Object Similarity Bootstraps Young Children to …...went one step further and investigated whether repeated experience of action-based verb extension supported by object similarity

Object Similarity Bootstraps Young Children to Action-Based

Verb Extension

Etsuko HaryuUniversity of Tokyo

Mutsumi ImaiKeio University

Hiroyuki OkadaTamagawa University

Young children often fail to generalize a novel verb based on sameness of action since they have difficultyfocusing on the relational similarity across events while at the same time ignoring the objects that areinvolved. Study 1, with Japanese-speaking 3- and 4-year-olds (N = 28 in each group), found that similarity ofobjects involved in action events plays a scaffolding role in children’s extraction of relational similarity acrossevents when they extend a verb. Study 2, with 4-year-olds (N = 47), further showed that repeated experienceof action-based verb extension supported by object similarity leads children to be better able to extend a novelverb based on sameness of action, even without support from object similarity.

Verbs refer to kinds of relations, whereas nouns,especially concrete nouns, denote kinds of objects.Thus, in order to learn a novel word, children needto identify the word’s form class and map the wordto a meaning that is characteristic of the form class.When children hear a novel noun, their task is tofind the referent object and generalize the word toother like objects. In contrast, when the novel wordis a verb, children must map the word to the relationthey find in the present scene and extend the wordbased on the sameness of the relation. However, it isnot a trivial task to figure out which relation that isobserved in the scene should be incorporated intothe meaning of a novel verb. For example, when wehear a novel verb in association with a scene inwhich someone is rolling a ball, the verb could meanthe agent’s action of releasing the ball, the rollingmotion of the ball, or the departure of the ball fromthe agent. Thus, researchers have argued that verblearning is more difficult for young children than

noun learning (e.g., Childers & Tomasello, 2002,2006; Gentner, 1982; Hirsh-Pasek & Golinkoff, 2006;Imai, Haryu, & Okada, 2005; Imai et al., 2008;Kersten & Smith, 2002; Maguire et al., 2002).

Supporting this view, Kersten and Smith (2002)introduced a novel verb for a scene in which anovel bug-like creature moved in a distinctive way,and found that 3-year-olds were not willing toapply this verb to a different creature that wasmoving in the same way. In contrast, children ofthe same age readily applied a novel noun intro-duced for the same scene to the same creature mov-ing in a different way. Maguire et al. (2002) foundthat 18-month-olds who were introduced to a novelverb during a video of an intransitive action failedto generalize the verb to the same action performedby a new agent, even after they heard the verbrepeatedly in association with the identical actionperformed by four different people. Other studiesalso showed that young children were reluctant togeneralize a novel verb associated with a noveltransitive action to the same action performed witha different object (Behrend, 1990; Forbes & Farrar,1993; Imai et al., 2005; Imai et al., 2008).

Young children have difficulty in extending anovel verb based upon sameness of action, especially

We thank the children, preschool teachers, and parents verymuch for their participation in this research. We are also verygrateful to anonymous reviewers who gave us insightful com-ments on the earlier version of our manuscript. This researchwas supported by Grants-in-Aid for Scientific Research from theMinistry of Education, Culture, Sports, Science and Technologyawarded to Etsuko Haryu (Grants 17330139 and 20330135) andMutsumi Imai (Grant 18300089).

Correspondence concerning this article should be addressed toEtsuko Haryu, University of Tokyo, Hongo, Bunkyo-ku, Tokyo,113-0033, Japan. Electronic mail may be sent to [email protected].

Child Development, March/April 2011, Volume 82, Number 2, Pages 674–686

� 2011 The Authors

Child Development � 2011 Society for Research in Child Development, Inc.

All rights reserved. 0009-3920/2011/8202-0016

DOI: 10.1111/j.1467-8624.2010.01567.x

when the action is performed by a different agent(Kersten & Smith, 2002; Maguire et al., 2002) orwith a different object (Behrend, 1990; Forbes &Farrar, 1993; Imai et al., 2005; Imai et al., 2008). How-ever, it is not the case that children of this age assumethat verbs can refer to objects. For example, in Imaiet al. (2005), 3-year-olds who were introduced to anovel verb in association with a novel action per-formed by a woman with a novel object would notextend the verb to another video showing the sameobject lying still on a table. Young children do under-stand that verbs should be extended by a differentprinciple than for noun extension, but they havedifficulty in identifying the core meaning of a novelverb, that is, the common relation between objects.

Young children’s difficulty in extracting rela-tional commonality across events has also beenreported in previous studies of analogical reasoning(Gentner, 1988; Gentner & Ratterman, 1991;Loewenstein & Gentner, 2005). For example, whenasked to interpret a metaphor such as ‘‘Plant stemsare like drinking straws,’’ young children werelikely to say, ‘‘They are both long,’’ focusing onlower order surface similarity. In contrast, olderchildren would say, ‘‘They are both used for draw-ing water,’’ attending to higher order relationalsimilarity (Billow, 1975; Gentner, 1988). Further-more, when relational similarity was pitted againstobject similarity in a task in which one of the testitems was to be matched to the standard item,young children tended to select on the basis ofobject similarity rather than on the basis of rela-tional similarity (Gentner & Ratterman, 1991). Thus,focusing on relational commonality, especiallywhen there is competing lower order surface simi-larity, is very difficult for young children, not onlyin verb extension but also in analogical reasoning.

Studies of analogical reasoning have also sug-gested that young children are better able to makerelational mappings when the objects involved inthe base and the target are perceptually similar(DeLoache, 1989, 1990; Gentner & Toupin, 1986;Holyoak, Junn, & Billman, 1984; Loewenstein &Gentner, 2001). For example, Gentner and Toupin(1986) asked young children to act out stories withnew toy characters and found that the childrencould more successfully map a plot structure fromone set of characters to another when the corre-sponding characters (i.e., the characters with thesame role in the stories) were similar (e.g., a squir-rel and a chipmunk) than when they were dissimi-lar (e.g., a squirrel and a moose). DeLoache (1989,1990) examined preschoolers’ ability to mapbetween a regular-sized, ordinary room and its 3-D

miniature and found that 38-month-olds’ perfor-mance was affected by object similarity betweenthe two rooms.

