Journal of Memory and Language 44, 250–273 (2001)doi:10.1006/jmla.2000.2751, available online at http://www.academicpress.com on

Less Really Is More for Adults Learning a Miniature Artificial Language

Alan W. Kersten and Julie L. Earles

Florida Atlantic University

Three experiments provide support for a key prediction of Newport’s (1988, 1990) “Less Is More” hypothesis.Adults were found to learn a miniature artificial language better when they were initially presented with onlysmall segments of language than when they were presented immediately with the full complexity of the language.Adults who were presented initially with individual words and only later with complex “sentences” composed ofseveral such words learned the meanings and morphology of those words better than did adults who were pre-sented with sentences throughout learning. The externally imposed constraint of processing only small segmentsof language is conjectured to be similar to the internal processing constraints of children that have been proposedto aid children in their acquisition of language.© 2001 Academic Press

Key Words: language; development; critical period; constraints; meaning; morphology.

Language learning is one of those rare tasksa

s

t

pmf gm

laaetoi

cally tested. One theory of the critical periodess).de-

ae

d-

e-sn-nlestas

ter-e

at which children seem to perform better thadults. Although adults are faster processorinformation and acquire most skills much moeasily than do children, there is ample colloquand observational evidence to suggest that cdren learning a language achieve higher umate competence than do adults learning same language. This evidence has promptedsearchers to propose that there is a criticalriod for language acquisition, extending frobirth to around puberty. Although the notion ocritical (or at least sensitive) period for languaacquisition is widely accepted, the mechanisunderlying this critical period remain uncleaMany theories that have been offered to expthis critical period involve little more than restatement of the basic finding that childrlearn better than adults, making it difficult generate further predictions that can be emp

We thank Cindy Barnes, Stephanie Danti, Andy Winc

0749-596X/01 $35.00Copyright © 2001 by Academic PressAll rights of reproduction in any form reserved.

25

and Hanako Yoshida at Indiana University and StephaRooney at Florida Atlantic University for their help in conducting these experiments. We thank Brian Bowdle, LinSmith, and Jesse Spencer-Smith for interesting discussregarding the project. We also thank Rob Goldstone for use of his lab in conducting Experiments 1 and 2. Animatexamples of the events used as stimuli in this researchavailable at http://www.psy.fau.edu/chez/awk.

Address correspondence and reprint requests to AKersten, Department of Psychology, Florida Atlantic Unversity, Boca Raton, FL 33431-0991. Fax: (561) 297-21E-mail: akersten@fau.edu.

n ofreialhil-lti-

that does make testable predictions is the “LIs More” hypothesis of Newport (1988, 1990The research presented in this article was signed to test a key prediction of this theory.

EVIDENCE FOR A CRITICAL PERIOD

,nieer

is-

--

’sn

n

dit

6

0

he re-e-

aes

r.in

n

ri-

IN LANGUAGE ACQUISITION

Evidence for a critical period comes fromnumber of different sources. Of course, thmost convincing evidence for a critical periowould come from depriving an otherwise normal child of language input until after the criti-cal period and then observing the child’s subsquent ability to learn language. Obviously, thiexperiment cannot be done due to ethical cosiderations. Occasionally, however, childrehave been discovered who had received littexposure to language prior to puberty. The mocelebrated example of one of these cases wGenie, a child who was locked in a separaroom and was allowed almost no social inteaction until her discovery at age 13 (Curtiss1977). Despite subsequent efforts to teach GeEnglish and American Sign Language, she nevachieved proficiency at grammar. This findingconsistent with a critical period for language acquisition. The case of Genie is difficult to interpret, however, because of the unusual circumstances of her upbringing. Genie’s fatherbelievedshe was mentally retarded from birth. Geniedifficulty at learning language may have bee

k,ie

-a

onsheedare

lani-0

e

n

e

r

n

be

e

t

of of ofne tog- of-

ngte

larng orer

&-ce

-ec-ortds,

ndtyto-

rbthts aestge

ostdt

ierng-n

rte

n-orve

LESS REAL

related to mental retardation rather than the lage at which she attempted to learn languaEven if she were not mentally retarded whshe was born, 13 years of social isolation couhave had dire consequences for Genie’s cogtive development, exacerbating any langualearning difficulties she may have had.

More conclusive evidence for a critical periocomes from research on the acquisition of Amican Sign Language (ASL). ASL is an interestitest case because, unlike spoken languages,ple are not necessarily exposed to it from birThere are a couple of reasons for this. First, age at which deafness is diagnosed can vary.viously, hearing parents would not expose thchild to sign if they did not know the child wadeaf. Second, even after a child has been dnosed, parents do not always immediately courage their child to learn ASL. Parents instesometimes attempt to teach their child to readlips of English speakers, a difficult task that donot generally result in a full command of Enlish. As a result, many deaf children are not posed to ASL until they reach elementary schage or even later. Thus, children who otherwreceive an entirely normal upbringing are not eposed to useful language input until much lathan most hearing children.

