{- i-{ ;\ l}'l- Ii R 3 3

Language and Cognitionin Development

SteIIa Christie and Dedre Gentner

lntroduction

The hypothesis that language can influencethought - generally known as the Whorfianhypothesis - has inspired strong opinionsin both directions. Early enthusiasm in ther95os and r96os was followed by decadesof disregard or worse. Now the wheel hasturned again, and the question of whetherand how language might influence cognitionis openly tested and debated. [See Gentnerand Goldin-Meadow, zoo3 and Gumperzand Levinson, 1996 for discussions of theforces behind this evolution.) However,work on language and thought remainsextremely contentious, and many of theclaims reviewed here are under challenge.

In some important ways, the field of cog-nitive development has been relatively opento the idea that language influences thought.Within adult cognitive psychology, the lnn-guage and thought hypothesis is associatedprimarily with Benjamin Lee Whorf and hismento4, Edward Sapir (e.g., Whor{ 1956),who proposed that specific properties of alanguage's grammar and lexicon could influ-ence cognition in speakers of that language:

"We dissect nature along lines laid down byour native language," fWhor{ ry56, p. n7).But in developmental theory the figuremost associated with the view that languageinfluences thought is Russian psycholo-gist Lev Vygotsky $962). Unlike Piaget, forwhom conceptual development proceededvia interactions with experiencg with lan-guage serving only as a means of communi-cation, Vygotsky saw language and cultureas critical to the development of thought.One could say that Piaget viewed the childas a tiny scientist, whereas Vygotsky viewedthe child as a cultural apprentice.

Vygotsky's view on the effects of lan-guage differed from that of Whorf andSapir. Rather than focusing on the effect$'of speaking one language versus another,Vygotsky theorized about the effects of lan-guage as such. He proposed that the inter-nalization of language provides childrenwith the means to direct their own thought:to achieve focused attention and will, and ameans of introspecting about one's own cog-nition. Vygotsky offered a yiew of languageas providing scaffolding for new concepts;he suggested that hearing a new term or a

6st

&KULVWLH��6���*HQWQHU��'����������/DQJXDJH�DQG�FRJQLWLRQ�LQ�GHYHORSPHQW������,Q��0��0��6SLYH\��.��0F5DH���0��-RDQLVVH��(GV���7KH�&DPEULGJH�+DQGERRN�����RI�3V\FKROLQJXLVWLFV��&DPEULGJH��&DPEULGJH�8QLYHUVLW\�3UHVV��SS���������

6s+ GENTNER AND CHRISTIE

new assertion, even when the child has poorinitial understanding, might both invite andguide the chlld to future learning.

Advances in cognitive psychology, lin-guistics, and linguistic anthropology haveled to progress on some classic questions.For example, a perennial question in thedevelopment of language and thought iswhich comes first - the concept or the lin-guistic term. Scholars like Bowerman (r98r;r9B9J have long raised challenges to thestanding Zeitgeist that concepts come first,with language merely naming them. Butuntil recently there was no way to addressthis directly. With the recent explosion oftechniques for studying infant cognition, itis becoming possible to address the ques-tion of whether and how prelinguistic cog-nition differs from postlinguistic cognition.As another example, comparative stud-ies of apes and young humans are anothernew source of insight. But this recent pro-gress has also made it clear that the ques-tion of whether language influences thoughtneeds to be decomposed into more specificquestions.

A set of fine-grained questions hasemerged concerning when and how lan-guage effects might occur [Gentner andGoldin-Meadow, 2oo3; Gumperz andLevinson, 1996; Wolff and Holmes, zou).Taking linguistic determinism - the hypoth-esis that the language we speak determineshow we perceive the world - as the start-ing point, one way to delimit the hypothesisis to specify whm we should expect to seeeffects of language on cognition. This wasthe move that Slobin (1996; zoo3) made inhis influential thinking for speahing hypoth-esis, which holds that language influencesthought only when language is activelyused. The initial statement was couched interms of "when actually speaking," but laterresearch has broadened the scope to includecomprehending language and perhapseven using language internally (though thismove makes the thinking-for-speaking viewharder to distinguish from linguistic deter-minismJ. Another way to delimit the ef[ectsof language is to assume that language aug-ments, but does not replacg other ways of

construing the world. This is the route takenby the Language as tool bit view - that acquir-ing a language provides new representationalresources - including new relational schemasas well as new categories - that augmentour capacity for encoding and reasoning[Gentne4 2oo], 2oro; Gentner and Christie,zoro; Loewenstein and Gentneq, 2oo5; seeFrank et al., zoo9, for a similar view). Theselinguistically inspired representations maybecome habitual, so that they are readilyaccessible even without the internal use oflanguage fHunt and Agnoli, r99t; Levinson,et al., zooz; Lucy, rg9 4) . In this view, lan guageprovides tools that facilitate forming andusing particular representations - represen-tations that may be intellectually potent -but does not replace all other encodingformats. Another distinction that needs tobe made is that between effects of languageon thought, and effects of language on lan-guage [Gleitman and Papafragou, zoo5). Forexample, the difference between the nam-ing patterns of English versus Japanese chil-dren for substances and objects, as in Imaiand Gentner's IISSZ) work, does not entitlethe conclusion that Japanese and Englishspeakers think about objects and substancesdifferently, though it does set the stage forfurther investigation [e.g., Imai and Mazuka,zoo1).

The distinctions among these views makeit clear that the language and thought issueis far subtler than the extreme version of theWhorfian hypothesis: that language acts todetermine our perception of the world - asa kind of permanent lens on the mind's eye.

From this perspective, the developmentalcourse oflanguage effects is ofcentral inter-est. In this paper we review evidence thatlanguage acts during cognitive developmentto promote certain kinds of conceptual struc-tures. In terms of the toolkit hypothesis, thequestion we ask is when and how languagefacilitates the acquisition of cognitive tools.Our discussion is organized around threedomains of active research: space, number,and theory of mind. There are other arenasthat could be discussed, including temporalrelations [e.g., Boroditsky, zoor), noun learn-ing and object individuation (e.g., Lupyan

LANGUAGE AND COGNITION IN DEVELOPMENT 6ss

et al., zooT; Xu, zooz), and object categori-zatron [e.g., Markman, 1989; Waxman, zooz;Waxman and Markow, t995), but given thelimitations of space we prefer to go deeplyinto a few areas.

r Space

The domain of space is an obvious placeto investigate the question of whether lan-guage affects cognition, for several reasons.First, spatial information is universally avail-able to humans regardless of where theylive. Indeed, we are all constrained by thesame universal laws of physics: An appleplaced in a bowl always rests on the bot-tom of the bowl rather than floating in midair within the bowl. Second, it is compara-tively easy to construct nonlinguistic situa-tions and events to which people can reactregardless of their language fHere spatialconcepts contrast with, say, counterfactualreasoning or concepts of justice.) Third,spatial knowledge is of fundamental impor-tance in human reasoning, both directly inactivities like navigation and manipulation,and tlrrough spatial analogies and meta-phors, which occur commonly in humanlanguage and thought. Finally, a prerequi-site for finding language-driven variabilityin concepts is variability in the language,and here space is an ideal domain. Recentresearch has revealed an astonishing varietyof ways in which languages have categorizedspatial configurations [e.g., Bohnemeyer andBrown, zooT; Bowerman, 1989, 1996; Brown,1994; Levinson and Brown, 1994; Casad andLangackel, 1985; Talmy, ry75, ry85). For allthese reasons, space has become an espe-cially active arena of investigation forlanguage and thought. The relational infor-mation that constitutes a given spatial con-figuration can be partitioned differentlyby different minds and different languages[Gentner, rg8z). For example, one can thinkof an apple in a bowl as being supported bythe bottom of the bowl, or as being con-tained inside the bowl, or as in loose contactwith the bowl. A11 of these possibilities arereflected across human languages.

tr Ftamcs of rcfercnre

Spatial relational terms provide framingstructures for the encoding of events andexperience. Across a range of languages,Levinson and his colleagues have identifiedthree spatial frames of reference that speak-ers use to describe the location of an object(Levinson, 1996; Levinson et al., zooz). Therelative (or egocentric) frame describes loca-tions relative to the speakeq, as in "the chairis left of the table." The intrinsic (or obiect-centered) frame describes locations relativeto a landmark object, as in "the chair is infront of the fireplace." Finally the absolute(or geocentic) frame describes locations rel-ative to a global frame, as in "the chair is inthe northwest corner." The term allocentricrefers to both object-centered and geocen-tric frames, in contrast to egocentric frames.Languages may use more than one of theseframes, but in many cases one frame is dom-inant. In particulal, when discussing close-range locations, the egocentric frame isdominant in English, Dutch, and German,while the geocentric frame predominatesin many other languages, including Tzeltal(MexicoJ, Arrernte fAustralia), and Haillom(Namibia) [see Majid et al., zoo4 for a com-prehensive review).

The question here is whether the habit-ual use of a particular linguistic frame ofreference has any more general effect onspatial cognition. Research by Levinson andcolleagues (Levinson, tgg6, zoo3; Levinsonand Brown, ry94; Levinson et a7., zooz;Pederson, rg95J suggests that the answer isyes; they find that people are influenced bytheir language's dominant frame of refer-ence even when carrying out a nonlinguis-tic spatial task, such as copying a scene brtracing a path through a maze fMajid et al.,zoo4;butsee Li and Gleitman, zooz for a dis-senting view).