Based on these findings, Gentner and her col-leagues (Gentner & Ratterman, 1991; Kotovsky &Gentner, 1996) argued that lower order concretesimilarity leads young children to find higher orderrelational similarity by heightening the overall simi-larity of the entities (scenes, stories, and so on) tobe compared and providing a scaffold for aligningthe representations of those entities. They furtherproposed that the repeated experience of makingrelational mappings by being guided by lowerorder concrete similarity prepares young childrento easily find relational similarity even when it isnot supported by lower order similarity. Gentnerand her colleagues called this process progressivealignment. For example, in a task that requiredchildren to find a hidden toy in a target room afterbeing shown its location in a model room, Loewen-stein and Gentner (2001) found that children whowere given an opportunity to simply compare theinitial model (Hiding Room 1) with another highlysimilar model (Hiding Room 2) became better ableto find the toy in the target room subsequently.

The process of alignment is also critical to theextraction of relational similarity across events inextending verbs. Based on the finding that lowerorder similarity helps young children identifyhigher order relational similarity in analogical rea-soning tasks (Gentner, Loewenstein, & Hung,2007; Kotovsky & Gentner, 1996), we predict thatyoung children’s verb extensions may also be fos-tered if relational similarity is highlighted bylower order similarity. Here, we consider the sim-ilarity of objects involved in action events becauseprevious studies have demonstrated that youngchildren have great difficulty in separating theaction itself from the involved object in extendingverbs (Behrend, 1990; Forbes & Farrar, 1993; Imaiet al., 2005; Imai et al., 2008). We predict that chil-dren’s verb extension will be fostered if theobjects involved in the same action are similar.We consider this process to be a variation of pro-gressive alignment proposed by Gentner and hercolleagues (Gentner et al., 2007; Kotovsky & Gent-ner,1996), and hereafter we call our predictionconcerning children’s verb extension the object-similarity bootstrapping hypothesis.

The primary purpose of the present research wasto test the object-similarity bootstrapping hypothe-sis. Two studies were conducted for this purpose.Study 1 examined whether object similarity wouldscaffold action-based verb extension. In Study 2, we

Object Similarity Scaffolds Verb Extension 675

went one step further and investigated whetherrepeated experience of action-based verb extensionsupported by object similarity would lead youngchildren to be better able to extract the same actionacross events even without support from objectsimilarity.

If we find support for the object-similarity boot-strapping hypothesis, it will provide importantinsights for theories of children’s verb learning.There has been a long debate concerning the mech-anisms by which young children learn verb mean-ings. Some researchers have argued for theimportance of domain-specific knowledge of themapping rules between argument structure andverb meanings (e.g., Fisher, Gleitman, & Gleitman,1991; Gleitman, 1990). Other researchers, in con-trast, have emphasized domain-general mecha-nisms such as analogical reasoning (e.g., Gentner,2006) and mindreading (Akhtar & Tomasello, 1996;Tomasello, 2003). If we find that object similaritynot only helps young children extend verbs inStudy 1 but also bootstraps them to action-basedverb extension even without support from objectsimilarity in Study 2, this will highlight the impor-tance of domain-general mechanisms in verb learn-ing, although it does not exclude the possibilitythat children use domain-specific knowledge aboutverbs in inferring the meanings.

Study 1

In Study 1, each child was assigned to one of twoconditions, the similar-object condition or the dis-similar-object condition. Children in both condi-tions heard a novel verb while watching a video inwhich a woman was performing a novel actionusing a novel object. They were then shown twotest scenes side by side and asked to select the oneto which the novel verb could be applied. The twotest scenes were an Action-Same ⁄ Object-Changetest event (henceforth, AS test event) in which thesame woman was performing the same action butwith a different object, and an Object-Same ⁄ Action-Change test event (OS test event) in which the samewoman was performing a different action with thesame object as that used in the original event. Ifchildren selected the AS test event, they could beregarded as having extended the novel verb basedon the sameness of the action, and if they selectedthe OS test event, they must have attended to thesameness of the object in extending the novel verb.In the similar-object condition, the object used inthe AS test event was highly similar in shape to the

object used in the original action event. In the dis-similar-object condition, the object used in the AStest event was not similar to the object used in theoriginal action. The object-similarity bootstrappinghypothesis predicted that children in the similar-object condition would extend a novel verb to theAS test event more successfully than those childrenin the dissimilar-object condition.

Method

Participants.. Twenty-eight 3-year-olds (M age =3 years 5 months, SD = 3.6 months, range = 3.0–3.11), and twenty-eight 4-year-olds (M age = 4 years5 months, SD = 2.8 months, range = 4.1–4.10) partic-ipated in this study. All participants were monolin-gual Japanese. They lived in a suburban city in thegreater Tokyo area and were mostly from middle-to lower-middle-class families. Each child wasrandomly assigned to one of the two conditions, thesimilar-object condition or the dissimilar-objectcondition. There was approximately the same num-ber of boys and girls in each condition.

Stimulus materials.. Six sets of video actionevents were used as stimulus materials (seeTable 1). Each set consisted of a standard event andtwo test events. In the standard event, a youngwoman performed a repetitive action with a novelobject. The two test events, that is, the OS test eventand the AS test event, were variants of the standardevent, as described earlier. The similar-object condi-tion and the dissimilar-object condition differedonly in the type of object used in the AS test events.In the similar-object condition, the object used inthe AS test event was highly similar in shape to theobject used in the standard event, but it could beclearly distinguished from the one that appeared inthe standard event and could be recognized as adifferent kind of object. In the dissimilar-objectcondition, the shape similarity of objects in thestandard event and the AS test event was low.