Research on the acquisition of ASL has vealed that people who were exposed to ASLan earlier age showed better ultimate comtence at ASL. For example, Newport (1990) hdemonstrated that early learners of ASL shbetter knowledge of verb morphology. In otest, ASL signers were asked to describe a sple event involving an object (e.g., a ball) moing along a particular path (e.g., a straight linwith a particular manner of motion (e.g., rollingSuch events can be appropriately descriusing a verb that includes a separate morphfor each of these three components of meanWhereas early learners of ASL produced thdifferent morphemes correctly, later learnersASL were more likely to produce verbs involving one or more incorrect morphemes. In parular, late learners of ASL often produced verinvolving combinations of morphemes that fr

quently go together but were not appropriate the situation being tested. For example, after

-

LY IS MORE 251

atege.nldni-ge

der-g

peo-th.theOb-eirsiag-en-adtheesg-x-

oolisex-ter

e-at

pe-asoweim-v-e)).edme

ing.se

of-ic-bse-for

peated exposures to a particular combinationmorphemes (e.g., a ball rolling), late learnersASL may start to use that same combinationmorphemes to describe a situation in which oof the morphemes is no longer relevant (e.g.,describe a ball bouncing). This finding sugested to Newport (1990) that late learnersASL were not analyzing verbs into their constituent morphemes, but rather were treatithem as unanalyzed wholes. In particular, lalearners of ASL may have treated a particucombination of morphemes as a unit, producithis combination even in situations where onemore morphemes were not appropriate. Othresearchers (Emmorey, Bellugi, Friederici, Horn, 1995; Emmorey & Corina, 1990; Mayberry & Eichen, 1991) have also found evidenof difficulties in morphological processing inlate learners of ASL.

Further evidence for a critical period in language acquisition comes from research on sond-language learning. Johnson and Newp(1989) tested knowledge of English syntax anmorphology in people who had, at varying ageemigrated to the United States from Korea aChina. Participants performed a grammaticalijudgment on a number of sentences designedtest different rules of English grammar. For example, to test a participant’s knowledge of vemorphology, participants were presented wisentences such as “Yesterday the hunter shoodeer” (p. 73) and were asked if such sentencwere grammatical. Participants’ ability to rejecsuch sentences was strongly related to their aof arrival in the United States, with people whhad arrived before the age of 7 performing beand people who had arrived during adulthooperforming worst. This finding suggests thapeople who were exposed to English at an earlage showed better ultimate competence at Elish grammar, consistent with a critical period ilanguage acquisition.

Although the results of Johnson and Newpo(1989) are suggestive of a critical period in thacquisition of the grammar of a second laguage, this conclusion remains controversial fa number of reasons. First, several studies hafound an effect of age of acquisition on gram

re-matical processing even when only adult arrivals

N

e

oc

tuc

a

e

h

n e

llbt

edsonla-on Aatinan.anded

ngm-

sr asastsilye inAs

derte-tym-yhiseat-of

-beanesot-s-

ssi-

terlter-on

n-ithrini-

252 KERSTEN A

are considered (Bialystok & Hakuta, 1999; Birdsong, 1999; Birdsong & Molis, 2001). If thefindings of Johnson and Newport (1989) indereflect a maturationally bounded critical perioin language acquisition, then some other eplanatory mechanism must be used to explathe continued decline in language acquisitioability with increased age in adulthood. Alternatively, it is possible that some mechanism oththan a critical period may be able to account fdeclines in language acquisition performanthroughout the lifespan.

Second, a number of studies have documenthat adult learners of a foreign language somtimes achieve nativelike competence in the gramar of that language (see Birdsong, 1999, foreview). If these learners truly perform idencally to native speakers of a language, this woseem to argue against the notion of a critiperiod. It remains possible, however, that theadult learners use different strategies than dotive learners to perform successfully on testsgrammatical knowledge. Even though these dferences may not be evident in off-line grammicality judgments, they may be revealed by oline (e.g., Emmorey et al., 1995) or physiologicmeasures (e.g., Weber-Fox & Neville, 1999) language processing.

Third, a number of variables are correlatwith age of arrival in a foreign country, and could be these variables rather than age ofrival per se that are responsible for the apparcritical period. For example, the older one when one first arrives in a foreign country, tmore established one’s native language is. Aresult, older learners of a second language mbe more strongly influenced by their native laguage, causing them to have difficulty acquirithose aspects of the second language thatdifferent from their native language (Fleg1999; Oyama, 1979). This may explain the dferent results of Johnson and Newport (198and a recent study by Birdsong and Mo(2001). Birdsong and Molis attempted to repcate the results of Johnson and Newport, with native Spanish speakers rather than naKorean and Chinese speakers. Birdsong aMolis found that age of arrival was unrelated

performance on English grammaticality judg

D EARLES

-

ddx-inn-

erre

tede-

m-r ai-ldalsena-ofif-t-

n-alof

ditar-entise

s aay

n-gare,

if-9)isi-ut

ivendto

ments for people who had arrived in the UnitStates before the age of 17. In contrast, Johnand Newport found a strong negative corretion between age of arrival and performance grammaticality judgments for this age range.possible explanation for this difference is thSpanish is much more similar to English terms of grammar than are Chinese and KoreAs a result, whereas increases in Chinese Korean grammatical competence with increasage may lead to greater difficulties in learniEnglish grammar, increases in Spanish gramatical competence may not.