How do such effects arise in cognitivedevelopment? Do we begin life with natu-ral proclivities or instead as "blank slates" onwhich language, culture, and other experi-ences impose spatial framesT This questionis diffrcult to answer for a topic like frameof reference, because very young infants are

6s6 GENTNER AND CHRISTIE

HIDING FINDING

I

&

E = EgocentricO = Ob.iect-centeredG = Geocentric

. " - ; ; ; : , ; ; ; . ; ; " ; " " " ' .

- =l = '='ele=;l= -

Figure 33.r. Experimental setup for an example trial in Haun et al.,zoo6. Ten identical cups were placed on trvo tables [five cups oneach tableJ. The participant watched while a target was hlddenunder the cup depicted as white (HIDINGJ. Then participantsmoved to the other table and indicated where they thought asecond target might be hidden IFINDINGJ.

limited in their response capabilities, whileolder infants and toddlers may already beinfluenced by culture and language. ThePiagetian tradition holds that there is an ini-tial egocentric bias and a shift from egocen-tric to allocentric over development [Piagetand Inhelder, ry67; Pick, 1993), althoughthere is also some evidence for flexibility ininfants fAcredolg r97B; Bremner, r97B).

Haun et a1., [zoo6) addressed this in aset of studies that combines cross-linguisticdevelopmental comparisons with cross-species comparisons between humans andour close relatives, the great apes. For thecross-linguistic comparison, Haun et al.compared Dutch speakers, whose language[like EnglishJ primarily uses an egocentricframe of reference, with speakers of Haillom[a Khoisan language spoken in Namibia),which primarily uses a geocentric frame.Haun et al. used two-dimensional arraysof five objects, which allowed them to dis-tinguish the three frames of reference [seeFigure 33.r).

Participants faced an array of five identi-ca1 cups on Table r and watched as an objectwas hidden in one of the cups. They then

moved around to Table z, whereidentical array of cups, but from,site direction. Their task was tohidden object. So far this task,:prior frame of reference tasls;salutary innovation in thiswhereas prior research hadspeakers' preference for usingt'or another, this research utilizedtask, allowing the researcher$participants' facility in learniframe over another. Particiten consecutive trials in whictl'answer required use of thethen fwithout any break) tengeocentric frame, then ten using;itcentered frame. This techniqrle.:whether people find it easier totial task when it is set in theence dominant in their language;paralleled prior findings of Iin adults: Dutch-speakingto-ten-year-olds were moreegocentric condition than inwhile Haillom-sPeaking adultswere most accurate in the geoction. The striking difference

LANGUAGE AND COGNITION IN DEVELOPMENT

is just what would be Predicteddominant frame of reference ofive languages. [Of coursg thesecould have stemmed from other

.ior environmental differences; butet al. [zoo4) for arguments against

et al. then went on to Probe theof frame of reference and to

it develops. Their studies built onresearch of Ca1l and Tomasello;Tomasello and Call, 1997) com-man patterns with those of our

:i.usins among the great apes. They, apes [three orangutans, tworee bonobos, and five chimpan-

four-to-five-year-old German-,children a simplified search task.'

saw an obiect hidden underline of three identical cups on

l,.As in the prior study, they thenround to Table z, where theyr identical array from the oppo-

and looked for the hiddenbefore, participants had to dis-rule that determined the correct

ron Table z. There were two con-i,ggocentric, where the hiding and

maintained the same positioniio the participants' view point [left; and allacentric, where the hiding

f!1rg cups maintained the same posi-tiVe to an external frame. (In this

task, we can distinguish betweenand allocentric responding, but

the two kinds of allocentricbecause the geocentric (e.g., the) and object-centered [e.g., the

to the experimenter) framesthe same response.) The results

great apes performed best incondition, consistent with

showing that many species usespatial information to navigate[, rggo). But what is more sur-

that German-speaking children

German (and also English) use basicallysystem of spatial reference, with the ego-atiirl reference frame as dominant.

showed the same allocentric pattern; theirperformance resembled that of our simiancousins rather than that of older Germanspeakers. As Haun et al. note, this suggestsa deep continuity between humans and thegreat apes in their native cognitive biaseswith respect to reference frame.

This finding of an allocentric prefer-ence in four-year-olds dovetails with evi-dence that infants can adopt allocentric aswell as egocentric frames (Acredolo, 1978;Bremner, r 978) . At the same time, this cross-species allocentric bias in young humansand great apes renders the later cross-lin-guistic divergence all the more striking. Byeight years of age, children whose languagefavors an egocentric frame have divergedfrom their native pattern and now find theegocentric frame easier to use, while thosewhose language is geocentric show a corre-spondingly geocentric bias. Important, thisentrainment by language is not absolute.For example, older Dutch children andadults performed above chance in the geo-centric condition, and as Li and Gleitman[zooz) note, speakers of a given languageare also influenced by contextual factorsfsee also Gleitman and Papafragou, zoo5).Nonetheless, the fact that each group per-formed better on the frame favored by theirlanguage is evidence for effects of habituallanguage on the way we most readily con-ceptualize space.

tz The setnantics of containment andsupport

Within cognitive development, Bowerman[r98o; r9B9) was among the first to challengethe idea that concepts come first in humandevelopment and are simply mapped ontolanguage. Noting the variability between lan-guages in how spatial relations are lexicalized(e.g., Bowerman, r98r; 1996; Bowerman andChoi, zoo3; Bowerman and Levinson, 2oouLevinson et al., zooz), Bowerman arguedagainst the common assumption that cer-tain words are acquired eadier than othersbecause they correspond to preexisting con-ceptual categories. Instead, she suggested

6sl

6r8 GENTNER AND CHRISTIE

that concept learning might be guided bylanguage from the start.'

One striking example of such semanticdiversity is the contrast between Korean andEnglish spatial terms first documented byChoi and Bowerman (r99r). In English, spa-tial attachment is divided into semantic cat-egories of containment and support (in andon). ln contrast, Korean speakers organizespatial attachments accor&ng to how thetwo objects fit with one anothe4 contrast-ing tight fit with loose fit. In English, puttinga videocassette in Lts case or an apple in abowl are both categorized as containment.However, Korean uses two different verbs: avideocassette/case event is described by theverb bbita [roughly, to join things tightly),and the apple/bowl event is described bythe verb nehta (to join things looselyJ. Inthe other direction, the English distinctionbetween containment and support is notlexicalized in Korean: for example, the sameverb (kbita) is used for putting the top on apen and for putting an earplug into an ear[Choi and Bowerman, r99r).

If children form particular spatial con-cepts that are then mapped onto language,then we would expect an advantage forwhichever,language best matches children'snatural concepts. But in fact, Choi andBowerman found that English and Koreanchildren acquired their very different spatialsystems at about the same rate. In both lan-guages/ the first relational terms appeared atabout fourteen to sixteen months. In Englishthese early relational words were down, out,on, off, andopen,with come,faII,walk,run, sit,and ridc by seventeen to eighteen months.In Korean the early terms were bhlta {fit),p p ayta (unfit), y eba [op en), and t att a [closeJ,with kata (go), ancta (sit), pwuthita Quxta-pose two surfaces), hka(b)ta [peel off), etc.,by seventeen to eighteen months. There aresome commonalities: Young children like totalk about opening and closing and about

z Bowerman's recent research has also explored theother direction: that some categories are more nat-ural than others and that trhese categodes will bothbe more frequent in the world's languages and morereadily learned by children (the typological preua-Ience hypothesis) (Gentner and Bowerman, zoog).

moving around. But although both groupstalk about spatial relations, they pick outvery different parts of the spatial wodd tolexicalize, and this selection is guided bytheir language.

Further work has explored the earlyeffects of language: that is, at what pointdo infants begin to form different semanticcategories corresponding to their linguis-tic terms? Choi et

"1. ftqqq) found that aseady as eighteen months, infants are sensi-tive to language-specific spatial categories.In this study, upon hearing the spatial termin, English-speaking children selectivelyattended to scenes depicting containment(matching scene) as opposed to nonmatch-ing scenes. Similarly, Korean-speaking chil-dren attend to scenes depicting tight-fitrelations upon hearing hklta. During con-trol trials, where the children did not hearthe target word (in for English or hhita forKorean), there was no preference for eitherthe matching or nonmatching scenes, sug-gesting the absence of nonlinguistic biases.Choi et al. used a variety of different spa-tial scenes - e.g., for the containment rela-tion (in), scenes included putting a peg in ahole, Lego blocks in a box, books in box cov-ers, and rings in a big basket. The fact thateighteen-month-olds could correctly maptheir respective linguistic spatial terms tothis variety of scenes suggests that they havea generalized understanding ofthe contain-ment relation entailed by the spatial term.

Another line of support for the role ofspatial language in shaping spatial semanticcategories is a study by Casasola (zoo5), inwhich eighteen-month-old English-speakinginfants formed an abstract category of sup-port only when they heard the word on dur'ing habituation triais. Infants in this studywere all habituated to four support events,two depicting tight support and two depict-ing loose support. Infants then viewed fourtest events in sequence: two depicting sup-port relation [familiar) and two depictingcontainment [a new relation). Infants whohad heard the spatial word on during habit-uation looked longer at the novel relation(containment) with both familrrar and novelobjects, indicating that they had formed an

absttioneralall cchartheyt o a

Tis irfornConin tlinfa:cate[zocoldeithpairniecdistitharableiarizthegor)objeevel

Fyouoldfornmerloosuat(everexaltainwermerseqlconltainthatgor)evetloosmerwitl

7 Ale

LANGUAGE AND COGNITION IN DEVELOPMENT 6ss

abstract representation of the support rela-tion. In contrast, infants who had heard gen-eral phrases, novel words, or no words atall during habituation failed to notice thechange in relations even for familiar objects;they attended to a change in objects, but notto a change in relations.

