’s; 50 relevant versus 4 irrelevant for the 5’s). It is probably unwise to attach much significance to the exception of the 3-year-olds’ nonselective references to parts of Dissimilar objects-especially since, with respect to mentions of function, these same children, just like their older counterparts, differentially mentioned the relevant and irrelevant with regard to both Similar and Dissimilar objects. For 3-, 4-, and 5-year-olds respectively, the numbers of mentions of relevant versus irrelevant functions with regard to Similar objects were: 27 versus 2,18 versus 2, and 20 versus 4; and with regard to Dissimilar objects were: 22 versus 7,28 versus 3, and 20 versus 5. Overall, there appears to be more frequent reference to the instrument function than to the painting function, but this is due to the fact that the former function could be easily demonstrated by the child on the test objects (by strumming with a finger rather than a guitar pick), but-with paper and paints absent in the test phase-the latter function could not be. In fact, when only verbal mentions of function are counted as references to function, there is a total of 41 mentions of relevant function by the Instrument groups, about equal to the 37 mentions of function by the Painter groups. With the same verbal constraint, the number of irrelevant mentions of the instrument function drops to only 5 overall, so a strong tendency for

Principle-Based Categorization

373

more references to relevant than irrelevant function is retained under the different scoring system. Children’s explanations of their categorization, then, like the categorizations themselves, appear to be centrally based in the functional affordances that the children infer from the structure of the objects. The children’s explanations also provide some insight regarding what had seemed to be two exceptional cases in the categorization data. Recall that the tendency to classify by experimenter-designated function was weaker (or even nonexistent in some age groups) for two specific objects-the two Instruments for which we judged the painting function to be eliminated by embedding the brushes in inaccessible positions within the object. For the Similar token of this type, one older child in the Painter group who accepted the item as a “stennet” offered, “You could dip this in some paint if you were very careful.” A kindergartner in the Painter group accepted the object as a stennet, and suggested, “I don’t know how you’re supposed to paint with these (brushes). Maybe you’re supposed to pull them out.” A second kindergartner said the object was a stennet and explained, “If you took some (brushes) off, then you could paint with it.” The other (Dissimilar) token elicited some related explanations of positive categorizations. A kindergartner opined, “You’d have to use these (brushes) one at a time.” Thus, it appears that for these particular objects, we were less successful than we had hoped in disabling the function of the objects in a way that was easily inferrable and widely agreed upon given the objects’ structure. These cases do not seem to be exceptions to our overall finding that children pay close attention to functional affordances in categorization. Instead, these are cases in which children base their categorizations on inferences about function that are different from ours. Indeed, these unanticipated cases actually serve to bolster our major conclusion in one way: The children’s verbalizations seem to indicate that they are engaged in just the kind of reasoning and “theory-testing” (Murphy & Medin, 1985) that is said to be characteristic of principle-based categorization. Some additional comments of the children are also suggestive of a true conceptual principle at work in their categorization. On several occasions, children noted that Dissimilar, -l-Function objects were stennets, but that they constituted “a different kind of stennet” or a “weird stennet.” Indeed, sometimes in accepting these objects as stennets, the children also modified the category name with reference to an irrelevant, but perceptually salient property of the test object, such as “a wooden stennet” or “a star stennet.” On other occasions, with regard to Similar, -Function objects, children explicitly noted the discrepancy between reality and appearance. Following a rejection of an object, one child said: “It looks just like a stennet, but it’s really not,” and, after rejection, another commented: “You think it’s a stennet, but it doesn’t make music.”

Deborah

374

C. Kemler Nelson

All of these verbalizations confirm that perceptual similarity relations alone cannot explain the findings. Instead, the children are actively engaged in a perceptual search process that is under the direction of a conceptual principle (Murphy & Medin, 1985). As a means for determining category decisions, the data that perception yields are evaluated by the children in relation to a principle of common function.