Although the results of Birdsong and Moli(2001) suggest that a native Spanish speakeold as 16 years of age can learn to performwell as a native English speaker on off-line tesof English grammar, this does not necessarmean that he or she is processing the languagthe same way as a native English speaker.discussed above, it remains possible that ollearners of a language may use different stragies to perform successfully on grammaticalijudgments than do younger learners. For exaple, a late learner of English may try to explicitlmap an English sentence onto a sentence inor her own language in order to determinwhether the sentence is grammatical. This stregy may be effective for a native speakerSpanish because of the similarity of the grammar of the two languages, whereas it mayquite unsuccessful for a native speaker of Koreor Chinese. Although these different strategiin native and late learners of a language may nbe revealed by off-line grammaticality judgments, they may be revealed by different meaures, as discussed above. Thus, it remains poble that there is a critical period insecond-language acquisition, but that lalearners of a language can sometimes use anative strategies to perform as well as nativescertain measures of language competence.

In addition to competence in one’s native laguage, a second variable that is confounded wage of arrival in a foreign country is the numbeof years of formal education one has receivedthe language of that country (Flege, YenKomshian, & Liu, 1999). Children who arrive

-in a foreign country receive their primary

r

la

rit

u

tgh

itit

h

geen

.rha-ina

ue

li

hs

vm

,l

r-t

elyofh-, itof

tod

nre-

nsn.nn,dsss

a possible mechanism underlying the critical

ssit-geta-tntoryes aendultnd

en-

-ofce

LESS REAL

schooling in the language of that countwhereas adults may receive relatively little fomal schooling in their second language unlethey take classes in English as a secondguage or classes at the university. Thus, formschooling rather than age of arrival may besponsible for the apparent critical periodsecond-language acquisition. Indeed, Flege e(1999) found that age of arrival in the UniteStates had no significant effect on the acqsition of English grammar by native Koreaspeakers when years of education in the UniStates was controlled. In particular, althouage of arrival and years of education were higcorrelated in their sample, they were ableselect two subgroups of participants who dfered in mean age of arrival in the United Sta(9.7 vs 16.6 years) but who were matchedterms of years of education in the United Sta(10.5 years). These two subgroups did not pform significantly differently on tests of Englisgrammar.

Although Flege et al. (1999) suggest that aof arrival in and of itself is not important to thacquisition of the grammar of a second laguage, an alternative interpretation remainsparticular, although the two subgroups wematched in terms of years of education in tUnited States, the nature of that education inlikelihood differed dramatically for the two subgroups. A group of participants who arrivedthe United States at about 10 years of agecompleted 10 years of education most likestarted in about fifth grade and completedleast some college in the United States. A groof participants who arrived in the United Statat about 17 years of age and completed 10 yeof education, on the other hand, most likecompleted not only high school and collegethe United States, but also must have eittaken a large number of classes in English asecond language or completed some gradustudy. The latter group of participants may habeen a highly motivated and/or intelligent saple, which may have allowed them successlearning English as a second language. Thusalternative interpretation of Flege et al.’s resu

is that high levels of motivation and/or intelligence may allow some late immigrants to

LY IS MORE 253

y,r-ssn-al

e-nal.

di-

nedhlytof-esneser-

foreign country to compensate for their late arival, allowing them to learn the language of thacountry as well as earlier immigrants. Thlate immigrants in this analysis, however, onaveraged 81% correct on Flege et al.’s testEnglish grammar, whereas no native Englisspeaking control scored less than 90%. Thusremains to be seen whether additional yearseducation would allow these late immigrantsfully compensate for their late arrival anachieve nativelike competence.

In summary, results from studies of ASLprovide strong evidence for a critical period ithe acquisition of one’s native language. Theis also evidence for a critical period in secondlanguage acquisition, although there remaisome controversy regarding this conclusioDespite this evidence for a critical period ifirst- and perhaps second-language acquisitiothe mechanisms underlying this critical perioremain unclear. In the next section, we discuone theory of the critical period that describe

- is

-Ineell

ndlyatpsarsynera

atee-atants

period.