In order to verify that the shape similarity ofobjects within each set was higher in the similar-object condition than in the dissimilar-object condi-tion, 15 undergraduates were presented with thetwo objects from each set in each condition andasked to rate the shape similarity between them ona scale of one (least similar) to seven (most similar).The mean rating score for the similar-object condi-tion was 5.3 (range = 3.7–5.9), and for the dissimi-lar-object condition, it was 1.8 (range = 1.5–2.5).The scores were significantly higher in the similar-object condition than in the dissimilar-object condi-tion for all the sets.

676 Haryu, Imai, and Okada

Tab

le1

Act

ion

san

dO

bjec

tsU

sed

inS

tudy

1


Procedure.. In both the similar-object conditionand the dissimilar-object condition, the procedurewas the same as that used in Imai et al. (2005),except that the stimuli were carefully constructedso that the shape similarity between the object usedin the standard event and the object used in the AStest event would be clearly higher in the similar-object condition than in the dissimilar-object condi-tion. The children were tested individually in aquiet room in the preschool they attended. Thestimulus videos were presented using PowerPointon a laptop computer.

As in Imai et al. (2005), children received fourwarm-up trials to make sure that they were able toindicate what they thought to be the correct answerby pointing to one of the two test videos. Care wastaken so that the warm-up trials would not leadchildren to have any bias toward objects or actionsduring the test trials. In the first two warm-uptrials, children were shown two familiar objects(e.g., cat vs. dog) on videos and were asked to iden-tify the object the experimenter referred to bysaying, for example, ‘‘Where is a doggy?’’ In thethird warm-up trial, they were shown two familiaractions (hand clapping vs. jumping) and the experi-menter asked, ‘‘In which [movie] is [she] jumping?’’dropping the words in brackets since doing that isvery natural in Japanese. In the last warm-up trial,the children heard a novel verb while watching astandard video in which an actor was performing anovel action with a novel object, and then wereshown two test events side by side. One of the testevents was exactly the same as the standard video;the other was totally different in both the actionand the object, although the actor was the same.The children were asked to which movie the novelverb should be applied. Thus, the last warm-uptrial was parallel to the test trials except for thestructure of the two test events.

After the warm-up trials, the children receivedthe test trials. In each set, the standard event (a 10-sclip) was repeated three times in a continuous loopfor a total of 30 s. While showing the standard event,the experimenter introduced a novel verb by saying,‘‘Mite (Look)! X-teiru (is X-ing).’’

In this Japanese sentence, X represents a verbstem that was replaced with a novel word like neke,ruchi, or heku; it was followed by the present pro-gressive suffix -teiru. Thus, just saying ‘‘X-teiru’’ inJapanese can be translated into English as ‘‘isX-ing,’’ with subject and object dropped. Droppingsubject and object from the sentence is neitherunnatural nor ungrammatical in Japanese, and thesekinds of sentences are as appropriate for transitiveT

able

1

Con

tin

ued


verbs as for intransitive verbs. For example, in not-ing that a rabbit is eating a carrot, Japanese speakersnaturally say ‘‘tabe-teiru (is eating)’’ without men-tioning the subject (i.e., rabbit) or the object (i.e., car-rot). In fact, in a similar verb extension task inprevious research (Imai et al., 2008), Japanese pre-schoolers performed slightly better when the verbwas introduced without subject and object thanwhen the verb was introduced with subject andobject, unlike their English-speaking age peers. ForJapanese children, hearing arguments of a verb in asituation in which the subject and the object wereboth obvious from visual information in the videomight have impeded their verb generalization. Thus,novel verbs were presented without arguments inthis study, as was done in Imai and colleagues (Imaiet al., 2005; Imai et al., 2008).

After the novel verb was repeated three timeswhile the standard event was displayed, the twotest events were shown side by side. The left–rightposition of the AS and OS test events was counter-balanced within each child. The experimenterpointed to each event in succession and waited sothat the children could watch and compare the twovideos as long as they wanted. Then the experi-menter asked the children to indicate which of thetwo test events the novel verb should be extendedto by saying, ‘‘X-teiru (X-ing) nowa (topic marker)docchi (which)?’’ This can roughly be translatedinto English as ‘‘In which [movie] is [she] X-ing?’’although the words in brackets were omitted. Thisprocedure was repeated for each of the six sets ofstimuli, and the presentation order was random-ized across participants.

Results

If children can extend a novel verb based onsameness of action, they should select the AS testevent. The selection of an AS test event was scoredas an AS response and the mean proportion of ASresponses was calculated for each condition, asshown in Figure 1.

To test whether children’s responses variedacross age groups and object similarity conditions,the proportion of AS responses was submitted to a2 (age: 3- vs. 4-year-olds) · 2 (object similarity con-dition: similar vs. dissimilar) analysis of variance(ANOVA). The main effect of object similarity wassignificant, F(1, 52) = 17.4, p < .001, gp

2 = .25, indi-cating that children in the similar-object conditionwere more likely to extend a verb to the sameaction than those in the dissimilar-object condition.The main effect of age also approached statistical

significance, F(1, 52) = 4.0, p = .052, gp2 = .07,

suggesting that 4-year-olds performed better than3-year-olds. The interaction between age and objectsimilarity was not significant, F(1, 52) = .20, p = .66,gp

2 = .004. Thus, both 3- and 4-year-olds showed abenefit of object similarity in novel-verb extension,and the prediction that object similarity scaffoldsaction-based verb extension was supported.

The 4-year-olds in the similar-object conditionsuccessfully extended a novel verb based on thesameness of the action, selecting AS test events82.1% of the time, which was significantly higherthan chance, t(13) = 5.21, p < .001, d = 1.39, two-tailed. In contrast, children of the same age in thedissimilar-object condition made AS responses39.3% of the time, which was not different fromchance, t(13) = 1.01, p > .10, d = .27, two-tailed. For3-year-olds, action-based verb generalization wasstill difficult even with a scaffold from object simi-larity. Their performance in the similar-object con-dition did not differ from chance level. They choseAS test events 59.5% of the time, t(13) = .87, p > .10,d = .23, two-tailed. The 3-year-olds in the dissimi-lar-object condition selected the AS test event only25.0% of the time, which was significantly belowchance, t(13) = 2.94, p < .05, d = .79, two-tailed.