GENERAL

DISCUSSION

We have reported three studies that offer compelling evidence that spontaneous inferences about function mediate category learning about novel artifacts by young children. These demonstrations that preschoolers’ categorization and naming are consistently influenced by functional information come as some surprise in light of previous research efforts that have found little evidence that it plays such a role. In fact, on one important count, the current test for the importance children attribute to functional information is far more demanding than the previous tests they have “failed.” The positive demonstration here is particularly remarkable because it emerges in a context in which children-on their own initiative-must infer whether or not a test object has a certain functional affordance. In much previous work where preschoolers seem to have ignored functional information in categorization, the function was directly demonstrated to them at the time of generalization. Why, then, do the children in the current study, relative to their agemates in previous research, appear to be so much more systematically principled in their categorization-constructing a new category of artifacts around a common functional core? We believe that the most crucial aspects of our test were the design of objects whose functional affordances (and disabilities) were related in a nonarbitrary manner to their perceptible attributes, and where the causal links between structure and function were simple for the children to understand. This is a context in which children have some principled background knowledge (about physical causality) that they can readily bring to bear. Apparently, even brief interaction with a single instance of a new artifact kind is sufficient motivation for the children to activate that knowledge and to reason from it. Is it also necessary that the object be labeled? A number of recent investigations suggest that children’s categorization when language labels are used is systematically different from categorization elicited in a nonlinguistic context (Baldwin, 1992; Imai et al., 1994; Jones, Smith, & Landau, 1991; Landau, Smith, & Jones, 1988; Markman & Hutchinson, 1984; Waxman, 1990). Because we did not systematically manipulate the labeling variable, it is impossible to tell whether supplying names for our new artifacts contributed to the use of the function principle in children’s categori-

Principle-Based Categorization

375

zation. It would not be surprising if the labels served at least a facilitative role of this sort (see Ward, 1990, for a related suggestion). Yet, Smith, Jones, and Landau (1995) have reported, to the contrary, that functional information influences young children’s categorization only outside of a naming context; that is, when children judge which objects are “like” the standard but not when they judge which objects get the same name. In line with the present results, these investigators show that functional affordances sometimes determine categorization in 3-year-olds, but, in contrast with the present results, they find that naming blocks this function-based mode of generalization. There are many differences between their procedures and ours that could account for the discrepant observations when objects are named. First, the novel names were introduced in a more natural way in our procedure: The adult used the name not only to label the standard object in isolation, but also to refer to the object a number of times in some variety of linguistic contexts as the child interacted with it. Second, some of the functions that Smith et al. described to the children seem to be rather arbitrary (e.g., “a toy dog sits in it”; “it holds pens”) or generic (e.g., “you can look through it”)-not the kind of novel function that could be expected to distinguish a truly new and basic-level artifact category. Third, our impression is that the objects we constructed tended to have more of the “look” of real, but truly novel artifacts that a child could encounter outside the laboratory. Fourth, because of the substantially fewer number of responses required in the experimental session, our participants may have responded to our queries in a more reflective and less hurried way. Any or all of these factors could explain why we might have been more successful in engaging children’s everyday strategies for interpreting names for basic-level artifacts. Factors of this sort also could have contributed to other reported failures to find evidence for preschoolers’ use of the functional principle in naming (Landau, Smith, & Jones, 1995; Merriman et al., 1993). In addition, a number of studies of the role of function in children’s categorization have used a rather simple methodology in which attention to function is pitted directly against attention to surface appearance (or shape identity specifically) as a basis for generalization (e.g., Landau et a1.,1995; Merriman et al., 1993). At the time of test, children are confronted with dissimilar-looking, but like-functioning objects and similar-looking objects that do not afford the critical function. So, what their responses can reveal is only which of the two factors in conflict-unprincipled similarity or functional affordance-is deemed more relevant to categorization by the child. In contrast to the present methodology, what such a procedure can obscure is whether the child invokes the functional information at all. Although the news in our data is children’s attention to functional affordance, the data by no means deny a role for unprincipled similarity