THE “LESS IS MORE” HYPOTHESIS

Newport (1988, 1990) has proposed the “LeIs More” hypothesis to explain the apparent crical period in language acquisition. Accordinto this hypothesis, it is no coincidence that thoptimal period of life for language learning is aa time when one’s information processing capbilities are quite limited. In particular, Newporhas proposed that children’s inferior informatioprocessing capabilities actually help themlearn language. Because of working memolimitations, young children can only perceivand remember small segments of language. Aresult, a child will only be able to map thessmall segments of language onto events fouin the external environment. In contrast, an adcan remember larger segments of language athus may be able to map entire phrases ontovironmental events.

According to Newport (1988, 1990), the necessity of children to focus on small segmentslanguage results in superior ultimate competenat the language. The reason for this advantage

athat, in language, meanings are constructed from

e

o

sof

rast

s

n

bh

o

ag

s

s-

ofeg-

w-ta-setgsoflean them-hentsallyld

ingnd

m-elyre-pli-ndrkted

- re-inhater-ester, in

-s-d.

-,-

nggnt

ges

within a sign rather than processing the entire

254 KERSTEN A

combinations of morphemes, each of which minvolve only a brief segment of speech or sign.order to achieve full proficiency in a languagone must learn the meanings of individual mophemes as well as the rules for combining themorphemes to create more complex meaninChildren’s inability to process more than limitesegments of language may encourage themlearn the relations of individual morphemes (possibly even individual syllables within thosmorphemes) to environmental events. In cotrast, because adults can process language alevel of entire words or even phrases, adults mhave a more difficult time determining which apects of meaning are relevant to a particular mpheme and which aspects are relevant to a difent morpheme within a segment of languagAdults may instead associate entire wordsphrases with environmental events. This wouexplain why older learners of ASL often producunanalyzed, holistic signs involving one or moincorrect morphemes. Older learners of ASL mbe able to remember entire ASL words and asciate those words with environmental evenfailing to appreciate the complex morphologicstructure underlying the composition of thowords.

There are at least two possible benefitsprocessing small segments of language, as cdren are proposed to do. One is that a learwho focuses on small segments of languagemore likely to learn the meanings of individuasegments. In contrast, a learner who focuseslarger segments of language may learn to asciate those larger segments with meanings,may be less likely to learn the meanings of tlower level components of language from whicthose larger segments are constructed. A secbenefit of processing small segments of laguage is that a learner who does so is mlikely to learn the rules for combining thossmall segments to convey more complex meings. In contrast, a learner who focuses on larsegments of language may learn a set of usephrases for expressing common meanings (e“Where is the bathroom?”), but is less likely tlearn how these phrases were constructed. Twould limit one’s ability to convey new phrase

(e.g., “Where is the kitchen?”), even if on

ND EARLES

ayIn,

r-segs.d

tor

en-t theay-r-

er-e.orldeeyo-s,ale

ofhil-eris

lonso-utehondn-re

en-erful.g.,ohis

knew the meanings of the all the words necesary to construct those phrases.

There is computational evidence for both these proposed benefits of processing small sments of language. First, Goldowsky and Neport (1993) have demonstrated that a computional model learns the mapping between a of linguistic forms and environmental meaninbetter when an input filter limits the amount form and meaning information that is availabto the learner on a given trial. Second, Elm(1993) has demonstrated that a model learnsgrammar of a language (i.e., the rules for cobining subjects and verbs) better when tmodel at first processes only small segmeof language and later processes incrementlonger language segments. This finding heboth when processing was limited by presentthe model with only small speech segments awhen processing was limited by imposing meory limitations on the model, thus limiting thnumber of words that could be simultaneousprocessed. It should be noted that other searchers have not always been able to recate Elman’s findings. In particular, Rohde aPlaut (1999) found that a connectionist netwolearned a language better when it was presenimmediately with the full complexity of the language. Rohde and Plaut were never able toproduce Elman’s findings despite variations parameter settings, and thus it is unclear wfactor or factors are responsible for these diffent patterns of results. Elman’s results suggthat at least under some conditions, howevprocessing limitations can lead to advantagesthe learning of grammar.

In addition to computational evidence, experimental evidence of a benefit from procesing limitations has recently been discovereCochran, McDonald, and Parault (1999) examined the learning of ASL morphology by adultseither under externally imposed processing limitations or under normal conditions. Processilimitations were imposed either by requirinparticipants to carry out a second, concurretask while learning ASL signs or by instructinparticipants to process only certain morphem

esign. In both cases, processing limitations led to

L

S

ita

hg

oigo

mtnewh.

itet

or

e

enoe

ia

o sigthig

iol

r

ot-

oin-llre-ingb-isinerif-is-

erb-

ta-nsse bend-Al- inmry ofi-esinsar-e

aine-ita-i-orethenly

testi-

ge.e in-to

LESS REAL

advantages in learning certain aspects of AFor example, in some trials participants hadcombine the relevant morphemes from two dferent signs, each of which involved the simulneous expression of an object, a manner of mtion, and a path (e.g., “person walking straigahead” and “animal running in circles”). Doinso would allow participants to produce a nesign involving the simultaneous expressionthe object and manner of motion from one sand the path from the other sign (e.g., “perswalking in circles”). Participants upon whoprocessing limitations were imposed did betat producing these signs. These participawere apparently better at decomposing thsigns into their constituent morphemes, alloing them to produce only those morphemes twere relevant to the situation to be described