An analysis of individual children’s performanceconverged with the results of the group analysis.Table 2 shows the number of AS-dominant childrenwho selected AS test events four times or more outof six trials, and the number of OS-dominant chil-dren who selected OS test events four times ormore. An asymmetrical log-linear model was fittedon the 2 (age) · 2 (object similarity condition) con-tingency table to see whether the ratio of AS-domi-nant children to OS-dominant children variedacross age groups and object similarity conditions.In both age groups, the proportion of AS-dominantchildren was greater in the similar-object condi-tion than in the dissimilar-object condition, v2(1,

0

10

20

30

40

50

60

70

80

90

100

3-Year-Olds 4-Year-Olds

% A

ctio

n-Sa

me

Res

pons

e

Similar-ObjectDisimilar-Object

Figure 1. Proportion of Action-Same responses in Study 1.


N = 56) = 13.26, p < .001, u = .49. Neither the maineffect of age, v2(1, N = 56) = 2.35, p > .10, u = .20,nor the interaction between age and object similar-ity condition, v2(1, N = 56) = .01, p > .10, u = .01,reached statistical significance.

Discussion

The results supported the object-similarity boot-strapping hypothesis. Children were more likely toextend a novel verb to the AS test event when theobject used in the AS test event was highly similarto the object used in the standard event. When thetwo objects used in the same action events (thestandard and the AS test) were similar to eachother, the overall similarity across the two eventsincreased, which helped children align the twoevents and extract the common relation.

In contrast, children were much less willing toextend a novel verb to the same action when theobjects used in the standard and AS test eventswere dissimilar. Three-year-olds in the dissimilar-object condition were especially prone to selectingthe test scene in which a different action was per-formed with the same object, doing so 75% of thetime. This result, however, should be interpretedwith caution, since there are three possible explana-tions for it. One possibility is that 3-year-olds maybelieve that a verb refers to an object. The secondpossibility is that they believe that a verb can beextended to any action performed with the sameobject. The third possibility is that 3-year-olds mayhave understood that a verb should be generalizedbased on the sameness of action, but they selectedthe OS test event because they were unable to findthe same action in the AS test event and resorted tothe sameness of objects.

In order to examine these possibilities, Study 1Awas conducted, using a yes-no paradigm with the

stimuli used in the dissimilar-object condition inStudy 1. To see if 3-year-olds think that a verbrefers to an object, a Still-Object test event (hence-forth, SO test event) was included in addition tothe AS and OS test events. In the SO test event, theobject used in the standard event was shown in avideo sitting alone on a table. Three-year-olds werefirst introduced to a novel verb while watching astandard event. One of the three test events wasthen presented to the side of the standard event.Children were asked whether the test event couldbe labeled with the verb.

If 3-year-olds’ performance in the dissimilar-object condition in Study 1 reflected their beliefthat a verb refers to an object, they will extendthe novel verb to the SO test event. If they believethat a verb refers to any action performed withthe same object, they will be willing to apply thenovel verb to the OS test event, but will acceptneither the SO test event nor the AS test event asa referent of the verb. Finally, if they know that averb should be extended based on sameness ofaction but have trouble extracting the same actionperformed with a different object, they will beunwilling to extend the novel verb to any of thethree test events.

Study 1AMethod

Participants.. Fifteen monolingual Japanese3-year-olds (M age = 3 years 6 months, SD =4.1 months, range = 3.0–3.10) took part in this study.There were 6 girls and 9 boys. All children lived in asuburban city in the greater Tokyo area. Their demo-graphic characteristics were the same as those of thechildren who participated in Study 1. Four addi-tional children were excluded due to failure to com-plete the experiment (1) or due to a yes bias, saying‘‘yes’’ to all the questions (3). None of the childrenhad participated in Study 1.

Stimuli and procedure.. The six sets of videostimuli used in the dissimilar-object condition inStudy 1 were used. In addition to the standardevent, the AS test event and the OS test event, avideo of an SO test event was included in each set.In the SO test event, the object used in the stan-dard event was shown in a video lying alone on atable. As in Study 1, the stimulus videos were pre-sented using PowerPoint on a laptop computer.The children were shown a standard event andintroduced to a novel verb by saying, ‘‘Mite(Look)! X-teiru (X-ing),’’ dropping subject and

Table 2

Number of Children Who Showed AS-Dominant and OS-Dominant

Performance in Study 1

AS-dominant

children

OS-dominant

children

Three-year-olds

Similar object 8 6

Dissimilar object 2 12

Four-year-olds

Similar object 11 3

Dissimilar object 4 10

Note. AS = Action-Same; OS = Object-Same.


object as in Study 1. Then one of the three testevents was shown on the right side of the standardevent. The experimenter pointed to the test eventand asked, ‘‘Kocchi mo X-teiru?’’ This sentence canbe translated into English as ‘‘Is [the woman] alsoX-ing [it] in this [movie]?’’ Again, the words inbrackets were not uttered in the Japanese sentence.The children saw each of the six standard eventsthree times with three different test events for atotal of 18 trials. Six verbs were used in this experi-ment, and a particular verb was always associatedwith a particular standard event. The order of the18 trials was randomized with the constraint thatno successive trials presented the same standardevent.

Results

If children agreed with the extension of the novelverb to the test event, the response was scored as ayes response. The proportion of yes responses toeach test event is shown in Figure 2. A repeatedmeasures ANOVA revealed a significant maineffect of test type, F(2, 28) = 4.33, p < .05, g2 = .24.Least significance difference post hoc comparisonsrevealed that the proportion of yes responses wasgreater for the AS trials than for the SO trials, andgreater for the AS trials than for the OS trials, allps < .05.