376

Deborah C. Kemler Nelson

also. A tendency to include more similar than dissimilar objects as members of the category was observed-functional affordance aside. This similarity effect was most unambiguous in Experiment 2, where use of the stennet materials allowed us to systematically unconfound the nature of the functional manipulation from the degree of unprincipled similarity, and where sufficient data were collected to allow for formal statistical tests. The appropriate inference from our data is not that functional affordance plays a more dominant role in children’s categorization of artifacts than does unprincipled similarity. The previous literature provides many counterindications to such a sweeping generalization. What we have been arguing is something quite different-namely, that there are circumstances under which even young children will invoke the function principle in categorizing basic-level artifacts. Perhaps most importantly, these will be circumstances when the structural properties that determine basic-level functions are easily perceived and when the relevant physical-causal principles that relate structure to function are well understood. Of course, this implies that in many cases, children will not invoke the function principle, and, in those cases, we and many other investigators believe that children will-quite reasonably-fall back on unprincipled similarity (e.g., Keit, 1989; Murphy & Medin, 1985) or shape similarity specifically (e.g., Imai et al., 1994). Moreover, given that high similarity is often predictive of functional equivalence in real-world categories, we should not be surprised to see-as we did in the current data-that even when children are inclined to attend to functional affordance, they may be inclined to take account of overall similarity also. Still, it is important to note that these qualifications on children’s use of the function principle do not detract from the importance of the current demonstration. If children sometimes do give weight to function when they have the necessary knowledge and perceptual support to invoke it, then their tendency not to take function into account under other circumstances may stem from their lack of domain-specific knowledge (e.g., of the intended function of certain objects, of their functionally relevant structural properties, or of the physical principles that mediate between the structural properties and the intended function) rather than to a general tendency to ignore functional affordance or to regard it as unimportant or irrelevant to categorization. So, just as we retreat from a general claim that young children always attend to functional affordances in naming artifacts, so must we reject a general claim that young children always base their linguistic categorizations of artifacts on a single nonfunctional principle-such as shape similarity. The sets of test objects that we employed provided ample opportunity to observe that similarity in shape is neither necessary nor sufficient for children to give two objects the same name. As to its not being necessary, we

Principle-Based

Categorization

377

observed both with the bindle materials and the stennet materials, multiple cases in which a test object that was very different in shape from the original (which was clearly the case with virtually every one of our Dissimilar objects-arguably, every one of them) was nevertheless given the same name as long as it functioned like the original. As to shape similarity’s not being sufficient, the stennet materials provides some compelling demonstrations. For example, when the paintbrushes were simply replaced by comparably configured strings of small beads, such that the overall shape of the object was highly similar, only a small minority of subjects in the Painter group thought it was a stennet. Analogously, when the fan of strings was simply replaced by a triangular piece of gauze, again maintaining high similarity of shape, only a minority of subjects in the Instrument group called it a stennet. To be sure, identity of shape did not obtain in any of these cases (in contrast to the studies of Landau et al., 1995), but high similarity in shape and high overall similarity surely did. It is also noteworthy that 3- to 6 year old children spontaneously invoke the function principle even when they are learning about novel artifact categories which are characterized by sometimes quite novel functions. Some investigators have suggested that children first begin to give up a naming strategy based wholly on unprincipled similarity or similarity in shape in the context of highly familiar categories. For example, Gentner and her colleagues (Genter & Rattermann, 1991; Imai et al., 1994) argued that experience with members of a category should lead to greater understanding of the causal and functional relationships in which the objects participate, and, accordingly, promote a shift away from physical similarity in extending the name of the category. Such a view has much in common with the current proposal. Both highlight the impact of domain-specific factors on the ways that children extend categories and both imply that, for artifacts, the function principle will come increasingly into play where more is known. However, it is notable that young children in our studies required only a few short minutes of experience with a single instance of a novel category before they attended to function. Hence, what appears to be important is not familiarity with the category per se, but adequate information about the intended, conventional function of at least one category instance (which, as we have seen, may be conveyed in a few short minutes) and command of the general, physical principles that account for relevant structure/function relationships. A related feature of our results is the failure to find any compelling developmental differences in the use of the function principle. This may be regarded as somewhat surprising for two reasons. One reason is that our stennet functions (e.g., painting parallel lines, strumming a tune) bear some similarity to functions that children become more and more familiar with over the period between three and six years of age. The other is that general