Although Cochran et al.’s (1999) results asuggestive of a benefit from processing limtions, there remain some difficulties in the intpretation of these results. First, although paripants who were subject to processing limitatiperformed better than participants with no pcessing limitations on certain aspects of ASparticipants with no processing limitations pformed better on other aspects. For examplesome trials participants had to combine the rvant morphemes from two different signs, oinvolving the simultaneous expression of an ject, a manner of motion, and a path (e.g., “pson walking straight ahead) and the othervolving the simultaneous expression of object and a manner of motion (e.g., “legs liming”) followed bythe expression of the path that object (e.g., “on a zig-zag path”). Doingwould allow participants to produce a new sinvolving the simultaneous expression of object and manner of motion from one s(e.g., “legs limping”) followed bythe expressionof the path from the other sign (e.g., “straigahead”). Participants upon whom no processlimitations were imposed performed better these signs. As a result, there was no overalvantage to the processing limitation groupany of Cochran et al.’s (1999) experiments.

In defense of Cochran et al., adults do perfo

better than children on certain aspects of laguage after limited amounts of learning (Snow

uld

Y IS MORE 255

L.tof--o-t

wfnn

ertsse-at

rea-r-ic-nso-L,r-, inle-eb-r-

n-n

p-fonen

htngnad-in

mn-

Hoefnagel-Höhle, 1978), and thus it may nbe surprising that adults with no processing limitations may similarly outperform adults whare subject to processing limitations on certaaspects of ASL within the brief context of an experiment. Children ultimately show an overaadvantage in language learning, however, wheas an analogous overall benefit from processlimitations has yet to be demonstrated. In the asence of this evidence of an overall benefit, itunclear whether processing limitations arefact beneficial to language learning or wheththey simply result in a change in emphasis to dferent types of material, with greater emphason some types of material in the processing limitation group and greater emphasis on othtypes of material in the group that was not suject to processing limitations.

A second issue with regard to the interpretion of Cochran et al.’s (1999) results concerthe generalizeability of their findings. BecauCochran et al. used ASL as their language tolearned, it remains to be seen whether their fiings would generalize to spoken languages. though ASL is equivalent to spoken languagescomplexity, the constraints on a visual mediuare quite different from those on an auditomedium. In particular, the visual presentationsigns in ASL allows for a greater amount of smultaneity in the presentation of morphemcompared to that in spoken languages. It remapossible that processing limitations may be pticularly useful in the acquisition of ASL becausthey encourage the processing of only certmorphemes within a set of several, simultanously presented morphemes. Processing limtions may be of more limited utility in the acqusition of spoken languages because the mserial nature of such languages may require presentation of and thus the processing of oone morpheme at a time.

The present research was designed to whether there are benefits from processing limtations in the acquisition of a spoken languaIn particular, if adults’ ability to process largsegments of language is responsible for theirferior language learning, then forcing adults process smaller segments of language sho

&significantly improve their language learning. To

e

o

r

ont

p

g

id

iy

mn

aa

nheheengsec-

-ngslta

nr,

edist.uss-rdaslyth,

of-as

toew

asesren

e-ese

-

256 KERSTEN A

test this hypothesis, adult participants wpresented with a miniature artificial languagParticipants saw a number of simple animaevents on a computer screen, similar to themployed by Kersten (1998a, 1998b). Each evwas accompanied by one or more novel woThe task was for participants to learn the meings of the novel words by observing their crelations with attributes of the accompanyievents. Critically, some participants heard en“sentences” of novel words throughout learninwhereas others were initially presented wsmaller segments of speech that became inmentally larger over the course of learning.the “Less Is More” hypothesis is correct, peowho are initially exposed to smaller segmentslanguage should show better ultimate learnof the language than should people who areposed to the full complexity of the languafrom the beginning of learning.

The present manipulation is similar to thatElman (1993), who compared the learning omodel that was exposed to long sentence strthroughout learning to the learning of a mothat was exposed to sentence strings whlength increased over the course of learnElman was interested in the acquisition of stax, however, whereas the present researchvestigated the acquisition of word meanings Experiments 1–3) and morphology (in Expement 3). The present manipulation is also silar to that of Cochran et al. (1999), who istructed some participants to concentrateindividual morphemes within an ASL signwhereas they instructed other participantsprocess entire signs. In contrast to the reseof Cochran et al., however, the present reseexamined the acquisition of a spoken langua

in which the different words were presented s

r

dfSr

ar-outnnar-outr-,--

rially rather than simultaneously.