The 3-year-olds extended the novel verb to theSO test event only 17.8% of the time, which wassignificantly below chance, t(14) = 3.85, p < .01,d = .99, two-tailed. They were also reluctant togeneralize the novel verb to a different actionperformed with the same object. They accepted theOS test event as a referent of the verb 16.7% of thetime, which was again significantly less thanchance, t(14) = 5.29, p < .001, d = 1.37, two-tailed.The children were more likely to extend a novelverb to the same action performed with a different

object, but the proportion of yes responses (43.3%)did not exceed chance level, t(14) = .70, p > .10,d = .18, two-tailed.

The analysis of individual children’s responseswas consistent with the results from the groupanalysis. Three of the 15 children accepted the SOtest event as a referent for the novel verb four timesor more in the six trials, which was significantlybelow chance (sign test, p < .05). Only 1 of the 15three-year-olds extended the novel verb to the OStest event four times or more in the six trials, whichwas also significantly different from chance,p < .01. The number of children who extended thenovel verb to the AS test event four times or morein the six trials was 5, which was not different fromchance.

Discussion

The results suggest that 3-year-olds believe nei-ther that a verb refers to an object nor that a verbrefers to any action performed with the sameobject, even though they selected the OS testevent significantly more often than chance level inthe dissimilar-object condition in Study 1. Theirunwillingness to extend a novel verb to the sameaction performed with a different, perceptuallydissimilar object in Study 1 should therefore beattributed to their difficulty in extracting the com-mon action.

The results from Study 1 indicate that youngchildren’s success in verb extension is stronglyaffected by the similarity of the objects used in per-forming the same action. This is presumablybecause object similarity enhances the overall simi-larity between the standard event and the AS testevent, and this helps the children compare the twoevents and extract the common relation (i.e.,action). As reviewed in the Introduction, a parallelphenomenon has been reported in the literature onchildren’s analogical reasoning (DeLoache, 1989,1990; Gentner & Toupin, 1986; Holyoak et al., 1984;Kotovsky & Gentner, 1996) as well as in previousresearch on word learning (Childers, 2008; Childers& Paik, 2009; Klibanoff & Waxman, 2000; Waxman& Klibanoff, 2000). Given the accumulating evi-dence for the scaffolding role of object similarity indeveloping analogical reasoning as well as in wordlearning, a further interesting possibility is thatrepeated experience of extending a novel verb tothe same action guided by object similarity maybootstrap young children to successful verb exten-sion in situations in which there is no support fromobject similarity, as predicted by the progressive

0102030405060708090

100

Still-Object Object-Same Action-Same

% Y

es R

espo

nses

Figure 2. Proportion of children’s agreement to extend the novelverb to each test event in Study 1A.


alignment hypothesis (Gentner et al., 2007; Kotov-sky & Gentner, 1996). This possibility was exploredin Study 2.

Study 2

We investigated whether repeated experience ofaction-based verb extension supported by objectsimilarity would lead children to be better able toextract commonality of action across events evenwithout support from object similarity. For thispurpose, 4-year-olds were assigned to either a simi-lar-then-dissimilar-object (henceforth, SD) conditionor a dissimilar-then-dissimilar-object (DD) condi-tion. The children in the SD condition were firstexamined on their novel-verb extensions with sup-port from object similarity on the first four trials.The last four trials tested whether they couldextend a novel verb to the same action without sup-port from object similarity. The children in the DDcondition were asked to extend a novel verb with-out support from object similarity both on the firstfour trials and on the last four trials. If the object-similarity bootstrapping hypothesis is valid, thechildren in the SD condition should perform betterthan those children in the DD condition, not onlyon the first four trials, where the two groups aretested using different types of stimuli, but also onthe last four dissimilar-object trials, where the twogroups receive the same stimuli.

Method

Participants.. Forty-seven monolingual Japanese4-year-olds participated in this study (M age =4 years 6 months, SD = 3.2 months, range = 4.0–4.11). They lived in a suburban city in the greaterTokyo area and were mostly from middle- tolower-middle-class families. None of these chil-dren took part in Study 1 or 1A. Twenty-fourchildren were assigned to the SD condition, and23 were assigned to the DD condition. There wereapproximately equal numbers of boys and girls ineach condition.

Stimulus materials.. Eight sets of video actionevents served as stimulus materials. As in Study 1,each set consisted of a standard event and two testevents, the OS test event and the AS test event. Inthe standard event, a woman performed a novelaction with a novel object. In the OS test event, thesame woman performed a different action usingthe same object as in the standard event. In the AStest event, the same woman performed the same

action as in the standard event, but with a differentobject.

On the first four trials, children in the SD andDD condition heard the same novel verb in associa-tion with the same standard event. They were thenpresented with the two test events, as in Study 1.For the first four trials, the OS test event was thesame across the two conditions, but the AS testevent depended on the condition the children wereassigned to. The children in the SD condition sawthe similar-object AS test events that were used forthe similar-object condition in Study 1, and the chil-dren in the DD condition were shown the dissimi-lar-object AS test events that were used for thedissimilar-object condition in Study 1. Thus, Sets 1,2, 4, and 6 in Table 1 were used on the first fourtrials in Study 2.

On the last four trials, children across the twoconditions saw the same four sets of stimuli; eachset consisted of a standard event, an OS test event,and a dissimilar-object AS test event. The two newsets of stimuli were used in addition to the two dis-similar-object sets from Sets 3 and 5 shown inTable 1. The two new sets were modeled after thedissimilar-object sets in Study 1.

Procedure.. Children were tested individually ina quiet room at the preschool they attended. Chil-dren in both the SD and DD conditions received atotal of eight trials. Throughout the session, theinstructions and the procedure for each trial wereidentical to those in Study 1 for both conditions.