These studies suggest that languageis instrumental in prompting infants toform stable spatial relaiional categories.Consistent with this claim, it appears thatin the absence of linguistic guidance, younginfants are ready to form a variety ofspatialcategories. McDonough, Choi, and Mandler(zoo 3J famili arized nine-to-fourteen-month-old English- or Korean-learning infants witheither pairs of tight containment events orpairs of loose containment events, accompa-nied only by music. Although the tight-loosedistinction is far more central in Koreanthan in English, both groups of infants wereable to extract the category during famil-iarization Both groups could distinguishthe familiar category from the new cate-gory when shown novel test pairs [with newobjects) consisting of a tight containmentevent and a loose containment event.3

Hespos and Spelke [zoo4) studied evenyounger infants and found that five-month-old English-speaking infants can readilyform either the English support/contain-ment distinction or the Korean tight fiVloose fit distinction. The infants were habit-uated either with a single tight containmentevent or with a loose containment event: forexamplg a cylinder entering another con-tainer that fit either tightly or loosely. Theywere then tested with both tight contain-ment and loose containment events (shownsequentially). Infants habituated to tightcontainment looked longer at the loose con-tainment event and vice versa, indicatingthat they had abstracted the respective cate-gory. More surprisingly, this pattern held upeven when infants had to transfer the tighVloose distinction from support to contain-ment. That is, when infants were habituatedwith either tight or loose support events, and

I A11 ages showed a familiarity preference in bothranguages.

then shown the tight containtnent and loosecontainment test events, they looked longerat the novel test event. This suggests thatfive-month-old infants can form the Koreantight-loose distinction, even when it cutsacross the English in/on disttnction.

Lining up the developmental studies dis-cussed so fa4 we have a rather perplexingcontrast. Five-month-olds showed sensitivityto the tight-loose distinction unmarked intheir native language in Hespos and Spelke's[zood study. But in Casasola's [zoo5) study,which also used a habituation paradigm,eighteen-month-olds failed to show sensitiv-ity to the support category which is markedin their language, unless they heard therequisite spatial term. We suggest that thisdifference may rely on the degree of gen-eralization that the infants needed to make(see also Casasola, zoo8J. In Casasola's study,the habituation events were quite varied andthe objects involved were perceptually richand differed across trials; in the Hespos andSpelke study, the habituation trials utilizedhighly similar events, both in the motionsinvolved and in the objects (which werevaried only slightly). Likewisg the test trialswere perceptually quite dissimilar from thehabituation trials in Casasola's study (espe'cially in the novel object trials) and percep-tually similar to the habituation trials inHespos and Spelke's study.

One might then ask "So which study isright? 'vVhen exactly do infants have thecategory of support?" We suggest that thisis the wrong question. Rather, the betterquestion is "When [and under what learningconditions) can infants form a category ofsupport at a given level of abstraction?" Ifwe consider that performance in these stud"ies derives in part from abstractions formedduring the study (rather than solely frompreexisting categories), then both kindsof study are informative. We can see the'sestudies as spanning a range. At one pole arestudies in which the intended relation is per-fectly aligned across exemplars with few dis-tracting surface differences (as exemplifiedin Hespos and Spelke's studies) - an idealsituation in which to form a generalization,albeit one that may not apply far beyond the

gs.

66o

initial stimuli. At the other pole are studieswith complex learning conditions, in whichthe relation is instantiated over differentkinds of objects (as in the Casasola stud-ies). That infants form the abstraction underoeifect conditions tells us that this potentialis there prior to language; and indeed theHespos and Spelke results show that infantsare multipotential learners at this eady stage'But the variable learning experience givento the infants in the Casasola studies moreclosely matches reallife learning conditions,in which children encounter a given spatialrelation instantiated over a wide variety ofspecific situations.+ From this perspective,tlis range of studies from ideal abstractionconditions to perceptually variable condi-tions can be seen as putting bounds on theconditions under which infants will formthe category.

t3 Rel"ational langaage and rclationalreptesent^tion

Gentner and colleagues have theorizedthat spatial relational language - and rela-tional language in general - can foster thelearning and retention of relational pat-terns [Centn er, zoo3, zoro; Loewenstein andGentner, zoo5), thus acting as a "cognitivetoolkit." During initial learning, hearinga relational term used for two situationsinvites children to compare them andderive their common abstraction (symbolicjuxtapositionJ. Once learned, a relationalterm can help to stabil ize the abstraction.Gentner fzoro) terme d this reification (seealso Lupyan, zoo8 for discussions of thisidea). The term can then be used to invitea particular construal of a given situation -one that may be advantageous for certain

o The McDonough et al. study also belongs on the"rich and varied" end of the continuum, with theadded important feature that they showed infantspairs of events (for example, two tight containmentevents) during Familiarization. Their finding that.,r"n tti.t"-rno.tth-old infants given pairs of eventscan abstract common relations from rich, complexstimuli is consistent with evidence that comparingtwo exemplars fosters the abstraction of common-alities, particularly common relations (Gentner andNamy, 1999; Oakes and Riba1, zoo5).

purposes. One particularly powerful kindof relational construal is a systematic tep-resentation: one in which the lower-orderrelations are interconnected by a higher-order constraining relation. For example,the set of terms top, middle, bottom forma systematic structure governed by thehigher-order relation of monotonicity inthe vertical dimension. This kind of con-nected relational system can be used tosupport inference and, as discussed laterin this chapter, analogical mapping andtransfer.

Recent evidence for the benefit of spa-tial relational language - and especially ofsystematic relational language - was offeredby Loewenstein and Gentner (zoo5) in aspatial mapping study. Preschool childrensaw two identical three-tiered boxes; theywatched an item being hidden in one boxand then searched for a similar item in thecorresponding location at the second box(see Figure g.z). Chlldren's performancewas better when they first heard the boxdescribed using spatial relational termssuch as on, in, under Fwther, when thetask was made more difficult by introduc-ing a competing object match (a cross-map'ping, Gentner and Toupin, 1986), childrenperformed far better with the terms top,middle, bottom (which convey a connectedsystem of relations) than with the termson, in, under, which lack a unifying higher-order structure. We infer that heating top'middle, bottom invited a representation otthe monotonic relational structure of thetwo boxes, and that this higher-order struc-ture helped the children to achieve a rela-tional mapping.

Further evidence of the influence of spa-tial relational language on spatial cogni-tion comes from a study by Dessalegn andLandau (zoo8). In this study, four-year-oldswere tested with a well-known problem invision: color and location conjunction (e.g.,a split square, with red on the left and greenon the rightJ. Children were presented witha target example [".g., "

red (left)-greenfrightJ square); their task was to find theexact match after a one second delay amongthree choices: the correct match [red-green

GENTNER AND CHRISTIE

t :.:..ri,'t,.:i:,t

Figure 33.:versions oIspatial ma1experimenbox and se

squareJ, 1or the disquare). Idren hearis on thehearing s1potentiallcues fflasof the re,languagerather thrthe task.s

e Howevesuch as pmetaphc

*

LANGUAGE AND COGNITION IN DEVELOPMENT 66t

Neutral

Relational Match Only

Cross-Map

Relational Match withCompeting object Match

Figure 33.2. Experimental setup for the twoversions of the Loewenstein and Gentner [zoo5Jspatial mapping task. Children watched theexperimenter place the "winner" card in the leftbox and searched for it in the risht box.

square), the reflection (green-red square),or the distractor (red-green diagonal splitsquare). Performance was best when chil-dren heard relational language (e.g., "the redis on the left"J. Crucial, the advantage ofhearing spatial language was not found forpotentially salient nonlinguistic attentionalcues (flashing, pointing, or changes in sizeof the red part], suggesting that relationallanguage affected the spatial representationrather than simply increasing attention tothe task.t

I However, interestingiy, other asymmetric termssuch as prettier also aided performance - possibly viametaphor mapping to the actual spatial situation.

t.4 Habitual construals: Beyond thinkingfor speahing

The studies reviewed so far suggest thatspatial language can influence children'sperformance on spatial tasks. Some of theseeffects couldbe explained bythe thinking forspeaking account. For example, Dessalegnand Landau's [zoo8) findings could beaccounted for by purely online, temporaryeffects of language. However, there are alsofindings that point to longer term effectsof language on the development of spatialrepresentations. For example, the effects oflanguage in the Loewenstein and Gentner[zoo5J spatial mapping task were durable,not fleeting. When children were broughtback to the lab two days later to "play thesame game," those who had heard system-atic language (top, middle, bottom) outper-formed those who had not, even thoughthe spatial relational terms were neverused during the second session. These chil-dren were also able to transfer the mappingtask to new, rather different-looking boxes.These findings suggest that the relationalterms induced a corresponding represen-tation which delineated the internal struc-ture of the boxes. Finally, in a recent studyin Istanbul, Gentner et al. [zoo8) comparedfive-year-old children who possess normallanguage with a group of deaf children whohad not been taught a sign language, andwhose self-developed homesign gesturalsystem fGoldin-Meadow, zoo3) was defi-cient in spatial terms. Neither group wasgiven any spatial language during the task.Nonetheless, the hearing children (who hadgood command of Turkish spatial terms)performed significantly better on the taskthan did the deaf children. ' :

There is also evidence from adult stud-ies that language can become internalizgdand come to influence our default concep-tual construals, for instance the frame of ref-erence studies discussed earlier. Studies ofsupport relation provide another example:In contrast to infants, adults show a strongpreference for the categories enshrined intheir native language. McDonough et al.[zoo3J gave adults an oddity task in which

GENTNER AND CHRISTIE

they had to say which of four events wasdifferent from the others. English-speakingadults could do this readily (seventy-eightpercent correct) when given three tightcontainment events and one loose supportevent; but when given three tight contain-ment events and one loose containmentevent [so that t]re choice had to be basedon the tight-loose distinction), only thirty-eight percent chose successfully. In contrast,Koreans ffor whom the tight-loose dis-tinction is part of habitual languageJ read-ily chose the odd item in this latter task

[eighty-percent correct). Likewise, Hesposand Spelke [zoo4) found tlat in making sim-ilarity judgments, English-speaking adultswere sensitive only to the Enghsh contain-ment-support distinction, and not to theKorean tight-loose distinction.