EXPERIMENT 1

In the first experiment, participants were psented with animated events accompaniedsentences involving up to three different worAlthough these sentences were composed odependent words, they were similar to the Asigns studied by Newport (1990) and Coch

et al. (1999) in that each was constructed from

ND EARLES

ree.

tedse

entds.an-r-gireg,ithcre- Ifle ofingex-e

off angseloseng.n- in-(inri-

i--on,torchrchge,e-

e- bys. in-L

an

morpheme referring to the moving object in aevent, a morpheme referring to the path of tmoving object, and a morpheme referring to tmanner of motion of the moving object. Somparticipants were presented throughout learniwith sentences involving one of each of thethree different types of words. These partiipants were expected to have difficulty decomposing a sentence into the component meaniexpressed by individual words, just as adulearners of ASL have difficulty decomposing verb into its component morphemes.

Other participants were forced to focus osmaller segments of language. In particulaeach of these participants was initially presentwith only one word along with each event. Thword referred to the moving object in the evenBecause these participants were forced to focon individual words, they were expected to eaily learn the meanings of those words. One-thiof the way through learning, a second word wpresented along with one of the previouspresented words. This word referred to the paof the moving object for some participantswhereas it referred to the manner of motion the moving object for other participants. Because the first word in each such sentence walready familiar, participants were expected be able to determine the meanings of these nwords. Finally, two-thirds of the way throughlearning, the final word in each sentence wadded. At this point in learning, the sentencheard by participants in the two conditions weidentical. Participants for whom words had beepresented incrementally, however, were prdicted to have an easier time decomposing thsentences into their component meanings.

This prediction of the “Less Is More” hypothesis is somewhat counterintuitive because pticipants who heard entire sentences throughlearning actually received more informatiothan participants who were forced to focus osmaller segments of language. For example, pticipants who heard entire sentences throughlearning had the opportunity to learn manneof-motion words from the beginning of learningwhereas participants who initially heard only object words had to wait until later in learning be

afore being given their first opportunity to learn

c

a

br

eo

i

teo

po

a

ref

iny

e.re

v-thehend.to-ndas

“geseju elnugop doochatig” would have been presented

LESS REAL

manner-of-motion words. In particular, partiipants who heard entire sentences throughlearning saw one of two different manners motion in each of 72 learning events and hethe corresponding manner-of-motion word the accompanying sentence to each of thevents. Participants who initially heard only oject words also saw one of the two differemanners of motion in each learning event, the words that corresponded to those mannemotion were not presented until later.

Because participants who heard entire stences throughout learning had all of the infmation available to them, a participant in thcondition who adopted the strategy of learnone word at a time should have performedleast as well as a participant who initially heaonly individual words. For example, if a particpant in the sentence condition decided to sby learning the first word in each sentence, hshe could have moved on to learning the secword in each sentence as soon as the first win each sentence had been learned. A particiwho initially heard only individual words had twait until one-third of the way through learninbefore starting to learn the second word in esentence, even if he or she had learned theword in each sentence much earlier in learnin

The “Less Is More” hypothesis, however, pdicted that participants who heard entire stences throughout learning would have difculty decomposing a sentence into its componmeanings. As a result, these participants wexpected to have difficulty when they wetested on their knowledge of the meanings ofdividual words. Participants who were forcedfocus on small segments of language, on other hand, were expected to perform better.

Method

Participants

One hundred twelve undergraduates at Inana University participated in this experimentpartial fulfillment of introductory psychologcourse requirements.

Stimuli

All events. Animated events were displayeon Apple Macintosh IIsi computers using th

le

d

Macromedia Director 3.1 software packagEach event was accompanied by one or mowords presented to participants in a fema

orally through headphones while the event took place.

LY IS MORE 257

-outofrd

inoseb-ntuts of

n-r-isng atrdi-art orndordant

gchfirstg.e-n-i-entererein-tothe

di-

FIG. 1. Three frames from an example event. The moing character starts the event at the top of the screen onleft. The stationary character appears at the bottom of tscreen. The event takes place against a marshy backgrouOver the course of the event, the moving character movesward the stationary character, angling its legs forward aback as it moves. A string of one or more words such

evoice through headphones that were attached to

258

r

d

r

o-aytlythen-ee

inor-

-nt byed on, ain.dedrt-ed

the computers. Three frames from an examevent are depicted in Fig. 1. In each event, bug-like creature moved while the other mained stationary, serving as a reference pfor the motion of the other bug. Events varion a number of attributes. Some of these abutes involved the static characteristics of objects in the events. In particular, each of characters in an event was composed of a ha body, and a pair of legs. The head, body, legs of the stationary character, as well as head of the moving character, each had fpossible values. These values are depicteFig. 2.. The body and legs of the moving chacter each had two possible values, chosendomly for each participant from the four valudepicted in Fig. 2.

