Conceptually coherent categories support label-based inductive generalization in preschoolers

Journal of Experimental Child Psychology 123 (2014) 1–14

Contents lists available at ScienceDirect

Journal of Experimental Child Psychology journal homepage: www.elsevier.com/locate/jecp

Conceptually coherent categories support label-based inductive generalization in preschoolers Amy E. Booth Roxelyn and Richard Pepper Department of Communication Sciences and Disorders, Northwestern University, Evanston, IL 60208, USA

a r t i c l e

i n f o

Article history: Received 4 October 2013 Revised 12 January 2014 Available online 12 March 2014 Keywords: Inductive generalization Inductive inference Preschoolers Conceptual knowledge Causal information Causal knowledge

a b s t r a c t Why do words support inductive generalization in preschoolers? The current study provides evidence that they do so, at least in part, by working with conceptual knowledge to establish kind membership. A sample of 30 4-year-olds learned new labels for novel items, sometimes along with additional non-obvious information, and were then asked to generalize a novel object property to a target item based on either visual similarity or shared label. Children were more likely to generalize properties based on shared labels (over perceptual similarity) if they initially learned causally coherent properties of items referenced by those labels than if they initially learned non-causal properties of those items or learned no properties at all. This finding suggests that novel words best support inductive inference when they are known by children to reference conceptually coherent categories. Therefore, conceptual information permeates the process of inductive inference in young children. Results are discussed with respect to their implications for the ‘‘word-as-feature’’ and ‘‘knowledgebased’’ accounts of early inductive inference. Ó 2014 Elsevier Inc. All rights reserved.

Introduction Inductive inference is a key cognitive process by which semantic information can be efficiently generalized beyond the context in which it was initially acquired. It is particularly useful to the extent E-mail address: [email protected] http://dx.doi.org/10.1016/j.jecp.2014.01.007 0022-0965/Ó 2014 Elsevier Inc. All rights reserved.

2

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

that it unfolds in accordance with the organization of the learner’s conceptual backdrop. That is, generalizations that are consistent with category boundaries are more likely to be accurate and, therefore, to facilitate understanding, appropriate expectations, and advantageous courses of action. It is clear that adults weigh categories heavily in their inductive generalizations (Deak & Bauer, 1996; Fisher, 2010; Sloutsky & Fisher, 2004). Considerable controversy, however, surrounds the developmental origins of this conceptual orientation. Much of this controversy has focused specifically on the role of words in supporting early inductive generalization. In the current study, we contribute to the ongoing discussion by seeking new evidence regarding the conditions under which shared labels take precedence over visual perceptual similarity in guiding children’s generalization of newly learned information. Words have figured prominently in this debate since its inception. For example, in Gelman and Markman’s (1986) groundbreaking work on children’s inductive inference, marking stimuli with shared labels served as the primary manipulation. In their study, 4-year-olds were first taught contrasting non-obvious properties of two pictured animals (e.g., a brontosaurus and a rhinoceros). They were then asked to judge which of these two non-obvious properties could be generalized to a third animal that was perceptually quite similar to one of the original standards but was actually drawn from the same category as the other standard (e.g., a triceratops that looked very much like the rhinoceros but was actually a dinosaur like the brontosaurus). The key finding was that children’s patterns of generalization were guided more heavily by labels than by visual similarity (e.g., children were more likely to generalize from the brontosaurus than from the rhinoceros to the triceratops). Based on this finding, Gelman and Markman advocated for a ‘‘knowledge-based’’ account of early inductive inference. They concluded that, by acting as cues to kind membership, labels provide children with a reliable conceptual framework for extending newly learned information. This study opened the floodgates to opinion and research articles aimed at supporting or disconfirming this conclusion. Naysayers championed more traditional conceptions of the child as concrete and perceptually oriented (e.g., Jones & Smith, 1993), whereas defenders extended Gelman and Markman’s initial finding to ever younger ages (e.g., Gelman & Coley, 1990; Gelman & Markman, 1987; Welder & Graham, 2001). Perhaps the strongest challenge to the knowledge-based account of early inductive inference originated with the ‘‘word-as-feature’’ alternative (Sloutsky & Lo, 1999; Sloutsky, Lo, & Fisher, 2001). Proponents of this account argue that shared names guide inductive inference in young children because names are salient perceptual features in and of themselves and, thus, will increase the overall perceptual similarity between identically named items. It is this perceptual similarity that is thought to be the primary determinant of inductive generalization patterns in preschoolers. Labels are thought to be particularly powerful in children’s similarity assessments because they are perceived through the auditory modality, which some recent research has identified as dominant in children’s perceptual processing (Robinson & Sloutsky, 2004; Sloutsky & Napolitano, 2003; but see Noles & Gelman, 2012b). The knowledge-based and word-as-feature accounts make a number of distinct predictions about the conditions under which labels should and should not guide inductive generalization (and other related cognitive processes) in preschoolers. The literature addressing these divergent predictions is already considerable but remains contentious (see, e.g., Gelman & Davidson, 2013; Gelman & Waxman, 2007; Graham, Booth, & Waxman, 2011; Hayes & Heit, 2004; Heit & Hayes, 2005; Noles & Gelman, 2012a; Sloutsky, 2008; Sloutsky & Fisher, 2005, 2012; Sloutsky, Kloos, & Fisher, 2007a; Waxman & Gelman, 2009; Wilburn & Feeney, 2008). In hopes of providing a new angle on the debate, here we focus on divergent theoretical predictions specific to circumstances under which children have no familiarity with either the items or the labels involved. These conditions can be particularly revealing because they most purely reflect children’s tendencies to weigh broad types of information (e.g., labels, visual appearance) independent of their knowledge of specific words and/or categories. According to the word-as-feature account, shared novel labels should be equivalent to shared familiar labels in their contribution to the similarity of items and, therefore, should be equally supportive of preschoolers’ inductive generalization. Consistent with these assertions, some studies have demonstrated an effect of novel words on preschoolers’ inductive generalization (e.g., Gelman & Davidson, 2013; Graham et al., 2011; Sloutsky & Fisher, 2004; Sloutsky et al., 2001). Of particular