This development from equipotentialityin infants to linguistically biased similarityin adults suggests that we possess an earlyabillty to form a large number of potential&stinctions. Habitual usage of language ren-ders certain spatial categories more domi-nant. This does not mean that adults cannotlearn a new spatial category under favor-able learning conditions (e.g', Borodiaky,Schmidt, and Phillips, 2oo3; Goldstone,r99B); but it does mean that such learningmaybe difficult in ordinary life.Which rela-tions become easy to notice appears stronglyinfluenced by the language we speak'

z Number

Mathematical structure seems so compel-ling that it must be an inevitable aspectof human cognition. Dehaene (tggl, p.r1z) quotes French mathematician ChadesHeimite: "I believe that the numbers andfunctions of analysis are not the arbitraryproduct of our spirits; I believe that theyexist outside of us with the same characterof necessity as the objects of objective real-ity. .." Yet there is evidence that even simplenumerical insight is not inevitable and thatlanguage plays a role in its development.Webegin by describing two possible precursorsof number concept, and then discuss how

their interactions with language may giverise to number knowledge.

TWo preverbal capacities that have beenimplicated in accounts of number develop-ment are the analog magnitude system and asystem for keeping track of small numbers ofitems. The analog magnitude system is a sys-tem shared broadly with other species thatallows approximate judgments of quantity.It is what allows us [or a hamster) to choosea larger pile of grain over a smaller pilg orto notice that the amount of liquid in a con-tainer has decreased. This skill operates overeven very large quantities, but its accuracyis limited by Weber's Law: the discrimina-bility between two amounts is a functionof their ratio. Thus, inaccuracies occur formagnitudes that are very close. The analogmagnitude system is often modeled withthe accumulator model [Meck and Church,19B5J.-

The other relevant nonverbal capacity isthe ability to keep track of a small numberof items. This ability can be thought of as apart of our general capacity for representingmental models of the world; some accounts(e.g., Carey, zoo4; Spelkg zooo) have also[n[ed it with Pylyshyn's (zoor) notion of apreattentive object file system' In contrastio the analog magnitude system, the objectfile system operates over discrete represen-tations and is capacity limited to roughlythree or four objects.

We will consider two main classes of the-ories that assign a major role to language innumber development. One theory centers onlanguage as a link between modules,'while a,e.ind b.oad class oftheories focuses on thecount system and other number terms as ameans of promoting numeric insight.

zt Langu^age as knhbetueen madubs

Spelke fzooo; zoo3) and her colleagues theo-rize that language serves as a combinatorialsystem that linls the two preverbal numericmodules discussed above - the object fileand the analog magnitude system. Since nei-ther of these two preverbal modules dealswith exaalarge numbers, the combinatorialpower of language is needed for the ability

LANGUAGE AND COGNITION IN DEVELOPMENT

rc.r gi*:1}lrir;i'j:j

:1

:;

661

to represent exact large numbers. One lineof support for this theory comes from astudy with adult Russian-English bilingualsfSpelke and Tsivkin, zoot), which showedeffects of language in the performance ofexact arithmetic calculations. Bilingual par-ticipants were trained in one language ontwo kinds of problems: exact calculationproblems and approximation problems.Participants were later tested in similarproblems in the other language. Spelke andTsivkin reasoned that iflanguage is necessaryfor representing exact large numbers, thenperformance in exact calculation problemsshould deteriorate. However, approximatecalculation skill should not be affected. Thisis indeed what they found: Bilinguals wereable to transfer the new approximation skillsacross languages, but not the exact calcula-tion skills.

z,z Numbet Innguage as cognitive toolbit'While the view that language acts as a linkbetween modules recruits both the objectfile and the accumulator system, anotherview of how language learning supportsnumber development relies primarily on theobject file capacity (Carey, zoo4;Mix, zooz).These accounts recognize that knowledgeofthe count routine does not by itself con-fer an understanding of numbers, but holdthat learning the linguistic count sequenceis crucial in the development of numbers.The binding of the numbers words to car-dinal sets occurs slowly. A child may under-stand that one refers to an individual, butstill regard two, three, and so on as referringto undefined larger sets. Counting seems tobegin as a social routine, akin to a chant,and only later to become linked to cardinalnumbers (Fuson, rgBB; Wynn, r99o). A strik-ing demonstration of this lag is the fact thateven when a young child has just correctlycounted a set of objects ["one, twg three,four"), she typically cannot respond "four"to the question "So how many are there?"

This suggests an intriguing possibility:that the linguistic count routine serves asan analogy that invites children to orga-nize numerical quantities into an ordinal

sequence. This possibility ls most clearlyarticulated by Carey (zoo4; zoog) in herbootstrapping account fsee also Gentne4zoroJ. According to this account, childrenfirst learn the counting routine as a kindof gamg with no understanding of how itconnects to cardinal numbers. Gradually,the child learns to attach number words tovery small set sizes. The learning is at firstpiecemeal - even after binding two to sets ofcardinality twg weeks or months may ensuebefore the child realizes thatthree refers to aset with three items (Carey, zoo4;Mix, zooz;Mix, Sandhofe4 and Baroody, zoo5). Butonce a child reaches an understanding ofroughly three, or sometimes/our,the patternchanges. The child rapidly binds succeedingnumbers to their cardinalities, and showsunderstanding of the successor principle,that every (naturalJ number has a naturalsuccessor. This insight - "If number word Xrefers to a set with cardinal value n, the nextnumber word in the list refers to a set withcardinal value n + r" fCarey, zoo4) - occursvia an analogy between counting one furtherin the verbal count sequence and increasingby one in the set size.

But the analogy between the countingonefurther in the count sequence and adding onein quantity is very abstract. As Mix [zooz;Mix et al., zoo5) documents, children's earlyinsights into how numbers connect to setsize are often concrete and context-specific.For example, in Mix's [zoozJ diary study,at twenty months Spencer spontaneouslybrought from another room exactly twotreats for the family's two dogs, and repeatedthis feat with perfect accuracy several timesover the next few weeks. But he failed whenasked to go get "train treats" for his twotoy trains, suggesting that his command of"twoness" was highly context-bound.

Mix and colleagues have noted severalkinds of early nonverbal experience thatcontribute to the gradual acquisition ofnumerical insight, including several kindsof routines that promote one-to-one cor-respondence (such as distributing candiesamong several peopleJ. They also suggestan important role for language in the devel-opment of cardinality: namely, that hearing

)e

66+ GENTNER AND CHRISTIE

tvvo sets labeled with the same count wordcould prompt a comparison process thatleads the child to notice their commonnumber [Mix et al., zoo5). This line is con-sistent with the idea that common languageinvites comparison (Gentne4 zoo3; Gentnerand Namy, 1999; Loewenstein and Gentneqzoo5) which in turn can support categortza-tion [Gelman and Markman, r987;Waxmanand Klibanoff zooo).

In opposition to the above proposals,Gelman and Gallistel argue that languagehas little if any role in the development ofnumber. In their account, the analog magni-tude system is the cognitive foundation ofnumber knowledge [Dehaene, 1997; Gallisteland Gelman, Lggz).Gallistel, Gelman, andCordes [zoo5) argue further that the analogmagnitude system, whose output is con-tinuous rather than discrete, represents thereal numbers. As how to language may playa role in number development, Gallistel andGelman posit that "[A] system for arith-metic reasoning witJr real numbers evolvedbefore language evolved. When languageevolved, it picked out from the real numbersonly the integers..." [Gall istel et al., zoo5, p.247).This position reverses the usual suppo-sition of developmentalists and historiansthat understanding of the natural numbersappears first, followed by the integers, therationals, and the reals.

2.7 Research on la.nguages that lark fullcount systen$

One line of support for the hypothesisthat count terms are causally related tothe development of number knowledgecomes from studies of the Pirahd (Everett,zoo5; Gordon, zoo4), an Amazonian tribalgroup that uses what has been described asa "one-two-many" system of counting [hdi,hoi, baagi lor aibail). (See also Pica et al',zoot for similar results for the MundurukuJ.Gordon administered several numericaltasks using objects familiar to the Pirahl,over numbers between one and ten. Forexample, the experimenter would place, say,five batteries on the table, and ask the par-ticioant to "make it the same" with another

set of batteries. In another task, some nutswere put in a can, and then nuts were drawnout of the can one by one by the experi-menter. After each withdrawal, participantswere asked whether the can still containednuts or was empty. The results were strik-ing. The Pirahd participants performedwith good accuracy for up to three items,but performance became merely approxi-mate after three items. Nevertheless, perfor-mance beyond three was not random; it wasconsistent with the Weber fraction foundin results of people performing magnitudeestimation tasks. The Pirahd have the sameability to estimate numerosity as do Englishor French speakers; what they lack is a ver-bal counting system.