378

Deborah

C. Kemler Nelson

experience with how members of artifact categories are named surely increases over that period. Relative to 3-year-olds, kindergartners have had more occasions and a more varied data base for learning that two objects that share the same intended function get the same name even when they are dissimilar in shape or dissimilar overall, and also for learning that two similar-looking objects get different names when they are intended to function differently. However, neither kind of experience seems to have played much of a role in our situation. This is likely because the functions that we created are quite compelling even to 3-year-olds, and because even they have had exposure to a sufficient data base of named artifact categories for them to draw the inferences about the importance of function. In addition, the structure-to-function inferences that we successfully tapped into here were deliberately very simple ones. Were the features of the stimulus materials different-either more subtle functions or more complicated structureto-function relations, developmental differences in the preschool period might well emerge. In addition, it is possible that below the age of three, with less broad exposure to how artifact categories are named-that is, according to common function-children might be generally less likely to spontaneously invoke the function principle even when they have the necessary domain-specific knowledge to do so. Ongoing studies in our laboratory address this possibility.

REFERENCES Baldwin,

D.A. (1992). Clarifying the role of shape in children’s taxonomic assumption.Journal

of Experimental

Child Psychology, 54,392-4X.

Brown, A.L. (1990). Domain-specific principles affect learning and transfer in children. Cognitive Science, 14,107-134. Carey, S. (1985a). Are children fundamentally different thinkers and learners than adults? In SF Chipman, J.W. Segal, & R. Glaser (Eds.), Thinking and learning skills, Vol. 2. Hillsdale, NJ: Erlbaum. Carey, S. (1985b). Conceptual change in childhood. Cambridge, MA: MIT Press. Cicchetti, D.V. (1972). Extension of multiple-range tests to interaction tables in the analysis of variance: A rapid approximate solution. Psychological Bulletin, 77,405--108. Clark, E.V. (1973). What’s in a word? On the child’s acquisition of semantics in his first language. In T.E. Moore (Ed.), Cognitive development and the acquisition of language (pp. 65-110). New York: Academic. Corrigan, R., & Schommer, M. (1984). Form versus function revisited: The role of social input and memory factors Child Development, 55,1721-1726. Gelman, R. (1982). Recent trends in cognitive development. In C.J. Schierer & A.M. Rogers (Eds), The G. Stanley Hall lecture series, (Vol. 3, pp. 145-175). Washington, DC: American Psychological Association. Gentner, D. (1978). A study of early word meaning using artificial objects: What looks like a jiggy but acts like a zimbo? Papers and Reports on Child Language Development, 15, l-6.

379

Principle-Based Categorization

Gentner, D., & Rattermann, M.J. (1991). Language and the career of similarity. In S.A. Gehnan & XP.Byrnes (Ed?.), Perspectives on thought and language: Inferrelations in development. London: Cambridge University Press. Gentner, D., & Toupin, C. (1986). Systematicity and surface similarity in the development of analogy. Cognitive Science, 10,277-300. Imai, M., Gentner, D., & Uchida, N. (1994). Children’s theories of word meaning: ‘Ihe role of shape similarity in early acquisition. Cognitive Development, 9,45-75. Jones, S.S., Smith, L.B., & Landau, B. (1991). Object properties and knowledge in early lexical learning. Child Development, 62,499-516. Keil, EC. (1989). Concepts, kinds, and conceptual development. Cambridge, MA: MIT Press Kolstad, VT., & Baillargeon, R. (1990). Context-dependent perceptual and functional categorization in lO.S-month-oZd infants. Unpublished manuscript. Landau, B., Smith, L.B., & Jones, S.S. (1988). The importance of shape in early lexical learning. Cognitive

Development,

3,299-321.