As in Study 1, warm-up trials were given priorto the experimental trials. The experimental trialsconsisted of two blocks, the first four trials and thelast four trials. On each of the first four trials, chil-dren were first introduced to a novel verb whilewatching a standard event. They were then shownthe two test events, the AS test event and the OStest event. The type of AS test event they sawdepended on the condition they were assigned to.The children in the SD condition were shown thesimilar-object AS test event, while the children inthe DD condition saw the dissimilar-object AS testevent. In both conditions, children were asked toindicate to which of the AS and OS test events thenovel verb should be applied. This procedure wasrepeated on each of the first four trials.

The last four trials were conducted without abreak after the first four trials were completed. Onthe last four trials, children in both conditionsreceived the same stimuli. That is, they were firstshown the standard event and then tested with theOS test event and the dissimilar-object AS test event.Any set that was used on the last four trials was not


used on the first four trials. The order of the foursets within the first four trials and the last four trialswas randomized and the left–right position of theAS and OS test events was counterbalanced withineach child. No feedback was provided concerningwhether the response was correct or not on the firstfour trials or on the last four trials.

Results

A choice of the AS test event was scored as anAS response. The proportion of AS responses wascalculated for the first four trials and the last fourtrials separately in each condition, as shown inFigure 3. The children in the DD condition selectedAS test events 51.1% of the time on the first fourtrials, and 55.4% of the time on the last four trials,neither of which was significantly different fromchance, t(22) = .13, p > .10, d = .03, and t(22) = .54,p > .10, d = .11, respectively. In contrast, the chil-dren in the SD condition selected AS test events86.5% of the time on the first four trials, which wassignificantly greater than chance, t(23) = 7.67,p < .001, d = 1.57. Unlike the children in the DDcondition, those in the SD condition also showedaction-based verb extension for the last four trials,selecting the AS test events 75.0% of the time,t(23) = 3.20, p < .01, d = .65.

The proportion of AS responses was submittedto a 2 (condition: SD vs. DD) · 2 (block: first fourtrials vs. last four trials) ANOVA with the factor ofblock as a repeated measure. The main effect ofcondition was significant, F(1, 45) = 7.72, p < .001,gp

2 = .15. Neither the main effect of block,F(1, 45) = 0.47, p > .10, gp

2 = .01, nor the interac-tion, F(1, 45) = 2.32, p > .10, gp

2 = .05, was signifi-cant. Thus, the 4-year-olds in the SD conditionperformed better than those in the DD conditionnot only on the first four trials but also on the lastfour trials, in which there was no support from

object similarity. The lack of an interaction effectsuggests that the performance of the children in theSD condition did not drop with the change fromthe similar-object trials to the dissimilar-object tri-als, whereas the performance of the children in theDD condition did not improve from the first fourdissimilar-object trials to the last four dissimilar-object trials.

Discussion

The results from the first four trials replicatedthe findings from Study 1. Specifically, 4-year-oldswere able to extend a novel verb based on thesameness of action when the same action was per-formed with a similar object, whereas they failedwithout support from object similarity. Moreimportant, however, is that the children who expe-rienced similar-object trials on the first four trialsalso performed very well on the following dissimi-lar-object trials. A general practice effect could notexplain the success on the last four trials for thechildren in the SD condition, since the performanceof the children in the DD condition did notimprove after their four dissimilar-object trials. Itthus seems legitimate to conclude that the differ-ence in performance on the last four dissimilar-object trials came from the difference in the chil-dren’s experience on the first four trials. Therepeated experience of action-based verb extensionsupported by object similarity bootstraps childrento be better able to generalize a novel verb basedon the sameness of action even without supportfrom object similarity.

Object similarity is also reported to be importantin young children’s adjective extension. Klibanoffand Waxman (2000) found that 3-year-olds couldextend a novel adjective successfully from oneobject (e.g., a bumpy green horse) to another (e.g., abumpy purple horse) if and only if both objectswere drawn from the same basic-level object cate-gory. In addition, 3-year-olds who had been givenan opportunity to extend a novel adjective withinthe same basic-level category became able to extendthe same adjective to other objects from differentbasic-level categories. Once children successfullyextend a particular adjective supported by objectsimilarity, this gives them another chance to dis-cover precisely what property the adjective refersto, which would lead them to generalize the adjec-tive to objects that are less similar to the original.The present study went one step further, however,demonstrating that children come to have generalexpertise in finding common relations acrossFigure 3. Proportion of Action-Same responses in Study 2.


events after they have extended several novel verbsbased on sameness of action supported by objectsimilarity.

General Discussion

Importance of Domain-General Cognitive Processes forVerb Learning

Young children were found to have great diffi-culty in generalizing a novel verb to the sameaction performed with a different, perceptually dis-similar object, which was consistent with the find-ings from previous research (Forbes & Farrar, 1993;Imai et al., 2005; Imai et al., 2008; Kersten & Smith,2002; Maguire et al., 2002). However, this does notmean that young children do not know that verbsshould be generalized by a different principle fromnouns. In Study 1A, 3-year-olds rejected an objectas a referent of a novel verb. They do understandthat verbs refer to relations between objects ratherthan objects per se.

Then, in what situations can young childrenextend a verb based on sameness of action? InStudy 1, both 3- and 4-year-olds could extend anovel verb to the same action more easily when theobjects involved in the original event and the same-action event were perceptually similar to each otherthan when they were dissimilar, which suggeststhat object similarity plays a scaffolding role in verblearning. Study 2 further demonstrated that chil-dren became better able to extend a novel verb tothe same action even without support from objectsimilarity after they had had repeated experience ofmaking action-based verb extension with supportfrom object similarity.

These results provide an insight into the mecha-nism by which young children learn verb mean-ings. Initially, young children may be veryconservative in generalizing verbs. They limitthemselves to extending a novel verb to events thatare massively similar to the event they originallyexperienced. Here object similarity can serve as ascaffold, by heightening the overall similarityacross events in which the same action is per-formed. Once children have chosen the same-actionevent guided by overall similarity, this in turn pro-vides them with an opportunity to compare theevents in more detail and to extract the commonaction. Repeated experience of extending novelverbs supported by object similarity then bootstrapschildren to action-based verb extension withoutscaffolding from object similarity. Thus, the resultsof the present research highlight importance of the

domain-general processes such as comparison andalignment (Gentner, 1982) in the domain of verblearning.