Striklng as it is, this finding has been rep-licated by a later study of the Pirah6 (Franket a1., zoo8). Frank et al. conducted the sametasks as in Gordon's study and again foundonly approximate performance for numbersbeyond three in tasls like the nuts-in-the-can task [though this time the Piraha per-formed better on the simpler versions of theone-to-one matching task than they had inGordon's study). An additional linguistictask administered by Frank et al. suggeststhat the count system of the Pirahd is evenless precise than the previously suggestedone-two-many system. In a numeral elic-itation task, speakers were shown a seriesof either increasing ffrom one to ten) ordecreasing (from ten to one) objects, andasked at each stage "how much/many arcthere?" While in the increasing elicitationcondition the word h6i was used only forone item, in the decreasing condition thesame word was used for quantities as largeas six. It appears that the Pirahd terms arenot true n-umbers, but are relative to thesize of the set - something more like "a few,more than a few, lots."

2.4 Effects of language onlaterdeu elo pment o f math ematic s

If number words are indeed crucial in thedevelopment of the number concept,- thenwe milht see different developmenlal pat'terns in number acquisition depending on

q

LANGUAGE AND COGNITION IN DEVELOPMENT 66s

the characteristics of number words in agiven language. Miller and Stigler (1987)suggest that one important difference isthe regularity of the number system. Theynoted that Chinese is more systematic thanEnglish in an important respect: Whereasboth languages have unique words for onethrough ten, Chinese is far more regularin its two-digit numbers than is Engfish. InChinese, eleven is "ten one," twelve is "tentwg" and so on throughout tJre teens, andthe system continues in this regular fashionwith "twenty-one," "t\,venty-twq" and so on.Contrast this system with English, in whicheleuen and twebe are opaque and althoughthirteen through nineteen are partly transpar-ent (ifthe child recognizes "teen" as "ten"), inaddition, the order oftens and units reversesafter twenty - we say twenty-one, twenty-two,and so on. Miller and Stigler hypothesizedthat a regular system like Chinese would beeasier to learn and use fsee also Fuson andKwon, r99z). Consistent with this predic-tion, they found that Chinese preschoolers(aged four and fiveJ were significantly bet-ter than their English-speaking peers in acounting task in which they had to count ashlgh as they could (Miller et a1., r995J. BothAmerican children [ninety-four percent)and Chinese children (ninety-two percent)could count to ten, but while seventy-fourpercent of Chinese children could count totwenty, only forty-eight percent ofAmericanchildren could do so.

What about the effects of linguistic vari-ability within a language - does the amountand quality of mathematical language influ-ence children's learning of that domain? Tofind out, Klibanoff et al. [zoo6) recordedthe kind of mathematical language used bypreschool or daycare teachers and relatedit to measures of the growth of children'sconventional mathematical knowledge overthe school year. They included language forordinality [e.g., "Point to the one that hasmore") and cardinality [e.g., "Point to fou4"given cards with varying numbers of itemsJ,as well as names for geometric shapes, theterm "half," and so on. The results showeddramatic differences in how much math-related talk teachers provided, and further,

that the amount of teachers' math-relatedtalk was significantly related to the growthof preschoolers' conventional mathematicalknowledge over the school year and unre-lated to their math knowledse at the start ofthe school year.

3 Theory of mind

Theory of mind refers to the ability to reasonabout mental states - beliefs, desires, inten-tions, and emotions. In large part, mentalstates are expressed via language, and hear-ing conversations about desires and inten-tions is one way of learning about others'mental states. But some researchers havetaken the link between language and theoryof mind further and have proposed that lan-guage plays a fundamental role in the devel-opment of theory of mind. Theories thatinvoke language differ as to which aspectsof language - whether pragmatics

"nd dir-

course structure, lexical semantics, or syn-tactic structure - are most fundamental fordeveloping a theory of mind.

A key question in the development oftheory of mind is how and when childrenbecome aware that other people's mindsmay not contain the same behefs as theirown mind. Performance on false belief tasksis one standard way of assessing whetherchildren have this understanding. Oneclassic false belief task, first introduced byWimmer and Perner (rg8lJ, is the unseendisplacement scenario. For example, three-to nine-year-old children are presentedwith a story in which Maxi puts his choc-olate in the kitchen cupboard. While Maxiis away, his mother moves the chocolate tothe drawer. Children are then asked whereMaxi will look for his chocolate when hereturns - in the cupboard or in the drawer.None of the three-year-olds correctly saidthat Maxi would look in tJre cupboard,whereas a majority of four-year-olds gavethe correct answer.

Another type of false belief task is the"smarties" study [Gopnik and Astington,r9B8; Perneq, Leekham, and Wimmet, tg87),which can be used to assess children's

666 GENTNER AND CHRISTIE

insight into their own minds. In this task,the child is shown a smarties box (smartiesbeing a type of candy), and asked what is inthe box. The child readily answers "candy"and is then allowed to look inside, where-upon she discovers to her surprise that thebox actually contains crayons. When theexperimenter asls what the child originallythought was in the box, five-year-olds cor-rectly say "candy," but most three-year-oldsinsist that they initially expected the candybox to contain crayons; some even claim tohave said so out loud.

Many results from theory of mind studiessuggest that the ability to represent people-'sepistemic states does not become firmly-established until around four to six years ofage.6 By this age, many cognitive and linguis-tic skills are already quite advanced. Manytheorists emphasize the importance of theseskills without assigning language a specialrole [e.g., Gopnik andWellman, r99z; Pernert99t). Others have argued that it is theory-ofmind that precedes language, rather than thereverse. In one version of this position, the-ory of mind stems from an innately develop-ing module [Baron-Cohen, 1999), which maysupport children's understanding of mentallangrrage but does not require language forits development. A weaker position is thatsome degree of interpersonal insight - nota-bly a sense of when joint attention is calledfor - is critical for language development[Baldwin, r99r; Tomaselo, 1998).

However, a sizable body of research hasargued for a role of language in the devel-opment of theory of mind. For examplgMilligan, Astington, and Dack [zoo7) foundin a metaanalysis of ro4 false belief studiesfor children under age seven that languageability is a significant predictor of false beliefunderstanding even when age is controlled.Overall, three major views have been pro-posed for how language may contribute tothe development of theory of mind: the dis-course pragmatic, the lexical semantic, and

6 However, some studies have found evidence ofan ability to represent at least some mental statesof others as ea.ly as fifteen months [Onishi andBaillargeon, zoo5). How tiese early sensitivitiesrelate to later patterns remain to be worked out.

the complementation syntax view. Althoughthese three views are not mutually exclu-sivg they make different bets as to whichaspect of language play a role in the devel-opment of theory of mind'

7t Discourse Ptagmntk

In the discourse pragmatic account, conver-sational pragmatics is critical in developingan understanding of other minds [Harris,r999J. Children first become aware of theirown mental states, and through simulationor role taking processes, they use this aware-ness to infer the mental states of others'Back-and-forth discourse allows childrento realize that they sometimes know whatothers do not, and vice versa. One line ofsupport for this view comes from a correla-tional study that showed that deaf childrenwho had more opportunities to participatein rich discourse interactions with othersalso performed better in false belief tasks

[Peterson and Siegal, zooo). In- anotherstudy, Dunn et al. [r99r) observed natural-istic conversations between two-year-oldsand their mothers. Seven months later, thechildren were queried on the understandingof other minds. They found that children'sengagement in family conversation aboutf".1it g states was positively correlated with-the ability to give.correct explanations offalse belief behaviors.

7.2 Lexical sem.antics

In the lexical semantics view, the acquisitionof mental state terms such as think., bnow,andbelieue plays a crucial role in the devel-opment of ihe understanding of false beliefs

[Astington, 1996; Bartsch andWellman, 1995;Bretherton and Beeghly, r98z). Childrenbegin to use mental state terms at about agetwo, especially perceptual and emotionalterms [see, hear;happy, sad, angry). Startingat age ihree, children also begin to produce.ognitiu. terms such as think, and hnow,butit is not until the early school years that chil-dren show clear discrimination among termssuch as think, know, and gtess [Bartsch andWellman, 1995; Bretherton and Beeghly,

.:!s'.1

i):LANGUAGE AND COGNITION IN DEVELOPMENT 662

r98z). The lexical hypothesis is that childrenacquire these mental state terms in con-versation; parents use them to refer to themental states that the child is experiencing,allowing the child to attach the terms to herown mental states. The child also noticesthat these terms can apply to other peo-ple, inviting the child to attribute the cor-responding mental states to others as well asto herself [Astington, 1996).

To test whether mothers' language influ-ences the development of theory of mind,RufFrnan, Slade, and Crowe fzooz) con-ducted a longitudinal study in which theyasked mothers to describe pictures to eighty-two children at three time points spanning aone-year period. They found that mothers'use of mental state utterances at eady timepoints was correlated with children's latertheory of mind understanding. The resultheld true even when a number of poten-tial intervening factors were accounted for,such as children's age, their language abil-ity, their own use of mental state languaggtheir earlier theory of mind understanding,and also mothers' education and other kindsof mothers' utterances. Ruffrnan et al. con-cluded that mothers' mental state utter-ances play a causal role in the developmentof theory of mind.