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

3

relevance is a study conducted by Sloutsky and Fisher (2004) in which they found no difference in the effect of familiar and novel labels on inductive generalization in 4- and 5-year-olds. Although at first blush these results might appear to run counter to the knowledge-based account, a more careful consideration of this theoretical position explains how these data could be construed as entirely consistent. Proponents of the knowledge-based view argue that shared novel labels should guide early inductive generalization (over visual similarity) only to the extent that children find them compelling as indicators of meaningful kinds. It is well documented that count nouns (even when they are novel) act as strong cues to categories for young children (e.g., Booth & Waxman, 2002a; Waxman & Booth, 2000). Importantly, however, preschoolers are not thought to automatically accept novel labels as absolute markers of new kinds (Gelman & Waxman, 2007; Noles & Gelman, 2012a). Rather, theory and evidence suggest that children take into account the reliability and intentions of the speaker, as well as the coherence of perceptual and conceptual properties of the labeled items, in evaluating the plausibility of a new label (Chapman, Leonard, & Mervis, 1986; Davidson & Gelman, 1990; Jaswal, 2004; Jaswal & Markman, 2007; McCarrell & Callanan, 1995; Mervis & Mervis, 1988; Sabbagh & Baldwin, 2001; Sobel & Corriveau, 2010). Perhaps then the documented cases of novel labels guiding inductive generalization in preschoolers referenced above (i.e., Graham et al., 2011; Sloutsky & Fisher, 2004; Sloutsky et al., 2001) all involved circumstances in which those labels were sufficiently compelling as markers of meaningful kinds to convince children of their inductive potential. A kind-based mindset might have been encouraged in a number of ways in these studies, including reference to familiar pictured categories (e.g., cats), provision of a plausible level of coherence among the properties of labeled exemplars, and pretraining with highly familiar categories from which analogies to the novel test items might have been drawn. Consistent with this interpretation, novel labels are sometimes less effective than familiar labels in facilitating inductive generalization in young children. For example, Davidson and Gelman (1990) reported such a difference in 4-year-olds (see also Welder & Graham, 2001, for similar evidence from infants). In their Yes/No judgment task, children were sequentially tested on their generalization of newly learned hidden properties to novel animals and artifacts that differed from the target in name, visual appearance, neither, or both. Davidson and Gelman (1990) suggested that, because children expect members of a kind to share many properties in common, they might have been confused by the fact that the new labels were used to refer to items that looked very different from each other and that different labels were used to refer to items that looked very similar to each other. As a result, children might not have trusted that the novel labels provided by the experimenter were meaningful and, therefore, fell back on another good (although generally somewhat less reliable) cue to category membership, visual similarity, to guide their generalizations. This explanation was further consistent with results from a follow-up study in which one of the conflict test items (i.e., either the perceptually similar/different name or the perceptually dissimilar/same name) was eliminated for each subject (Davidson & Gelman, 1990). Here the evidence against the coherence of labeled categories was reduced and children relied on both familiar and novel labels equally. The importance of the conceptual status of novel labels as markers of kinds is further highlighted by recent work with both infants and preschoolers indicating that novel labels presented as count nouns, but not as adjectives, support inductive generalization (Graham et al., 2011; Keates & Graham, 2008). Because theoretical debate regarding the role of words in supporting early inductive inference has yet to be resolved (e.g., Badger & Shapiro, 2012; Deng & Sloutsky, 2012; Fisher, Matlen, & Godwin, 2011; Gelman & Davidson, 2013; Graham et al., 2011; Noles & Gelman, 2012a), here we sought to provide a direct test of the predictions of the knowledge-based account. To this end, we implemented a procedure in which children were taught different types of information about novel to-begeneralized-from items prior to completing a classic triad inductive generalization task (Davidson & Gelman, 1990). In our two key experimental conditions, children were taught information that is more or less relevant to establishing novel names as indicators of conceptually coherent kinds. Specifically, children were taught either causal or non-causal properties of labeled items. We reasoned that if words guide inductive generalization because they provide additional information over which children can calculate perceptual similarity (i.e., the word-as-feature account is correct), novel labels should have equally strong effects regardless of what type of additional information is provided during this pretesting period. If, however, words guide inductive generalization because they signal the

4

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

existence of meaningful kinds (i.e., the knowledge-based account is correct), different patterns of generalization should be evident across these conditions. Because causal information is fundamental to conceptual structure and coherence (e.g., Gopnik, 2003; Wellman & Gelman, 1992), causal descriptions should provide stronger evidence that novel labels refer to meaningful kinds than do non-causal descriptions. As a result, causal descriptions should lead to stronger name-based inductive generalization than do non-causal descriptions. To provide a broader interpretive context, performance in these two experimental conditions was assessed relative to two control conditions in which children were provided with names alone or with no information at all (beyond what is perceptually obvious in the pictured stimuli themselves) on which to base their inductive generalizations. Method Participants Data from 30 4-year-olds (M = 53.6 months, range = 48.2–59.9), half of whom were girls, were included in the final analysis. Peabody Picture Vocabulary Test scores averaged 114 (SD = 16). An additional 10 children participated but were excluded because of non-compliance (n = 2), experimenter error (n = 5), parental report of developmental disability (n = 2), or less than 25% exposure to English (n = 1). Children were recruited through mailings, advertisements, and word of mouth. Although the majority of participants were Caucasian, 6 were African American and 3 were Asian. Regarding parents’ education, 13 mothers had an advanced graduate degree, 10 had a bachelor’s degree, 4 had completed some college, and 3 had earned a high school diploma/GED. Materials Black and white line drawings of 24 novel items were created. This relatively simple pictorial format reflects classic work in the area of inductive inference. Indeed, several of the stimuli were adapted from those used in Davidson and Gelman (1990), as provided to us by the authors. Following Davidson and Gelman, half of our pictorial stimuli represented novel animals and half represented novel artifacts and, therefore, were broadly representative of domains in which inductive inference is critical for efficient generalization of knowledge. The 24 images were organized into eight triads, each consisting of two training items and one target (see examples in Fig. 1). Both training items bore some perceptual resemblance to the target, but one was consistently rated by 14 adults on a 7-point Likert scale (where 7 = maximally similar) as perceptually more similar to the target than the other (mean difference in similarity ratings = 2.60, SD = 1.16; mean similarity rating of training items relative to each other = 3.00, SD = 0.90). Design Although our initial plan was to test each child on all of the stimuli, piloting revealed that many children were unable to continue through eight trials of this information-heavy task. Therefore, we halved the trials by presenting only artifacts to half of the children and only animates to the others. Our experimental manipulation, however, proceeded entirely within participants. That is, each child completed one trial in each of our four experimental conditions (Baseline No Label, Label Alone, Label + Causal, and Label + Non-Causal). Although visual stimuli were always presented in the same order, conditions were assigned to these stimuli in one of four unique orders such that the key experimental manipulation was temporally counterbalanced. Trials were identical across experimental conditions except in the type of information children learned about training stimuli. Children in the Baseline No Label condition heard descriptions that offered no information at all about the unique properties of the pictured item (e.g., ‘‘I really like this one. Isn’t it neat? It’s really cool’’). Children in the Label Alone condition heard descriptions that consisted of an attention-getting phrase followed by four repetitions of a novel label for the pictured item (e.g., ‘‘It’s called a dur. See, it’s a dur. Can you say ‘dur’? Remember, it’s a dur’’). Children in the