Optimal Levels of Object Similarity for Comparison andProgressive Alignment

An interesting question is the kind and degree ofobject similarity that are needed to foster progres-sive alignment in verb (as well as adjective) learn-ing. The object similarity that promoted adjectivelearning in Klibanoff and Waxman (2000) was thesimilarity that arose from the same basic-level cate-gory. Kersten and Smith (2002) also reported that3-year-olds could extend a novel verb to the sameaction performed by a different agent from thesame category. In contrast, object similarity in thepresent research was purely perceptual, since allthe similar objects were of different kinds.

On one hand, it is possible that children receivestronger scaffolding when object similarity is bothconceptual and perceptual than when it is only per-ceptual. Thus, we might expect that children mayhave performed even better if the objects from thesame basic-level category were used for the stan-dard action event and the AS test event. On theother hand, it is also possible that if the objectsused for the standard-action event and the AS testevent were too similar, the children may havethought that the two action events were identical,and hence may not have initiated comparisonbetween the two events. If so, progressive align-ment would not take place. In fact, Childers andPaik (2009) found that children who were shownseveral dissimilar events associated with a particu-lar verb became better able to extend the verb toother dissimilar events than those children whowere provided with the same number of similarevents in association with the verb. It may be whenthere are both sufficient similarity and variationacross events denoted by the same verb that chil-dren initiate comparison and alignment. It is alsoprobable that the optimal level of similarity inter-acts with the size of children’s verb lexicons (seeMaguire, Hirsh-Pasek, Golinkoff, & Brandone, 2008,for the possibility that very young children benefitmore from the repetition of one exemplar ratherthan from many different exemplars). This issue isworth pursuing in future research.

Nature of Children’s Verb Learning

The present research has demonstrated thatyoung children’s verb extension is fostered by


object similarity. At the same time, it suggests thatchildren as old as 4 years of age still need scaffold-ing from object similarity in verb extension. Thisfact is striking, since Waxman and her colleagues(Waxman, Lidz, Braun, & Lavin, 2009) havereported that 24-month-olds are ready to extend anovel verb to a different event involving the sameaction. However, in our view, their findings andours should not be taken as contrasting.

In Waxman et al.’s (2009) research, 24-month-olds received four different examples of the sameaction (e.g., in each example, the same actor waswaving one of four different balloons) in associa-tion with a novel verb in the familiarization phase.In addition, the children were shown a completelydifferent action (e.g., the same actor was playing asaxophone) and told that the verb should not beapplied to the action. With such rich scaffolding,24-month-olds could extend a novel verb to a verysimilar event in which the same actor was perform-ing the same action with another object from thesame basic-level object category (e.g., a new balloonthat the child did not see in the familiarizationphase). Thus, Waxman et al.’s results suggest that24-month-olds need rich scaffolding to extend anovel verb to a very similar event. The fact thatchildren at 3 and 4 years of age still need somescaffolding in verb extension indicates that theirunderstanding of verb meanings is not as robust asadults, and has room for developmental progress.

Even when children understand that verbsshould be generalized based on the sameness ofrelation, finding the ‘‘same’’ relation is not easy.For example, adult English speakers may think thatthe English verb carry refers to fairly concreteactions and its meaning is easy to grasp. However,when we consider the range of events that could bedenoted as ‘‘carrying’’ actions, it becomes obviousthat extracting commonality across ‘‘carrying’’actions is not simple. For example, a scene of some-one ‘‘carrying’’ a briefcase in the hand is distinc-tively different from a scene of someone ‘‘carrying’’a tennis racket on the shoulder or ‘‘carrying’’ abackpack on the back. For English speakers, theseare all the same action to which the verb carry isapplied, although they look so different. In Chinese,however, these actions are denoted by differentverbs. Thus, the semantic invariant for each verbcan be very abstract, even though each referentaction seems to be concrete.

Furthermore, what counts as a semantic invari-ant varies depending on the semantic domain ofthe verb. The current experiments focused on arepetitive action performed with some object, such

as ‘‘waving a flag’’ or ‘‘twirling a baton.’’ Otherverbs, however, can be extended based on thesameness of the result of the action, or based onthe sameness of the manner of motion. Given thevariety of relations to which verbs can refer, it isnot sufficient for children to understand thatverbs refer to kinds of relations. Children alsoneed to find out what kind of common relationshould be preserved as a core meaning of a verbin the particular domain. In this regard, weshould be cautious about concluding that childrenhave abstract understanding of verb meaningseven if they show some level of understanding.Even when children are able to extend a novelverb to the same action without scaffolding, theymay still have a long way to go before achievingadult-like comprehension of verb meaning (Sajiet al., 2008; Theakston, Lieven, Pine, & Rowland,2002).

In any case, we emphasize that the resultspresented in the current research should not betaken to suggest that young children lack the abilityto extend a novel verb (see Imai et al., 2005, for asimilar discussion). Instead, they suggest that ittakes children a long time to gain the robust, adult-like representation of verb meanings that allowsthem to successfully extend a novel verb evenwhen scaffolding is scant. Our research offers anaccount of how children go through this longdevelopmental trajectory, and of how they build upexpertise in verb learning that requires less and lessscaffolding.

References

Akhtar, N., & Tomasello, M. (1996). Two-year-olds learnwords for absent objects and actions. British Journal ofDevelopmental Psychology, 14, 79–93.

Behrend, D. A. (1990). The development of verb concepts:Children’s use of verbs to label familiar and novelevents. Child Development, 61, 681–696.

Billow, R. M. (1975). A cognitive development study ofmetaphor comprehension. Developmental Psychology, 11,415–423.

Childers, J. B. (2008). The structural alignment and com-parison of events in verb acquisition. Proceedings of the30th Annual Meeting of the Cognitive Science Society, 681–686.