73 Compbmentation srytqx

Another prominentview is that acquiring thesyntax of sentential complements is a criti-cal factor in the development of false beliefunderstanding (de Villiers and de Villierqzooo). In a sentential complement construc-tion, a sentence takes a fuIl clause as its obiectcomplement: for example, "M"ry thinks ihatJohn is at home." This construction makes itrelatively transparent to see that the truthvalue of the sentence as a whole can differfrom that of the embedded proposition: thatis, the fact that Mary thtnhs that John is athome does not necessarily mean that Johnis at home. De Villiers and de Villiers notethat communication constructions such as "xsays that p" provide overt evidence for thisdisassociation when p is known to be untrue.In this way, communication verbs can serve

to bootstrap children's understanding of theuse of think: children learn to deal with falsecomplements via say, and by analogy cometo understand that think. too can take a falsecomplement.

One line of evidence for this hypoth-esis comes from a longitudinal study withpreschool children that found that perfor-mance on false belief tasks was predicted byperformance in interpreting sentences con-taining mental and communication verbswith complements [de Vlliers and Pyers,zoozJ. To gauge mastery of sentential com-plements, children were given scenarios likethe following: "She said she found a monsterunder her chair, but it was really the neigh-bor's dog" and then were asked "What didshe say?" Children's responses were countedcorrect as long as they said "a monster." Thechildren also carried out false belief taslasuch as the unseen displacement task andthe unexpected content task describedearlier. A positive correlation was foundbetween mastery of sentential complementsand success in false belieftasks.

Further evidence was found among oraldeaf children who are delayed in languagelearning (de Villiers and de Villiers, zoo3). Tocontrol for any effects of language requiredby the false belief tasls themselves, nonverbalfalse belief tasls were used. In one task, theexperimenter hid a sticker in one of the fouridentical boxes while a screen obscured thehiding from the child. On the test trial, two"helping" adults - one wearing a blindfoldand one who could see the hiding event -each pointed to a box after the screen wasraised, and the child had to choose whoseadvice to follow. In the other task, childrenwere shown a sequence of pictures depict-ing an unexpected content event and had tocomplete the sequence with either a surpriseface fcorrect) or a neutral face. The perfor-mance of these children on false belief taskswas predicted by their performance in thecomplement comprehension task.

Most recently, Pyers and Senghas (zoo9)took advantage of a naturally occurringchange in the linguistic affordances ofan emerging language - Nicaraguan SignLanguage [NSL) - to test whether language

668 GENTNER AND CHRISTIE

promotes false belief understanding over andabove social experience. NSL first appearedin the rgTos among deaf children enteringspecial education schools [Senghas, Kita,and Ozyiirek, zoo4).When the roughly fiftyoriginal children [who typically had devel-oped their own idiosyncratic homesign ges-ture systems; Goldin-Meadow, zoo3) werebrought together, they developed an eadyform of NSL as a common language. Thiswas further enriched by the second cohortof children in the mid-rg8os. Senghas andcolleagues report that even today, the sec-ond cohort exhibits a more developed formof the language than the older first cohort(senghas and Coppola, zoor; Senghas et a1.,zood. In particulal, the second cohort's lan-guage includes more mental state verbs thanthe first cohort's version - setting the stagefor a test of whether possession of mentalstate verbs supports reasoning about others'mental states. Important, aside from theirlanguage differences, the two cohorts havesimilar histories of schooling and socialinteraction.

Pyers and Senghas used a low verbal falsebelief task to test speakers from the twocohorts of Nicaraguan signers: Participantswere given a sequence of pictures depict-ing unseen displacement events [followinga false belief plotJ, and then had to choosewhich oftwo final pictures correctly depictedthe final event. The results showed a strongeffect oflanguage: The second cohort (aver-age age r7.5 years) by far outperformed thefirst cohort [average age 26.8 years). Out offour test trials, the second cohort solved onaverage 3.5 trials, as contrasted with .5 trialsfor the first cohort. Interesting, when testedtvvo years later, some of the first cohort whohad gainedmental state verbs also performedbetter in the false belief task. Overall, theseresulG support the idea that language pro-vides cognitive tools that support theory ofmind, even well into adulthood.

4 Summary'vVhile the issues are far from resolved,the evidence reviewed here suggests that

language may influence the developmentof conceptions of space, mind, and number.Language can foster cognitive developmentthrough various routes. The language wespeak provides us with tools for dissectingspace into finer categories, and for connect-ing those concepts into systems that permitcombinatorial inferences. Language can alsoinvite a systematic representation [as in theChinese numerals, or in the fluent numeri-cal abilities of English speakers as comparedto the Pirahi speakers). These representa-tions can support real-world tasls such asnavigation and calculation. They also sup-port abstract thinking, such as using spa-tial relations to reason about time, to plotkinship relations, or to comprehend graphsand other figures that use spatial patterns todepict nonspatial phenomena.

An interesting further question is: Whatis the time course of this linguistic influ-ence? The available research suggests thatlanguage effects are not immediate. It isnot enough to simply learn a set of terms(even supposing that their full meaningsare understood - a dubious assumption inearly development). We suspect that somedegree of entrenchment must occur beforethe new representation is sufficiently robustto have cognitive effects. For example, inthe Loewenstein and Gentner [zoo5J spatialmapping task, hearing spatial terms suchas top, middle , bottom improved the perfor-mance of youngeq but not oldel, children.Apparently, the younger children knew theterms well enough to benefit from hearingthem, but not well enough to access themspontaneously. The performance of theolder children is consistent with the possi-bility that the spatial system conveyed bythe terms was suffrciently entrenched tocome to mind with or without the terms.

In some cases linguistic patterns maygive rise to a habitual mode of construal'But in general, the evidence suggests thatthe influence of language is far from abso-lute. Human representation is rich and var-ied, and no one system of encoding is able togovern all of human representation. Therei, alro reason to believe that there is influ-ence in the reverse direction, from cognition

LANGUAGE AND COGNITION IN DEVELOPMENT 66g

prevalence of a given semantic systemoss langrr"g.t may afford an estimate of

ilbognltiu" naturalness and concomitantiof learning.

are many open questions. Forap1e, in which semantic arenas do we.the largest effects of language - are

larger effects of language in relativelyarenas, such as mathematics and

e, than in more concrete domains suchfu61or? Does language for emotions andthought processes influence the way in

we construe our own minds? Doesfng a technical language influence adult

oition in ways similar to the develop-iial patterns discussed here? Addressing

questions will give us a deeper under-of how language affects the devel-

of thought.

L. P. (1978). Development of spatial ori-€ntation in infancy. D eueLtpmental Psycholo g,

224-34.J.W. [r996J. What is theoretical

the child's understandins of mind?Vygotskian view of its development. In

P. & Smith, P. K. [Eds.] Theoriestheories of mind (pp. I8+-SSJ. Cambridge

ty Press.D.A. ftSStJ. Infants' contribution to

reference. Childachievement of ioint62,875-go.

K. & Wellman, H. (tSSl). Children talkthe mind. New York: Oxford Universitv

S. ft999J.The evolution ofa theory

guage. Some systems of categories arenatural - easier to learn and use - than

For example, Gentner and Bowermanbqj have suggested that the typologi-

conceptions of time. Cognitiue Psycholngt,4f t) , rzz.

Boroditsky, L., Schmidt, L., & Phillips, W (zoo3J.Sex, Syntax, and Semantics. In Gentne4 D.& Goldin-Meadow, G. (Eds.J Language inMind: Aduances in the study of Language andCognition [pp. 6r-8oJ. Cambridge: MIT Press.

Bowerman, M. (r98o). The structure and ori-gin of semantic categories in the language-Iearning child. In Foster, M. L. & Brande, S.(Eds.) Symbol as sense (pp. ,ll-gg). New York:Academic Press.

[r98rJ. The child's expression of meaning:Expan&ng relationships among lexicon,syntax and morphology. In Winitz, H. (Ed.)Natiue language and foreign Language acqui-sition [/ol. 779, pp, r7z-89). New York: NewYork Academy of Sciences.

(r989J. Learning a semantic system: What roledo cognitive predispositions play? In RicgM. L. & Schiefelbusch, R. L. [Eds.J The teach-ability of language [pp. tll-6S). Baltimore,MD: Paul H. Brooks.

[1996). Learning how to structure space for lan-guage: A crosslinguistic perspective. In Bloom,P., Peterson, M. A., Nadel, L., & Garrett, M. F.(Eds.) Lang.rage and space (pp. Z8S-+16).Cambridge: MIT Press.

Bowerman, M. & Choi, S. (zoo3J. Space underconstruction: Language specific spatial cat-egorization in first language acquisition. InGentne4, D. & Goldin-Meadow, S. [Eds.)Language in Mind: Aduances in the studyof Language and Cognition (pp. Z8l-+t8).Cambridge: MIT Press.

Bowerman, M. & Levinson, S. [Eds.J. (zoorJ.Language acquisition and conceptual deuelop-mmt (preface, introduction). Cambridge, UK:Cambridge University Press.

Bremner, J. G. (1978). Egocentric versus allocen-tric spatial coding in nine-month-old infants:Factors influencing the choice of code.Deuelnpmental Psycholng, 4, 1,46-55.

Bretherton, L & Beeghly, M. (r98zJ. Talking aboutinternal states: The acquisition of an explicittheory of mind. Deuelapmental Psycholog, fi,9o6-zr.

Brown, P. [rSS+J.The ins and ons of Tzeltai loc-ative expressions: The semantics of staticdescriptions of location. Linguistics, 32, 7 4;'-9o.