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

Target

dur

Different-label (visually similar) training item

This is a fot.

Same-label (visually dissimilar) training item

This is a dur.

Non-causal:

Has bright green buttons on it.

Has soft pink buttons on it.

Causal: Test property:

Turns orange juice into bubbly soda. Has a round ball inside.

Turns bananas into chewy candy. Has a stretchy rubber band inside.

jop

This is a yup.

Test property:

Makes a whistling sound when it sleeps. Searches the bottom of rivers for clams to eat. Has white blood.

This is a jop. Makes a growling sound when it sleeps. Searches underneath rocks for slugs to eat. Has black blood.

nin

This is a wace.

This is a nin.

Non-causal: Causal:

5

Non-causal:

Has blue stripes on the bottom.

Has yellow bumps on the bottom.

Causal:

Tells people when the mail has come.

Tells people when it is going to rain.

Test property:

Has lots of switches inside.

Has twisty tubes inside.

zav

This is a pume.

This is a zav.

Non-causal:

Has tiny bumps on its belly.

Has cold spots on its belly.

Causal:

Spits smelly goo at animals that try to get it.

Throws sand in the eyes of animals that try to get it.

Test property:

Has a striped tongue.

Has a polka-dotted tongue.

Fig. 1. Examples of novel drawings, along with their assigned labels, non-causal versus causal training descriptions, and test generalization properties.

Label +Non-Causal and Label + Causal conditions heard descriptions that provided four repetitions of a novel label plus a non-obvious fact about the pictured item (see Fig. 1). Causal and Non-Causal descriptions were matched in length (M = 7.93, SD = 2.11 vs. M = 7.25, SD = 2.50 words) but were distinguished from each other by the degree to which they provided coherent conceptually relevant information about the pictured item. For animals, Causal descriptions focused on survival-related behaviors such as eating, shelter building, and defense (e.g., ‘‘This spits smelly goo at animals that try to get it’’). For artifacts, they focused on object function (e.g., ‘‘This tells

6

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

people when it is going to rain’’). Non-Causal descriptions also provided new information that was not perceptually obvious but did not explicitly identify a purposeful behavior or use of the pictured item (e.g., ‘‘This has blue stripes on the bottom’’). See Fig. 1 for examples. Although we use the term ‘‘NonCausal’’ to refer to these descriptions, it is important to recognize the possibility that participants could spontaneously imagine some causal basis for, or use for, the highlighted properties. Nevertheless, we confirmed successful manipulation of causal richness by asking 29 adults to rate each description (on a 5-point scale) on the degree to which it provided information about the causal properties of the pictured item (where 1 = low and 5 = high). Causal descriptions were rated as significantly more causally informative (M = 3.49, SD = 0.75) than Non-Causal descriptions (M = 1.84, SD = 0.76), t = 7.53, p < .001. We further confirmed that this manipulation was, as intended, relevant to perceived conceptual coherence by asking adults to rate each description (on the same 5-point scale) on the degree to which it provided them with a greater understanding of those categories (see Murphy & Medin, 1985, for relevant discussion). Adults rated the Causal descriptions (M = 3.23, SD = 0.65) as imparting significantly greater understanding than the Non-Causal descriptions (M = 1.52, SD = 0.75), t = 11.67, p < .001. Procedure Overview After a warm-up period with the experimenter, children were escorted into a quiet testing room. Parents were encouraged to wait outside, but in the unusual circumstance where children were uncomfortable separating, the parents were invited to sit quietly in a corner of the testing room behind their children. Children were told that they were going to play a game in which they would see lots of new things that they have never seen before. To minimize children’s tendency to assimilate unfamiliar items into known categories, they were told that all of the things were from a far-away planet and could not be found on Earth. Children then completed a two-phase procedure for each of four sets of stimuli, thereby providing a single response per experimental condition (i.e., Baseline No Label, Label Alone, Label + Non-Causal, and Label + Causal). Training The experimenter first introduced both training items from a stimulus triad to children. While holding each item forward toward children in turn, the experimenter described the item according to the condition-appropriate script. Whether the visually similar (different-label) or visually dissimilar (same-label) training item was described first was alternated across trials. On Label Alone, Label + Non-Causal, and Label + Causal iterations, children were then probed on their memory for the labels just provided. The experimenter briefly removed the two stimuli from view and replaced them on the table in random order, asking, for example, ‘‘Which one is the traw? Point to the traw.’’ If children answered correctly, the experimenter said, ‘‘That’s right! There’s the traw.’’ If they answered incorrectly, the experimenter said, ‘‘Oops! Is that the traw? Point to the traw.’’ Children invariably switched their choice at this time. The stimuli were then briefly shuffled out of children’s view before the experimenter placed them on the table again and asked children to identify the referent of the second label they were just taught. The same feedback protocol was followed. If children failed to identify the correct referent of either label, the entire training procedure was repeated. If children identified both referents correctly, and they had just been taught non-obvious properties in addition to the new words (i.e., they were experiencing a Label + Non-Causal or Label + Causal iteration), they were next probed on their memory for those properties. After another brief reshuffle and random placement of the training stimuli on the table, children were asked, for example, ‘‘Which one has blue stripes on the bottom? Point to the one with blue stripes on the bottom.’’ Children were again given feedback on their responses. If children answered either question incorrectly, the entire training procedure was repeated. Training was completed when children either responded correctly to all memory probes or completed three full repetitions of the training procedure. Because no content information was provided in the Baseline No Label condition, no label or non-obvious property memory probes were administered for those trials. Because children had a fairly high (25%) chance of reaching criterion on any round of testing, and because correct responses were eventually elicited on every trial, we did not use