Childers, J. B., & Paik, J. H. (2009). Korean- and English-speaking children use cross-situational information tolearn novel predicate terms. Journal of Child Language,36, 201–224.

Childers, J. B., & Tomasello, M. (2002). Two-year-oldslearn novel nouns, verbs and conventional actions from


massed or spaced exposures. Developmental Psychology,38, 967–978.

Childers, J. B., & Tomasello, M. (2006). Are nouns easierto learn than verbs? Three experimental studies. In K.Hirsh-Pasek & R. M. Golinkoff (Eds.), Action meetswords (pp. 311–335). New York: Oxford UniversityPress.

DeLoache, J. S. (1989). The development of representationin young children. In H. W. Reese (Ed.), Advances inchild development and behavior (Vol. 22, pp. 1–39). NewYork: Academic Press.

DeLoache, J. S. (1990). Young children’s understanding ofmodels. In R. Fivush & J. Hudson (Eds.), What childrenremember and why (pp. 94–126). Cambridge, UK: Cam-bridge University Press.

Fisher, C., Gleitman, H., & Gleitman, L. (1991). On thesemantic content of subcategorization frames. CognitivePsychology, 23, 331–392.

Forbes, J. N., & Farrar, J. (1993). Children’s initialassumptions about the meaning of novel motion verbs:Biased and conservative? Cognitive Development, 8, 273–290.

Gentner, D. (1982). Why nouns are learned before verbs:Linguistic relativity versus natural partitioning. In S. A.Kuczaj (Ed.), Language development: Vol. 2. Language,thought and culture (pp. 301–334). Hillsdale, NJ: Erl-baum.

Gentner, D. (1988). Metaphor as structure mapping: Therelational shift. Child Development, 59, 47–59.

Gentner, D. (2006). Why verbs are hard to learn. In K.Hirsh-Pasek & R. M. Golinkoff (Eds.), Action meetswords (pp. 544–564). New York: Oxford UniversityPress.

Gentner, D., Loewenstein, J., & Hung, B. (2007). Compari-son facilitates children’s learning of names of parts.Journal of Cognition and Development, 8, 285–307.

Gentner, D., & Ratterman, M. J. (1991). Language and thecareer of similarity. In S. A. Gelman & J. P. Byrnes(Eds.), Perspectives on language and thought: Interrelationsin development (pp. 225–277). New York: CambridgeUniversity Press.

Gentner, D., & Toupin, C. (1986). Systematicity and sur-face similarity in the development of analogy. CognitiveScience, 10, 277–300.

Gleitman, L. L. (1990). The structural sources of verbmeanings. Language Acquisition, 1, 3–55.

Hirsh-Pasek, K., & Golinkoff, R. M. (2006). Action meetsword: How children learn verbs. New York: Oxford Uni-versity Press.

Holyoak, K. J., Junn, E. N., & Billman, D. O. (1984).Development of analogical problem-solving skill. ChildDevelopment, 55, 2042–2055.

Imai, M., Haryu, E., & Okada, H. (2005). Mapping novelnouns and verbs onto dynamic action events; Are verbmeanings easier to learn than noun meanings for Japa-nese children? Child Development, 76, 340–355.

Imai, M., Li, L., Haryu, E., Okada, H., Hirsh-Pasek, K.,Golinkoff, R. M., et al. (2008). Novel noun and verblearning in Chinese-, English, and Japanese-speakingchildren. Child Development, 79, 979–1000.

Kersten, A. W., & Smith, L. B. (2002). Attention to novelobjects during verb learning. Child Development, 73, 93–109.

Klibanoff, R. S., & Waxman, S. R. (2000). Basic level objectcategories support the acquisition of novel adjectives:Evidence from preschool-aged children. Child Develop-ment, 71, 649–659.

Kotovsky, L., & Gentner, D. (1996). Comparison and cate-gorization in the development of relational similarity.Child Development, 67, 2797–2822.

Loewenstein, J., & Gentner, D. (2001). Spatial mapping inpreschoolers: Close comparisons facilitate far map-pings. Journal of Cognition and Development, 2, 189–220.

Loewenstein, J., & Gentner, D. (2005). Relational languageand the development of relational mapping. CognitivePsychology, 50, 315–353.

Maguire, M. J., Hennon, E., Hirsh-Pasek, K., Golinkoff, R.M., Slutzky, C., & Sootsman, J. (2002). Mapping wordsto actions and events: How do 18-month-olds learn averb? In B. Skarabela, S. Fish, & A. H. Do (Eds.), Pro-ceedings of the 27th annual Boston University conference onlanguage development (pp. 371–382). Somerville, MA:Cascadilla Press.

Maguire, M. J., Hirsh-Pasek, K., Golinkoff, R. M., & Bran-done, A. C. (2008). Focusing on the relation: Fewerexemplars facilitate children’s initial verb learning andextension. Developmental Science, 11, 628–634.

Saji, N., Saalbach, H., Imai, M., Zhang, Y., Shu, H., &Okada, H. (2008). Fast-mapping and reorganization:Developmenht of verb meanings as a system. Proceed-ings of the 30th annual meeting of the Cognitive ScienceSociety, 35–40.

Theakston, A. L., Lieven, E. V. M., Pine, J. M., & Row-land, C. F. (2002). Going, going, gone: The acquisitionof the verb ‘‘go.’’ Journal of Child Language, 29, 783–811.

Tomasello, M. (2003). Constructing a language: A usage-based theory of language acquisition. Cambridge, MA:Harvard University Press.

Waxman, S. R., & Klibanoff, R. S. (2000). The role of com-parison in the extension of novel adjectives. Develop-mental Psychology, 36, 571–581.

Waxman, S. R., Lidz, J. L., Braun, I. E., & Lavin, T. (2009).Twenty four-month-old infants’ interpretations of novelverbs and nouns in dynamic scenes. Cognitive Psychol-ogy, 59, 67–95.


Documents

Object Similarity Bootstraps Young Children to …...went one step further and investigated whether repeated experience of action-based verb extension supported by object similarity