Call, J. (zoorJ. Object permanence m orang-utans (PongopygmaeusJ, chimpanzees(Pantroglodytes), and children (Homosapiens).Journal of Comparatiue Psycholng, :n5, ry9-71.

Carey, S. fzood. Bootstrapping and the origin ofconcepts. Daedalus, t3j.

r.nind. In Corballis, M. C. & Lea, SEG (Eds.)E,descmt of mind: Psycholngical perspediueshomini"d euolution (pp.26v7fl. New York:

University Press.; J. & Brown , P. (zoo7). Standing

and loca-languages.

: dispositional predicatespredications in two Mayan

, ' t ) l )1 .

L. (zoorl.7 Mandarin

Does language shapeand English speakers'

67o GENTNER AND CHRISTIE

(zoo9]. The origin of concepts. Oxford UniversityPress, USA.

Casad, E. H. & Langacker, R.W. [r985J. "Inside"and "outside" in Cora Grammar. InternationalJoumal of Amencan Lingtistics, 5t,247-8r.

Casasola, M. [zoo5). Can language do the driv-ingT The effect of linguistic input on infants'categorization of support spatial relations.Deuelnpmental Psychologt, 4t, t83iz

[zoo8J. The development of infants' spatial cat-egories. Current Direc'tions in PsychologicalSaence, 17,2r-5.

Choi, S. & Bowerman, M. (r99rJ. Learning toexpress motion events in English and Korean:The influence of language-specific lexicaliza-tion patterns. Cognitian, 4t, 87-rzr.

Choi, S., McDonough, L., Bowerman, M., &Mandler, J.M. 0SSS). Early sensitivity to lan-guage-specific spatial categories in English andKorean. Cognitive D euelopmmt, t4, 24148 -

Dehaene, S. GSSZJ. The number sense: How themind creates mathematics. New York: OxfordUniversity Press.

Dessalegn, B. & Landau, B. (zoo8J. More thanmeets the eye: The role of language in bind-ing and maintaining feature conjunctions.Psychologcal Science, ry, t89-95.

de Villiers, J. & de Mlliers, P. (zood. Linguisticdeterminism and the understanding of falsebeliefs. In Mitchell, P. & Riggs, K. J. (Eds.)Children's reasoning and the mind (p:p' r9t-zz8j. Hove, England: Psychology Press.

[zoo3J Language for thought: Coming tounderstand false beliefs. In Gentnel, D.& Goldin-Meadow, S. [Eds.) Language inMind: Aduances in the Study of Language andThougftt (pp. :lS-8+). Cambridge, MA: MITPress.

deVilliers,J. G. &Pyers,J. E. [zooz). Complementsto cognition: A longitudinal study ofthe rela-tionship between complex syntax and false-belief understan&ng. Cognitiue Deuelapmmt,t7, ro37-6o.

Dunn, J., Brown, J. R., Slomkowski, C., Tesla,C., & Youngblade, L. (r99r). Young children'sunderstanding of other people's feelings andbeliefs: Individual &fferences and their ante-cedents. Chil"d Develapment, 62, ry52-66.

Everett, D. (zoo5). Cultural constraints on gram-mar and cognition in Pirahd: Anot}er lookat the design features of human language.Cunent AnthroPologt, 76, 6z-46.

Frank, M.C., Everett, D., Fedorenkg E., & Gibson,E. [zoo8J. Number as a cognitive technology:Evidence from Pirahd language and cognition.Cognition, to9,8ry-24.

Fuson, K. C. (r98SJ. Child,rm's counting and con-cepts of number New York: Springer Verlag'

Fuson, K. C. & Kwon, Y. 0SSr). Learning addi-don and subtraction: Effects of number wordsand other cultural tools. In Bideaud, J., Meljac,C., & Fische4 J. P. {Eds.) Pathways to number:Children's deuelaping numencal abilities fpp.283-3o6J. Hillsdale, NJ: Erlbaum.

Galllitel, C. R. [r99o). The organization of bam'ing. Cambridge, MA: MIT Press.

Gallistel, C. R. & Gelman, R' (1992). Preverbal andverbal counting and computation. Cogninon,44,43-74.

Gallistel, C. R., Gelman, R., & Cordes, S. (zoo5J.The cultural and evolutionary history of thereal numbers. In Levinson, S. & Jaisson, P.(Eds.) Culture and euoluhon. Cambridge, MA:MIT Press.

Gelman, S.A. & Markman, E. M' [1987). Youngchildren's inductions from natural kinds: Therole of categories and appearances. ChildD eueloPmmt, 58, ry32-4r.

Gentner, D. (1982). Why nouns are learned beforeverbs: Linguistic relativity versus natural par-titioning. In Kuczaj, S. A. (Ed.] Language dcvel'opment: VoI. z. Language, thought and culture(pp. ;ot-l+).Hillsdale, NJ: Lawrence ErlbaumAssociates.

[zoo3J. Why we're so smart. In Gentner, D. &Goldin-Meadow, S. (Eds.) Language in min"d:Aduances in the study of lanpage and thought(pp.tg5-46). Cambridge, MA: MIT Press.

(z,oroJ. Bootstrapping children's learning:Analogical processes and symbol systems'Cognirtue Scimce, l+G), I Sz-1 5.

Gentner, D. & Bowerman, M' [zoo9). Why somespatial semantic categories are harder tolearn than others: The Typological Prevalencehypothesis. In Guo, J., Lieven, E. Ervin-tipp,S., Budwig, N.,. Ozgaliskan, S., & Nakamura,K. (Eds.] Crosslinguistic approaches to the psy--chologr of language: Research in the tradition ofDan Isaac Slobin (pp' 465-8o). NY: LawrenceErlbaum Associates'

Gentner, D. & Christie, S. (zoro). Mutual boot-strapping between language and analogi-cal processing. Language and Cognition, z(z),z6r-83.

Gentner, D. & Goldin-Meadow, S. [zoo3).Whither\Mhorf In Gentner, D. & Goldin-Meadow, S'(Eds.) Language inmind: Aduances inthe studyi1 longuog, ond thought (pp. l-t+J. Cambridge,MA: MIT Press.

Gentner, D,, & Namy, L' ISSSJ. Comparisonin the development of categories. Cogn'itiueDeueloPmmt,4, 487-19.

LANCUAGE AND COGNITION IN DEVELOPMENT 67,

Gentneq, D., Ozyrirek, A., Goldin-Meadow, S., &Gurcanli, O. [zoo8J. Spatial language potenti-ates spatial cognition: Ettidence from deaf home-signers. Presented at the American Associationfor the Advancement of Science annualmeeting.

Gentner, D. & Toupin, C. (r986J. Systematicityand surface similarity in the development ofanalogy. Cognitiue Science, rc, 277-7oo.

Gleitman, L. & Papafragou, A. [zoo5J. Languageand thought. In Holyoak, K. & Morrison, B.(Eds.) Cambridge Handbooh of Thinhing andReasoning (pp. 633-6rJ. Cambridge : CambridgeUniversity Press.

Goldin-Meadow, S. (zoo3). The resilience of lan-guage: What gesture creation in deaf childrencan tell us about how all children lBam lan-guage. NewYork: Psychology Press.

Goldstone, R. L. (1998]. Perceptual leaming.Annual Rertiew Psycholo gt, 49, 585-612.

Gopnik, A. & Astington, J. W. (1988). Children'sunderstanding of representational change andits relation to the understanding of false beliefand the appearance-reality distinctio n. Chil"dDeuelopment, 59, 2637.

Gopnik,A. &Wellman, H. (1992).Whythe child'stheory of mind really is a theory. Mind andLanguage, T,145-7r.

Gordon, P. (zoo4). Numerical cognition withoutwords: Evidence from Amazonia. Science,3o6,4c,6-99.

Gumperz, J. J. & Levinson, S. C. (r996J. Rethinbinglinguistic relatirtity. Cambridge, England:Cambridge University Press.

Harris, P. L. (r999J. Individual differences inunderstanding emotion: the role of attach-ment status and psychological discourse.Attachment & Human Deuelopment, 3,307-24.

Haun, DBM, Rapold, C., Call, J., Janzen, G.,Levinson, S. C. (zoo6). Cognitive cladisticsand cultural override in Hominid spatial cog-nition. Proceedings of the National Academy ofSciences, rc3, ry568-73.

Hespos, S. & Spelke, E. (zoo4). Conceptual pre-cursors to language. Nature, 43o, 457-6.

Hunt, E. & Agnoli, F. 0SSr). The Whorfianhlpothesis: A cognitive psychology perspec-tive. Psychological Reutew, g8, yl-8g.

Imai, M. & Gentner, D. ISSZJ.A crosslinguisticstudy of early word meaning: Universal ontol-ogy and linguistic influence. Cognition, 62,r69-zoo.

Imai, M. & Mazuka, R. [zoo3). Reevaluatinglinguistic relativity: Language-specific cat-egories and the role of universal ontological

knowledge in the construal of individuation.In Gentner, D. & Goldin-Meadow, S. [Eds.)Langtage in mind: Advances in the study of lan-guage and cognition fOp. +S-6+).Cambridge,MA: MIT Press.

Klibanoff, R. S., Levine, S. C., Huttenloche4J., Vasilyeva, M., & Hedges, L.V. (zoo6).Preschool children's mathematical knowl-edge: The effect of teacher "math talk."D euelopmmtal P sycholog, 42, Sc-f,g.