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

7

performance during training as an exclusion criterion. Rather, training was intended to provide some modicum of assurance that children had sufficient exposure to the training information to support its potential influence on subsequent testing. This was particularly important for the non-obvious descriptions because, in contrast to the labels, the experimenter did not repeat these during testing. In a retrospective analysis, children took an equal number of trials to reach the learning criterion in all conditions. Inductive generalization testing The experimenter then tested children’s generalization of new non-obvious properties from the training items to the target (which you may recall was perceptually more similar to one of the training items than the other). With both training items placed in random order on the table, the experimenter began by describing a new non-obvious ‘‘generalization property’’ of each. To highlight the contrast between pairs of training items with respect to these properties, they were crafted to be maximally alignable (Gentner & Gunn, 2001; Markman & Gentner, 1993). Specifically, distinctive aspects of the same internal part were highlighted as contrasting generalization properties within each pair (e.g., a striped vs. polka-dotted tongue). The experimenter then introduced the target item and called attention to it. In the experimental conditions, this was achieved by labeling the target with the same name as was previously applied to the visually dissimilar training item (e.g., ‘‘This is a zav’’). In the Baseline condition, the experimenter instead used a neutral phrase, for example, ‘‘Look at this one.’’ Importantly, the experimenter never described any non-obvious properties of the target so that only the label and visually obvious properties of the target were ever known to children. Children were told that they were going to be asked a question about the target item and that it was important for them to listen carefully to the whole question and answer only at the very end. In the experimental conditions, the experimenter then asked, for example, ‘‘Do you think that this zav has a striped tongue like this pume over here, or do you think that this zav has a polka-dotted tongue like this zav over here?’’ In the Baseline condition, the novel count nouns were all replaced by ‘‘one.’’ The experimenter pointed to each item as she referred to them in the question. Which training item was referenced first was randomized across participants for each condition and was alternated across trials for each participant. After children indicated their selection by pointing and/or saying the name of one of the training items, the experimenter indiscriminately responded with a cheery, ‘‘Thank you! You are doing great!’’ The experimenter then allowed children to take a brief break to choose a sticker and place it in their sticker book before proceeding with the next iteration of training and testing or, if all four stimulus triads (one per condition) were already presented, ending the experimental procedure. Coding Children’s selections of the same-label item at test were recorded by the experimenter during task implementation and then checked from video by a secondary coder who was blind to the experimental hypotheses. The experimenter and secondary coder were in 100% agreement, including with respect to two responses (one in the Causal condition and one in the Non-Causal condition) that needed to be eliminated from analyses due to lack of clarity in children’s choice. Results The dependent variable of interest was the number of children selecting the dissimilar test item. In all but the No Label control condition, this corresponded to label-based responding. The within-participants and dichotomous nature of these data dictated that we examine the effect of condition (No Label, Label Alone, Label + Causal, or Label + Non-Causal) with a mixed effects logistic regression. Although we included domain (animate or artifact) as a between-participants factor in preliminary analyses, as expected, it produced no significant main effect, Wald 2(df = 1, N = 30) = 0.671, ns, or interaction with condition, Wald 2(df = 3, N = 30) = 4.99. Given that we had no a priori predictions regarding differential performance across domains, we dropped this factor from subsequent consideration. The

8

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

Proportion of children generalizing from the perceptually dissimilar test item

** 90

*

80 70

*

60 50 40 30 20 10 0

No Label

Label Alone

Label + NonCausal

Label + Causal

Fig. 2. Proportion of children (with standard errors) who selected the perceptually dissimilar test item in each condition. In all but the No Label condition, this further reflects children’s reliance on names (rather than perceptual similarity) to guide inductive inference. An asterisk (⁄) indicates significantly more perceptually dissimilar responses in both the Label and Label + Causal conditions relative to the No Label baseline. The double asterisk (⁄⁄) indicates significantly more label-based (perceptually dissimilar) responding in the Label + Causal condition than in the Label + Non-Causal condition.

resulting omnibus analysis revealed a significant main effect of condition, Wald 2(df = 3, N = 30) = 16.49, p = .001. See Fig. 2 for an overview of performance in each condition. To test whether conceptually coherent information influences the likelihood with which children rely on labels to guide their inductive generalizations, we next specifically compared performance in the Label + Non-Causal and Label + Causal conditions. As predicted, more children relied on labels (rather than perceptual similarity) in the Label + Causal condition (72%) than in the Label + Non-Causal condition (28%), Wald 2(df = 1, N = 30) = 11.35, p = .001, Cramer’s V = .62 (large effect size). Although less central to our hypotheses, it is worth noting that label-based responding in the Label + Causal condition was also higher than that in the Label Only condition (47%), Wald 2(df = 1, N = 30) = 3.71, p = .05, Cramer’s V = .35 (medium effect size). It was also higher than choices of the dissimilar test item in the No Label condition (20%), Wald 2(df = 1, N = 30) = 13.63, p = .001, Cramer’s V = .67 (large effect size). Although label-based responding (i.e., dissimilar selections) in the Label Alone condition was also more likely than dissimilar item selections in the No Label condition, Wald 2(df = 1, N = 30) = 4.74, p = .03, Cramer’s V = .40 (medium effect size), no significant differences were observed in the remaining comparisons of the Label + Non-Causal condition with the Label Alone condition, Wald 2(df = 1, N = 30) = 2.30, p = .13, or the No Label condition, Wald 2(df = 1, N = 30) = 0.41, p = .52. Discussion The goal of this investigation was to directly test the role of conceptual information in mediating the effect of words on early inductive generalization. This goal was motivated by ongoing debate between advocates of two divergent accounts of why young children rely heavily on shared labels in generalizing newly learned information. The word-as-feature account argues that shared labels contribute by enhancing the perceived similarity between items (Sloutsky, 2003; Sloutsky & Lo, 1999; Sloutsky et al., 2001). According to this perspective, conceptual knowledge does not begin to exert any influence on inductive generalization until children are well beyond preschool. In contrast, the knowledge-based account argues that shared labels contribute by clarifying the kind membership of items for young children. According to this perspective, both general conceptual knowledge regarding