Landau, B., Smith, L., & Jones, S. (1995). Object shape, object function, and object name. Manuscript submitted for publication. Markman, EM., & Hutchinson, J.E. (1984). Children’s sensitivity to constraints on word meaning: Taxonomic versus thematic relations. Cognitive Psychology, i6,1-27. Medin, D.L. (1989). Concepts and conceptual structure. American Psychologist, 44,1469-1481. Merriman, WE, Scott, PD., & Marazita, J. (1993). An appearance-function shift in children’s object naming. Journal of Child Language, 20,101-118. Mervis, C.B. (1987). Child-basic object categories and early lexica1 development. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization (pp. 201-233). Cambridge: Cambridge University Press. Mervis, C.B., & Greco, C. (1984). Parts and early conceptual development: Comment on Tversky and Hemenway. Journal of Experi~nental Psychology: General, 113,194-197. Miller, G.A., & Johnson-Laud, PN. (1976). Language andpercept~on. Cambridge, MA: Harvard University Press. Murphy, G.L., & Medin, D.L. (1985). Role of theories in conceptual coherence. Psychological Review, 92,289-3X.

Nelson, K. (1973). Some evidence for the cognitive primacy of categorization and its functional basis. Merrill-Palmer Quarterly, 19,21-39. Nelson, K. (1974). Concept, word, and sentence: Interrelations in acquisition and development. Psychological

Review, 81,267-285.

Nelson, K. (1978). Semantic development and the development of semantic memory. In K.E. Nelson (Ed.), Children’s language. New York: Gardner. Nelson, K. (1979). Expioration in the development of a functional semantic system. In W. Collins (Ed.), Children’s language and communication. Minnesota Symposium on Child Psychology, Vol. 12. HilIsdaIe, NJ: Erlbaum. Nelson, K. (1983). Concepts, words, and experiments: Comment on “When is a cup not a cup?” by Anderson and Prawat. Merely-Farmer Quarterly, 29,387-394. Prawat, R., & Wildfong, S. (1980). The influence of functional context on children’s labeling responses. Child Development, 51,1057-1061. Rips, L.J. (1989). Similarity, typicality, and categorization. In S. Vosniadou & A, Ortony (Eds), Similarity and analogical reasoning (pp. 21-59). Cambridge: Cambridge University Press. Rosch, IS., Mervis, C.B., Gray, W., Johnson, D., & Boyes-Braem, I? (1976). Basic objects in natural categories. Cognitive Psychology, g382-439. Smith, L.B., & Heise, D. (1992). Perceptual similarity and conceptual structure. In B. Burns (Ed.), Percepts, concepts, and categories: The representation and processing of information (pp. 233-272). Amsterdam: Elsevier. Smith, L.B., Jones, S.S., & Landau, B. (1995). Naming in young children: An encapsulated attentional mechanism.

Deborah

380

C. Kemler Nelson

Tomikawa, S.A., & Dodd, D.H. (1980). Early word meanings: Perceptually or functionally based? Child Development, 51,1103-1109. Tversky, B. (1989). Parts, partonomies and taxonomies. Developmental Psychology, 25,983-995. Tversky, B., & Hemenway, K. (1984). Objects, parts, and categories. Journal of Experimental Psychology:

General, 113,169-193.

Ward, T.B. (1990). The role of labels in directing children’s attention. In J.T. Enns (Ed.), The development ofattention: Research and theory (pp. 321-342). Amsterdam: Elsevier. Waxman, S.R. (1990). Linguistic biases and the establishment of conceptual hierarchies: Evidence from preschool children. Cognitive Development, 5,123-150.