Levinson, S. C. (1996). Relativity in spatial con-ception and description. In Gumperz, J.J.& Levinson, S. C. (Eds.) Rethinking linguis-tic relativity (pp. r77-zoz). Cambridgg MA:Cambridge University Press.

(zoo3). Language and mind: Let's get tJre issuesstraightl In Gentneq, D. & Goldin-Meadow,S. (Eds.) Langmge in mind: Aduances in thestudy of language and cognition [pp. ,S-+6).Cambridgg MA: MIT Press.

Levinson, S., Kita, S., Haun, D., & Rasch, B. (zooz).Returning the tables: Language affects spatialreasoning. Cognition, 84, y5-88.

Levinson, S. C., & Brown, P. (tSS+). ImmanuelKant among the Tenejapans: Anthropology asapplied philosophy. Ethos, zzft), 7-4t.

Loewenstein, J. & Gentneq, D. [zoo5J. Relationallanguage and the development of relationalmapping. Cognitiue Psychologt, 50, 315-53.

Li, P. & Gleitman, L. [zooz). Turning the tables:Language and spatial reasoning. Cognition, 83,26s-94.

Lucy, J.A. (t994). Grammartcal categories andcognition. Cambridge: Cambridge UniversityPress.

Lupyan, G. (zoo8J. From chair to "chair": A repre-sentational shift account of object labelingeffects on memory. Journal of ExperimentalPsycholog : General, 97 Q), 348-69.

Lupyan, G., Rakison, D. H., & McClelland,I.L. (zoo7). Language is not just for talking:labels facilitate learning of novel categories.P sychological Saence, t8 ftz), rc7 7 -83.

Majid,A., Bowerman, M., Kita, S., Haun, DBM, &Levinson, S. C. (zoo4J. Can language restruc-ture cognition? The case for space. TRENDSin Cognitiue Sciences,8, ro8-r4.

Markman, E. M. GSSS). Categorization andnaming in children: Probbms of induction.Cambridge, MA: MIT Press.

McDonough, L., Choi, S., & Mandler, J. (zoq).Understanding spatial relations: Flexibleinfants, lexical adults. Cognitiue Psycholog, 46,229-59.

Meck, W. H., Church, R. M., & Gibbon, J.[r985J. Temporal integration in duration and

67, GENTNER AND CHRISTIE

numb er discri min ation. J ournal of Exp erime nt alPsychobg: Animal Behauior Processes, n,

591-7.Mifer, K.F. & Stiglel, J.\M [1987)' Counting in

Chinese: Cultural variation in a basic cogn!tive skill. Cognitiue D evelopment, 2' 27 9-3c5'

Millea K. F., Smith, C.M.,Zhu, J., &Zhang,H'(r99d. Preschool origins of cross-national dif-i.;;;;* in mathematical competence: Therole of number-naming systems' PsycholngicalScience,6, 564o.

Milligan, K., Astington, J'W., & Dack, L'A'[zoo/. Language and theory of mind: Meta-

"""lytit of the relation between language

abillty and false-belief understanding ' ChiAD euelaPmmt, 77, 6zz-46.

Mix, K. S. (zooz). The construction of numberconcepts. Cognitiue DeueLopment, t7, t31543'

Mix, K. S., Sandhofer, C. M., & Baroody, A' J'(zoo5) Number words and number concepts:the interplay ofverbal and nonverbal quanti-fication in early childhood. In Kail, R' V' [Ed')Aduances in Child Develnpment and Behavior[Vol. ]3, pp. 3o5-46).New York: AcademicPress.

Oakeg L. M. & fubar, R. J. (zoo5). A compar-ison of infants' categorization in paired andsuccessive presentation familiarization tasksInfancy,7,85-98.

Onishi, K. H., & Baillargeon, R. [zoo5]. Dor 5 -month-old infants understand false beliefs?Scimce,3o8,255-8.

Pederson, E. 0SS5)' Language as context, lan-guage as means: Spatial cognition and habituallanguage use. Cognitiue Linguisncs, 6, 33-62'

Pemer, J. [r99r). Understanding the representa-tionalmind. Cambridge, MA: MIT Press'

Perner, J., Leekham, S. R., & Wimmer, H' [IS8ZJ'Three-year-olds' dlfficulty with false belief:The case for a conceptual deficit' BrinslrJournal of Deuelopmental Psycholog, 5, 125-37'

Peterson, C. C. & Siegal, M' (zooo)' Insighs intoa theory of mind from deafness and autism'Mind Et Language, ry, n3-45.

Pica, P., Leme1, C., lzard, V., & Dehaene, S'[zoo4). Exact and approximate arithmetic inan Amazonian indigene gtoup- Science, 3o6,499-503.

Piaget, J. & Inhelder, B. [1967). The chil"d's concep-tion of sPace. New York: Norton.

Pick, H. L. Jr. (1993). Organization of spatialknowledge in children. In Eilan, N., McCarthy,R., & Breweq, B. (Eds.) Spatial Representanon:Probbms in Phibsoplry and Psycholag (pp'3r-42). Blackwell Pubhshing.

Pvers, J. & Senghas, A' [zoo9)' Language-pro-' -ot". faise-belief understanding: Evidence

from Nicaraguan Sign Language' PsycholngicalSamce, zo,8o5-rz.

Pylyshyn, Z. (zoot). Visual indexes, preconcep-' irrai obiects, and situated vision' Cognirion, Bo,

127-56.Rrff-u.,, T., Slade, L., & Crowe, E' [zooz)' The

relation between children's and mothers'mental state language and theory of mind

understandin g. Chill Deuel'opment, 73' 734-5r'Senghas,A. & Coppola, M' (zoorJ' Ciildren creat-

ing language: How Nicaraguan Sign Language

".i"i..d a spatial grammar' Psychological

Science, tz,3z7-8.Senghas, A.,

'Kita, S., & Ozytirek, A' (zoo4)'

Childt".t creating core properties of language:Evidence from an emerging sign language inNicaragua. Science, 3o5, t779-82'

Slobin, D. I. [1996). From "thought and language"to "thinking for speaking'" In Gumperz, J' J'-&Levinson, S-. C. (Eds.) Rethlnhingllnguistic rela'til)ity (pp. 7 0-96).Cambridge, MA: CambridgeUniversitY Press.

[zoo3J. Language and thought online: Cognitive- .oi-rr"orl..r..t of linguistic relativity' InGentner, D. & Goldin-Meadow, S' [Eds')Langttage in mind: Advances in the study of langu"g, i"d cognition (pp. tSl-')'Cambridge,MA: MIT Press.

Spelke, E. S. [zooo). Core knowledge' AmericanP sychologist, 55 ftt), nY-43

(zoo3). What makes us smart? Core knowl'- edge and natural language. In Gentner, D', &Goldin-Meadow, S. [Eds.) Langtage in mind':Aduances in the sudy of language and cognition(p:p. z773tz)' Cambridge, MA: MIT Press'

Sp"iL, g. S. A Tsivkin, S' [zoorJ. Initial knowl-edge and conceptual change: space and num-b.I In Bo*".man, M. & Levinson, S' C' [Eds']Language acquisition and concepual deuelgO'**1 1pp.7"-rooJ . Cambridge, UK: CambridgeUniversitY Press.

Talmy, L. (1975). Semantics and syntax of motion'IrrKimball, J. (Ed.) Syntax and semantics $oL4, pp. r8r-238). New York: Academic Press'

fro8cl. Lexicalization patterns: Semantic struc-' it." i" lexical forms. In Shopen, T' [Ed ]Langtnge, typolog and syntaaic dc.scr-rOtion:VoI.-2. Grammatical categories and the bxicon

bp.il-t+g).New York: Cambridge UniversityPress.

Tomasello, M. 0SS8). Reference: Intending,that others jointly attend. Pragmatics and

Cognition,6, zzg-43.

LANCUACE AND COGNITION IN DEWLOPMENT 6zlTomasellq M. & Call, J. QggT). Primate Cogtition.

New York, NY: Oxford Universitv press.Vygotsky, L. S. (r96zJ. Thought and language.

Cambridge, MA: MIT Press. (Original workpublished 1934).

Waxman, S. R. (zooz). Linla between object cat-egorization and naming: Origins and emer-gence in human infants. In Rakison, D. H. &9"br,f.M. (Eds.J Early category and conceptdeuelnpmmt: Mahing sense of the bloomiig,buzzing confusion tpp. zt3-41. Ny, New yorf:Oxford University Press.

Waxman, S. R. & Klibanoff R. S. fzoooJ. The roleof comparison in trhe extensionof novel adiec-tiv es. Deuelopmmtal Psychobgt, 36, 57 vtl..

Waxman, S. R. & Markow, D. B. (1995). Wordsas invitations to form categories: Evidence

fuT y- to r3-month-old infants. Cognith.teP sycholo g, z9 (3), 257 -3o2.

mo{ B. L. (r956J. Language, thought andreakty: Selcaed writings. Cambridge, MA:Technology Press of Massachusetts InstituteofTechnology.

Wq"T.l H. & Perner, i. 0SS3). Beliefs aboutbeliefs: Representation and constraining func-tion ofwrongbeliefs in young children,s under-

_. - :9"$"g of deception. Cognition, g, 4r4g.Wo_ltl & Holmes, K. (zorr). Linguistic relativity.- WREs Cognitive Science, z,253-265.Wynn, K. (r99oJ. Children,s understanding of_ - counting. Cognition, 36, ry5-tg3.Xu, L (zooz). The role of language in acquiring

object kind concepts in infancy. Cognitinn, gl,1.r->w.

I ii it'i { .i L:l. E