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

9

the status of count nouns relative to meaningful categories and specific conceptual knowledge of the categories involved in any particular task can have a powerful impact on inductive generalization in preschoolers and even infants (Gelman & Markman, 1986, 1987; Gelman & Waxman, 2007; Keates & Graham, 2008; Waxman & Gelman, 2009). Our approach involved devising conditions under which the availability of labels was held constant, whereas the information available for supporting a construal of those labels as reliable markers of coherent kinds varied. Although conceptual coherence might have been manipulated in a number of ways (e.g., identifying items as representing subordinates of familiar global categories, describing ‘‘essential’’ features, making connections to existing conceptual knowledge), we chose to vary the causal richness of information conveyed about training items prior to the inductive generalization task (Gelman, 2003; Murphy & Medin, 1985). If labels support inductive generalization when they clearly index coherent kinds, a larger influence of labels should have been evident in the Label + Causal condition than in the Label + Non-Causal condition. If, on the other hand, labels simply contribute to perceived similarity among items, their influence should have been unaffected by this manipulation. Results were consistent with the former possibility, revealing greater label-based generalization in the Label + Causal condition than in the Label + Non-Causal condition. It is difficult to see how this result could be reconciled with the word-as-feature account. First, precisely the same words (and visual stimuli) were provided in the Label + Causal and Label + Non-Causal conditions. Thus, similarity calculations among training and target stimuli should have been equivalent. Although proponents of the word-as-feature account might well argue that the descriptions provided ‘‘linguistic’’ rather than ‘‘conceptual’’ information that could have entered into similarity calculations, because this information was provided only for the training items (not the target) and was presented before (rather than during) the inductive generalization task, the non-obvious properties could not be used in calculating similarity between the target and training items. Thus, it is not clear that this recharacterization of the stimuli improves the viability of the word-as-feature account (see Smith & Samuelson, 2006, and Booth & Waxman, 2006, for a similar argument and response in the context of early word learning). Second, the manipulation was implemented within participants, with the ordering of conditions balanced across children. Thus, differences in performance cannot be easily attributed to spurious group characteristics or design factors. Third, when (as an added check on validity) we calculated the number of training trials experienced by children in the Label + Causal and Label + Non-Causal conditions before reaching the learning criterion, we detected no differences. Thus, we can rule out any potential influence of differences in the time children spent viewing the stimuli or in the number of repetitions of the words and/or non-obvious information they heard. Although it is true that 2 children failed to show consistent evidence of having explicitly learned the non-obvious information (i.e., they did not supply correct answers to both of the memory probes even after three rounds of training) in the Label + Non-Causal condition, and only 1 child failed to do so in the Label + Causal condition, elimination of these data points from the analyses did not alter the key findings reported here. The difference in responding observed in the Label + Causal and Label + Non-Causal (and Label Alone) conditions is, however, highly consistent with the knowledge-based account of early inductive generalization. When conceptual support was provided for interpreting labels as reliable markers of kinds (i.e., through the provision of causal descriptions), those labels were significantly more effective in guiding inductive generalization than when that support was not provided. It is interesting to further note that, although there was no significant difference in performance across the Label Alone and Label + Non-Causal conditions, only the former promoted label-based responding above baseline. One possibility that could be usefully evaluated in future work is that highlighting superficial and arbitrary object properties actually draws children’s attention away from their kind membership, thereby reducing reliance on labels. Of course, the compatibility of the current results with the knowledge-based account hinges on a relatively nuanced view of the relationship between labels and kinds in the minds of young children. On this view, children do not automatically assume that all items labeled by count nouns belong to coherent kinds with strong inductive potential (Gelman & Waxman, 2007; McCarrell & Callanan, 1995; Noles & Gelman, 2012b). Rather, although count nouns remain a powerful cue in and of themselves, children also take a variety of supplemental forms of information into account in making this

10

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

determination. Some of these factors have already been identified in the literature, including the reliability and intentions of the speaker, the linguistic context in which labels are presented, and the conceptual coherence of the properties of labeled items (Chapman et al., 1986; Davidson & Gelman, 1990; Jaswal, 2004; Jaswal & Markman, 2007; McCarrell & Callanan, 1995; Mervis & Mervis, 1988; Sabbagh & Baldwin, 2001; Sobel & Corriveau, 2010). It is important to note that novel labels alone did successfully promote label-based inductive generalization (above baseline) in the current work despite the fact that we provided little basis for children to interpret these words as reliable markers of kind membership in the Label Alone condition. In particular, our stimuli were entirely novel and unnaturally dissociated perceptual similarity from labels. Nevertheless, our stimuli might well have been sufficiently complex and detailed to be convincing to children as sketches of real objects from real categories of animals and artifacts. Indeed, despite our ‘‘alien planet’’ pretense, children frequently supplied their own familiar labels for our novel stimuli (e.g., dinosaur, dog, spaceship). An important direction for future research will be to further specify the conditions under which children do and do not accept labels as convincing markers of new kinds. It seems likely that a complex interplay among factors that affect generalization in a probabilistic fashion will be identified. More broadly, how can the knowledge-based view be reconciled with the extant evidence as a whole? One important thing to keep in mind when reviewing this literature is that demonstrations of the contributions of associative learning, attentional weighting, and perceptual similarity to inductive generalization in preschoolers are not in and of themselves incompatible with the knowledgebased account. Proponents of this view acknowledge the ubiquitous contributions of these general processes to a wide range of cognitive tasks. They do not eschew the processes at the heart of the word-as-feature account in the same way as proponents of the word-as-feature account eschew the role of conceptual factors. Therefore, it is cases in which conceptual information failed to influence inductive generalization that present the greatest challenge to the knowledge-based view. Three such types of demonstrations have recently been reported in the literature. First, Fisher and colleagues (2011) reported that synonymous words (e.g., bunny and rabbit) support inductive generalization in 4-year-olds only if they also have high co-occurrence frequencies in the natural language input heard by young children. This finding is seemingly problematic for the knowledge-based account because two words that access the same kind would be predicted to support inductive generalization regardless of their statistical representation in the speech input (see Gelman & Markman, 1986). However, although paired words in this work were always semantically related, they were not always actually synonyms (e.g., dolphin and whale). As a result, they did not always reference the same kind and, therefore, would not be expected by proponents of the knowledge-based view to enhance inductive generalization. The fact that more low- than high-co-occurrence word pairs referred to non-isomorphic concepts substantially weakens the relevance of these findings to the current theoretical debate. Moreover, because the check on children’s knowledge of the words used in this study was quite superficial (forced-choice picture recognition), it is unclear whether the youngest children were even capable of spontaneously calling to mind the concepts associated with them, let alone assessing their semantic equivalence to each other. A second approach to disconfirming the influence of conceptual knowledge on early inductive generalization has involved first teaching children the names of contrasting novel categories and then testing their inductive generalizations over unlabeled exemplars from those categories (Badger & Shapiro, 2012; Sloutsky, Kloos, & Fisher, 2007b). The advantage of this approach is that any potential perceptual influence of labels can be removed from the inductive generalization task itself, thereby allowing a more targeted view on the influence of categories. The problem, however, is that, as pointed out by Gelman and Waxman (2007), the knowledge-based view predicts that kinds will guide inductive generalization, not just any old categories. Despite arguments to the contrary (Sloutsky et al., 2007b), the arbitrary categories taught in these studies might not have been plausible as conceptually meaningful kinds. Even though exemplars of one category were described by Sloutsky et al. (2007a) as ‘‘friendly pets’’ while exemplars of the other category were described as ‘‘vicious wild animals,’’ the pretense explained that both were sold in pet stores, thereby making this potentially meaningful distinction difficult to interpret. In addition, although children in a supplemental control condition were given additional information meant to explain the relevance of the definitive category-differentiating