The motion of the moving character also vied on two attributes. First, the moving charater moved on one of two different paths with spect to the stationary character. These

FIG. 2. The four different values of ea

paths were chosen randomly for each par

pleonee-ointedttri-thetheead,andtheour in

ar-ran-es

ar-c-e-two

ipant from the following four paths: (1) directlytoward the stationary character, (2) indirectly tward the stationary character, (3) directly awfrom the stationary character, and (4) indirecaway from the stationary character. Second, manner of motion of the moving character ivolved two different motions of the legs of thmoving character with respect to its body. Somcharacters angled their legs forward and backsynchrony, whereas others angled their legs fward and back in opposite directions.

A final attribute of variation in the events involved the static background on which an evetook place. This background was representedtwo identical pictures appearing in two unuscorners of the screen. Each event took placeone of four possible backgrounds: a desertswamp, a mountainous terrain, or a rocky pla

Each event began when a black screen faaway to reveal the two characters in their staing positions. The stationary character start

ch of the static object attributes in Experiment 1.

KERSTEN AND EARLES

tic-each event at a random location near the center

i

ane

e

pa

itput

o

”fw

tj

oman

he

n

tas

ord

stedter.ried

oncter ont to toam-

awdifferent objects and paths. The pairings of words to cate-

LESS REAL

of the screen. The moving character started eevent separated vertically, horizontally, or diaonally from the stationary character. The dtance between the two characters was chosuch that a moving character with a path towthe stationary character would just come icontact with the stationary character at the of the event. Each event lasted about 5 s.

Learning events. The first 72 events were dsigned to teach participants the meanings ofnovel words. For participants in the sentencondition, every learning event was accomnied by three words. For example, a participmay have heard the words “geseju elnugdoochatig” while seeing the event depictedFig. 1. These sentences were presented wflat intonation contour, with brief pauses serating the three words. One word in each s“sentence” ended in “-ju” and was related to body and legs of the moving character in event. Two different words with this ending weheard during learning. For example, a givparticipant may have always heard the w“geseju” in the presence of a moving characwith a particular body and pair of legs, wherehe or she may have heard the word “mogajuthe presence of a moving character with a difent body and pair of legs (see Fig. 3). The twords of this type that were heard by a givparticipant were chosen randomly from a sesix possible words (i.e., bidiju, geseju, hetilemuju, mogaju, and yuboju).

A second word in each “sentence” ended“-gop” and was related to the manner of motiof the moving character in an event. For exaple, a given participant may have always hethe word “ontigop” in the presence of one maner of motion, whereas he or she may haalways heard the word “elnugop” in the presenof a different manner of motion (see Fig. 3). Ttwo words of this type that were heard by a givparticipant were chosen randomly from a setsix possible words (i.e., ajegop, aubeegop, elgop, issogop, ontigop, and utegop).

A third word in each “sentence” ended in “-tigand was related to the path of the moving chacter with respect to the stationary charac

For example, a given participant may have ways heard the word “neematig” in the presen

th

”ar-er.al-

of one path of the moving character, wherehe or she may have always heard the w“doochatig” in the presence of a different pa

gories was determined randomly for each participant.

LY IS MORE 259

achg-s-senrdtond

-sixcea-ntopinh aa-ch

heanreenrdteras iner-

oen ofu,

inn-

rd-

veceenofu-

FIG. 3. Examples of the three different kinds of wordlearned in Experiment 1. Object words were differentiaon the basis of the body and legs of the moving characThe appearance of the head of the moving character varandomly. Manner of motion words were differentiated the basis of the motions of the legs of the moving charawith respect to its body. Path words were differentiatedthe basis of the path of the moving character with respecthe stationary character. The particular pairings of wordscategories in this figure are only meant to serve as an exple. Different participants heard different words and s

ce(see Fig. 3). The two words of this type that

set

c

”thio

rt

wehb

no

n

na

d

o

esdh

e

as

m-g

sta-edmre

teds-ts

the ofac-eresme ofonere.” ofieddth

r-ony-ctter-edofre-ar-r-e

ic-hts

eddar-he

260 KERSTEN A

were heard by a given participant were chorandomly from a set of six possible words (i.doochatig, farnitig, neematig, quopatig, segoand vustutig).

The three words in a “sentence” always ocurred in the same order for a given partipant. For half of the participants in the setence condition, the first word (ending in “-jureferred to the moving object in the events, second word (ending in”-gop”) referred to tmanner of motion of that object, and the thword (ending in “-tig”) referred to the path. Fthe other half of the participants, the first woreferred to the object, the second word referto the path, and the third word referred to manner of motion. These two sentence ordwere chosen because they were consistent English grammar. For example, both the stences “The boy ran into the room” and “Tboy entered the room running” are acceptain English.