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

11

feature (ratio of fingers to body buttons), the explanation (that a chemical in the blood of animals from one of the categories makes their fingers, which they use to catch food, sticky) really only highlighted the relevance of fingers, generally speaking, rather than the number of fingers or, more critically, the ratio of that number to the number of body buttons. As a result, it adds little convincing conceptual coherence to the categories. In addition, although Badger and Shapiro (2012) provided a modestly more convincing causal explanation for why members of their contrasting bug categories have differently shaped heads, the overall conceptual coherence and distinctiveness of these categories remains relatively weak. Indeed, when children are taught conceptually more distinct categories (e.g., from different ontological domains) that can be perceptually differentiated on the basis of a conceptually relevant feature (e.g., eyes for animals and bolts for artifacts), young children do use categories over global perceptual similarity to guide their inferences in this type of task (Gelman & Davidson, 2013). Third, and finally, Deng and Sloutsky (2012) recently demonstrated that a single salient feature (type of head movement) could override shared novel labels in guiding inductive inference in 4-year-olds (see also Deng & Sloutsky, 2013). But the same kinds of problems identified above are relevant to this study as well. Children were taught novel categories that had no deeper conceptual structure whatsoever. Moreover, on the critical inductive inference trials, labels for these categories were dissociated from the feature that children had just learned to be criterial for category membership (head movement), thereby providing strong evidence that the speaker (who in this case was not even a human) was unreliable as a namer of things. As a result, children may have discounted the value of the labels as a useful guide to their responses on these induction trials. The fact that, as a group, adults responded at chance on these trials is particularly telling regarding the pragmatically confusing nature of this scenario. In addition to these concerns, Deng and Sloutsky (2012) study could be explained in terms of developmental differences in inhibitory control. Given how powerful motion can be as an attentional magnet (especially when presented in an otherwise static scene), it seems entirely possible that the young participants were simply unable to inhibit their response to the head motion (Franconeri & Simons, 2003; Kellman, 1993; Lavie, 2005; Wolfe & Horowitz, 2004). In conclusion, each of the seeming challenges to the knowledge-based account presented in the literature, thus, can be alternatively explained by task demands that are disproportionately burdensome on the limited executive functioning resources of young children and/or by insufficient attention to presenting children with compelling kinds over which to generalize. In contrast, the current work provides strong positive evidence that conceptual information contributes in critical ways to inductive generalization in young children. Specifically, it suggests that children capitalize on conceptual information to flexibly determine, on a case-by-case basis, how heavily they should rely on specific names to generalize novel object properties. This conclusion parallels that reached by Gelman and Davidson (2013) with respect to children’s selective reliance on categories that are conceptually rich and distinctive in inductive inference tasks. Together, these findings are more broadly consistent with a wealth of other evidence suggesting that young children, and even infants, are sensitive to conceptual information and that they capitalize on this information in categorization, naming, and inductive inference tasks (e.g., Ahn, Gelman, Amsterlaw, Hohenstein, & Kalish, 2000; Booth, 2001, 2008, 2009; Booth & Waxman, 2002b; Booth, Waxman, & Huang, 2005; Diesendruck, Gelman, & Lebowitz, 1998; Diesendruck, Markson, & Bloom, 2003; Florian, 1994; Gelman & Markman, 1986; Gopnik & Schulz, 2007; Gopnik & Sobel, 2000; Graham & Kilbreath, 2007; Graham et al., 2011; Heyman & Gelman, 2000; Jaswal, 2006; Jaswal & Markman, 2007; Kemler Nelson, Frankenfield, Morris, & Blair, 2000; Kemler Nelson, Russell, Duke, & Jones, 2000; Lombrozo, 2009; McCarrell & Callanan, 1995; Ware & Booth, 2010; Kemler Nelson, O’Neill, & Asher, 2008). We believe that the weight of this evidence is too great for any theory seeking to deny a role for conceptual information in early cognition to withstand. Importantly, this is not to say that perceptual and attentional factors make no contribution to early inductive generalization and to related processes more generally speaking. We believe that they in fact make critically important contributions and emphasize that the current results cannot rule out the specific possibility that labels contribute as perceptual features in addition to their role as markers of kinds. Therefore, we advocate for an inclusive perspective and propose that the best way forward is to now devote our collective energies to specifying how conceptual information might interface with other types of input, as well as with core

12

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

cognitive processes such as associative and perceptual learning, attention, and memory, to support development. Acknowledgments This research was generously supported by a Grant from the National Science Foundation (BCS084352). We also thank Susan Gelman for sharing her stimuli (from Davidson & Gelman, 1990) with us. We are also indebted to Kathy McGroarty-Torres, Aubry Alvarez, and Brianne Williams for their help in collecting data. Finally, we thank all of the children and parents who generously shared their time and minds with us so that we could pursue this research. References Ahn, W.-K., Gelman, S. A., Amsterlaw, J. A., Hohenstein, J., & Kalish, C. W. (2000). Causal status effect in children’s categorization. Cognition, 76, B35–B43. Badger, J. R., & Shapiro, L. R. (2012). Evidence of a transition from perceptual to category induction in 3- to 9-year-old children. Journal of Experimental Child Psychology, 113, 131–146. Booth, A. E. (2001). The facilitative effect of agent-produced motions on categorization in infancy. Infant Behavior and Development, 23, 153–174. Booth, A. E. (2008). The cause of infant categorization? Cognition, 106, 984–993. Booth, A. E. (2009). Causal supports for early word learning. Child Development, 80, 1243–1250. Booth, A. E., & Waxman, S. R. (2006). Déjà Vu all over again: re-revisiting the conceptual status of early word learning: comment on Smith and Samuelson (2006). Developmental Psychology, 42(6), 1344–1346. Booth, A. E., & Waxman, S. R. (2002a). Object names and object functions serve as cues to categories for infants. Developmental Psychology, 38, 948–957. Booth, A. E., & Waxman, S. R. (2002b). Word learning is ‘‘smart’’: Evidence that conceptual information affects preschoolers’ extension of novel words. Cognition, 84, B11–B22. Booth, A. E., Waxman, S. R., & Huang, Y. T. (2005). Conceptual information permeates word learning in infancy. Developmental Psychology, 41, 491–505. Chapman, K. L., Leonard, L. B., & Mervis, C. B. (1986). The effect of feedback on young children’s inappropriate word usage. Journal of Child Language, 13, 101–117. Davidson, N. S., & Gelman, S. A. (1990). Inductions from novel categories: The role of language and conceptual structure. Cognitive Development, 5, 151–176. Deak, G. O., & Bauer, P. J. (1996). The dynamics of preschoolers’ categorization choices. Child Development, 67, 740–767. Deng, W., & Sloutsky, V. M. (2012). Carrot eaters or moving heads: Inductive inference is better supported by salient features than by category labels. Psychological Science, 23, 178–186. Deng, W., & Sloutsky, V. M. (2013). The role of linguistic labels in inductive generalization. Journal of Experimental Child Psychology, 114(3), 432–455. Diesendruck, G., Gelman, S. A., & Lebowitz, K. (1998). Conceptual and linguistic biases in children’s word learning. Developmental Psychology, 34, 823–839. Diesendruck, G., Markson, L., & Bloom, P. (2003). Children’s reliance on creator’s intent in extending names for artifacts. Psychological Science, 14, 164–168. Fisher, A. V. (2010). What’s in the name? Or how rocks and stones are different from bunnies and rabbits. Journal of Experimental Child Psychology, 105, 198–212. Fisher, A. V., Matlen, B. J., & Godwin, K. E. (2011). Semantic similarity of labels and inductive generalization: Taking a second look. Cognition, 118, 432–438. Florian, J. E. (1994). Stripes do not a zebra make, or do they? Conceptual and perceptual information in inductive inference. Developmental Psychology, 30, 88–101. Franconeri, S. L., & Simons, D. J. (2003). Moving and looming stimuli capture attention. Perception & Psychophysics, 65, 999–1010. Gelman, S. A. (2003). The essential child: Origins of essentialism in everyday thought. New York: Oxford University Press. Gelman, S. A., & Coley, J. D. (1990). The importance of knowing a dodo is a bird: Categories and inferences in 2-year-old children. Developmental Psychology, 26, 796–804. Gelman, S. A., & Davidson, N. S. (2013). Conceptual influences on category-based induction. Cognitive Psychology, 66, 327–353. Gelman, S. A., & Markman, E. M. (1986). Categories and induction in young children. Cognition, 23, 183–209. Gelman, S. A., & Markman, E. M. (1987). Young children’s inductions from natural kinds: The role of categories and appearances. Child Development, 58, 1532–1541. Gelman, S. A., & Waxman, S. R. (2007). Looking beyond looks: Comments on Sloutsky, Kloos, and Fisher (2007). Psychological Science, 18, 554–555. Gentner, D., & Gunn, V. (2001). Structural alignment facilitates the noticing of differences. Memory & Cognition, 29, 565–577. Gopnik, A., & Nazzi, T. (2003). Words, kinds, and causal powers: A theory theory perspective on early naming and categorization. In D. H. O. Rakison & M. Lisa (Eds.). Early Category and Concept Development. New York: Oxford University Press. Gopnik, A., & Schulz, L. (2007). Causal learning: Psychology, philosophy, and computation. New York: Oxford University Press. Gopnik, A., & Sobel, D. M. (2000). Detecting blickets: How young children use information about novel causal powers in categorization and induction. Child Development, 71, 1205–1222. Graham, S. A., Booth, A. E., & Waxman, S. R. (2011). Words are not features of objects: Only consistently applied nouns guide 4year-olds’ inferences about object categories. Language Learning and Development, 8, 136–145.

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

13

Graham, S. A., & Kilbreath, C. S. (2007). It’s a sign of the kind: Gestures and words guide infants’ inductive inferences. Developmental Psychology, 43, 1111–1123. Hayes, B. K., & Heit, E. (2004). Why learning and development can lead to poorer recognition memory. Trends in Cognitive Sciences, 8, 337–339. Heit, E., & Hayes, B. K. (2005). Relations among categorization, induction, recognition, and similarity: Comment on Sloutsky and Fisher (2004). Journal of Experimental Psychology: General, 134, 596–605. Heyman, G. D., & Gelman, S. A. (2000). Preschool children’s use of trait labels to make inductive inferences. Journal of Experimental Child Psychology, 77, 1–19. Jaswal, V. K. (2004). Don’t believe everything you hear: Preschoolers’ sensitivity to speaker intent in category induction. Child Development, 75, 1871–1885. Jaswal, V. K. (2006). Preschoolers favor the creator’s label when reasoning about an artifact’s function. Cognition, 99, B83–B92. Jaswal, V. K., & Markman, E. M. (2007). Looks aren’t everything: 24-month-olds’ willingness to accept unexpected labels. Journal of Cognition and Development, 8, 93–111. Jones, S. S., & Smith, L. B. (1993). The place of perception in children’s concepts. Cognitive Development, 8, 113–139. Keates, J., & Graham, S. A. (2008). Category markers or attributes: Why do labels guide infants’ inductive inferences? Psychological Science, 19, 1287–1293. Kellman, P. J. (1993). Kinematic foundations of infant visual perception. In C. Granrud (Ed.), Visual perception and cognition in infancy: Carnegie Mellon Symposia on Cognition (pp. 121–173). Hillsdale, NJ: Lawrence Erlbaum. Kemler Nelson, D. G., Frankenfield, A., Morris, C., & Blair, E. (2000). Young children’s use of functional information to categorize artifacts: Three factors that matter. Cognition, 77, 133–168. Kemler Nelson, D. G., O’Neill, K., & Asher, Y. M. (2008). A mutually facilitative relationship between learning names and learning concepts in preschool children: The case of artifacts. Journal of Cognition and Development, 9, 171–193. Kemler Nelson, D. G., Russell, R., Duke, N., & Jones, K. (2000). Two-year-olds will name artifacts by their functions. Child Development, 71, 1271–1288. Lavie, N. (2005). Distracted and confused? Selective attention under load. Trends in Cognitive Sciences, 9, 75–82. Lombrozo, T. (2009). Explanation and categorization: How ‘‘why?’’ informs ‘‘what?’’. Cognition, 110, 248–253. Markman, A. B., & Gentner, D. (1993). Structural alignment during similarity comparisons. Cognitive Psychology, 25, 431–467. McCarrell, N. S., & Callanan, M. A. (1995). Form–function correspondences in children’s inference. Child Development, 66, 532–546. Mervis, C. B., & Mervis, C. A. (1988). Role of adult input in young children’s category evolution: I. An observational study. Journal of Child Language, 15, 257–272. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289–316. Noles, N. S., & Gelman, S. A. (2012a). Effects of categorical labels on similarity judgments: A critical analysis of similarity-based approaches. Developmental Psychology, 48, 890–896. Noles, N. S., & Gelman, S. A. (2012b). Preschool-age children and adults flexibly shift their preferences for auditory versus visual modalities but do not exhibit auditory dominance. Journal of Experimental Child Psychology, 112, 338–350. Robinson, C. W., & Sloutsky, V. M. (2004). Auditory dominance and its change in the course of development. Child Development, 75, 1387–1401. Sabbagh, M. A., & Baldwin, D. A. (2001). Learning words from knowledgeable versus ignorant speakers: Links between preschoolers’ theory of mind and semantic development. Child Development, 72, 1054–1070. Sloutsky, V. M. (2003). The role of similarity in the development of categorization. Trends in Cognitive Sciences, 7, 246–251. Sloutsky, V. M. (2008). Recognition memory and mechanisms of induction: Comment on Wilburn and Feeney. Cognition, 108, 500–506. Sloutsky, V. M., & Fisher, A. V. (2004). Induction and categorization in young children: A similarity-based model. Journal of Experimental Psychology: General, 133, 166–188. Sloutsky, V. M., & Fisher, A. V. (2005). Similarity, induction, naming, and categorization (SINC): Generalization or inductive reasoning? Reply to Heit and Hayes (2005). Journal of Experimental Psychology: General, 134, 606–611. Sloutsky, V. M., & Fisher, A. V. (2012). Linguistic labels: Conceptual markers or object features? Journal of Experimental Child Psychology, 111, 65–86. Sloutsky, V. M., Kloos, H., & Fisher, A. V. (2007a). What’s beyond looks? Reply to Gelman and Waxman. Psychological Science, 18, 556–557. Sloutsky, V. M., Kloos, H., & Fisher, A. V. (2007b). When looks are everything: Appearance similarity versus kind information in early induction. Psychological Science, 18, 144–151. Sloutsky, V. M., & Lo, Y.-F. (1999). How much does a shared name make things similar? Linguistic labels and the development of similarity judgment. Developmental Psychology, 35, 1478–1492. Sloutsky, V. M., Lo, Y.-F., & Fisher, A. V. (2001). How much does a shared name make things similar? Linguistic labels, similarity, and the development of inductive inference. Child Development, 72, 1695–1709. Sloutsky, V. M., & Napolitano, A. C. (2003). Is a picture worth a thousand words? Preference for auditory modality in young children. Child Development, 74, 822–833. Smith, L. B., & Samuelson, L. (2006). An attentional learning account of the shape bias: Reply to Cimplian and Markman (2005) and Booth, Waxman and Huang (2005). Developmental Psychology, 42(6), 1339–1343. Sobel, D. M., & Corriveau, K. H. (2010). Children monitor individuals’ expertise for word learning. Child Development, 81, 669–679. Ware, E. A., & Booth, A. E. (2010). Form follows function: The role of artifact function in the development of the shape bias. Cognitive Development, 25, 124–137. Waxman, S. R., & Booth, A. E. (2000). Principles that are invoked in the acquisition of words, but not facts. Cognition, 77, B33–B43. Waxman, S. R., & Gelman, S. A. (2009). Early word-learning entails reference, not merely associations. Trends in Cognitive Sciences, 13, 258–263.

14

A.E. Booth / Journal of Experimental Child Psychology 123 (2014) 1–14

Welder, A. N., & Graham, S. A. (2001). The influences of shape similarity and shared labels on infants’ inductive inferences about nonobvious object properties. Child Development, 72, 1653–1673. Wellman, H. M., & Gelman, S. A. (1992). Cognitive development: Foundational theories of core domains. Annual Review of Psychology, 43, 337–375. Wilburn, C., & Feeney, A. (2008). Do development and learning really decrease memory? On similarity and category-based induction in adults and children. Cognition, 106, 1451–1464. Wolfe, J. M., & Horowitz, T. S. (2004). What attributes guide the deployment of visual attention and how do they do it? Nature Reviews Neuroscience, 5, 495–501.