The two different object words combined iterchangeably with the two different manner-motion words and the two different path wordThus, there were eight possible sentences, rering to the eight possible combinations of thethree attributes. Each of these eight sentenwas presented nine times during learning. Torder in which these sentences were presewas determined randomly for each participawith the constraint that each sentence had topresented three times during the first 24 evethree times during the second 24 events, three times during the final 24 events.

The events seen by participants in the invidual word condition were generated in thsame way as those seen by participants insentence condition. The first 24 events seenthese participants, however, were each accpanied by only a single word, in particular onof the two object words. The next 24 evenwere each accompanied by an object woalong with a second word. This was a mannof-motion word for participants who were asigned the Object–Manner–Path sentence orwhereas it was a path word for participants wwere assigned the Object–Path–Manner stence order. The last 24 learning events w

each accompanied by a “sentence” involvin

ND EARLES

en.,

ig,

c-i-

n-)heerdr

rdedheersithn-ele

-f-s.fer-secesheted

nt, bets,nd

i-ethebym-etsrdr--er,o

en-re

three words in the assigned word order, justfor participants in the sentence condition.

Attributes that were not related to the accopanying words (i.e., the head of the movincharacter; the head, body, and legs of the tionary character; and the environment) varirandomly throughout learning. This randovariation was included to make the task mochallenging to participants.

Isolated word meaning test events. After the72 learning events, participants were presenwith 12 trials testing for knowledge of relationbetween individual words and individual attributes of the events. In each trial, participanwere presented with two events, one after other. The two events in a trial varied on onethe three attributes that were related to the companying words during learning. The othtwo relevant attributes held constant valuacross the two events in a test trial. The saisolated word was presented along with eachthe two events in a trial. Events that varied the body and legs of the moving character wboth accompanied by a word ending in “-juEvents that varied on the manner of motionthe moving character were both accompanby a word ending in “-gop.” Events that varieon the path of the moving character were boaccompanied by a word ending in “-tig.” A paticipant’s task was to choose which of the twevents was a better example of the accompaing word. The choice of whether the correevent was presented first or second was demined randomly for each trial. Four trials testfor knowledge of each of the three types words. The order in which these trials were psented was determined randomly for each pticipant. All attributes that varied randomly duing learning continued to vary randomly in thtest trials.

Embedded word meaning test events. Afterthe isolated word meaning test events, partipants were presented with 12 trials in whicthey were asked to choose which of two evenwas the better example of a word embeddwithin a “sentence.” As in the isolated wormeaning test events, the two events in a trial vied on one of the three relevant attributes. T

gtwo events in a trial were both accompanied by

p

n

n

ap

r

in

a

er

oc

wtO

ns,

otedri-erentire

hisionedordct–lsoirdestu-ntialordts.

of

inlli-

ari-in-

of

teddidgh-

pe,tned

han

ifi-

ar-ticipants learned the meanings of words in a

LESS REAL

the same “sentence” involving three words. Oevent was entirely consistent with the accomnying “sentence,” whereas the other event winconsistent on the value of the varying attribuin that trial. Because only one of the three worin a “sentence” was related to the value of thattribute, these trials again tested for knowledof relations between individual words and indvidual attributes. Unlike the isolated word meaing test events, however, these words were psented in the context of entire “sentences.”

The embedded word meaning test evewere included in this experiment along with thisolated word meaning test events because oparticipants in the individual word conditiohad ever heard an event accompanied by onsingle word during learning. As a result, if thindividual word condition performed better thathe sentence condition in the isolated womeaning test events, one could attribute thisvantage to the sentence condition’s lack of exrience with events accompanied by individuwords. On the other hand, if the individual wocondition also performed better than the setence condition in the embedded word meantest events, this alternative hypothesis wouldapply, given that both conditions had heaevents accompanied by sentences.

Procedure

Participants were instructed that they wouldlearning words from a language spoken onother planet. They then put on a pair of heaphones,afterwhich theyclickedonabutton in tlower right hand corner of the screen label“Begin” to view the first learning event. Afteeach learning event, a button labeled “NeEvent” appeared in the lower right-hand cornerthe screen, allowing a participant to move onthe next event.After watching 72 learning evenparticipantswerepresentedwith12 isolatedwmeaning test trials. After the first event in eatrial, a button labeled “Next Event” appearedthe lower right-hand corner of the screen, alloing a participant to view the second event. Afthe second event, three buttons appeared.button, labeled “Repeat,” allowed a participantview the two events again. The other two butto

labeled “First Event” and “Second Event,” a

LY IS MORE 261

nea-astedsatgei-n-re-

tsenly

ly aenrdd-e-

aldn-ngot

rd

ben-d-hed

xtoftots,rdh

in-

erne

to

lowed participants to choose which of the twevents was a better example of the presenword.After the 12 isolated word meaning test tals, 12 embedded word meaning test trials wpresented using the same procedure. The eexperiment generally lasted about 40 min.

Design

There were four independent variables in texperiment. One was the presentation condit(i.e., individual word vs sentence), manipulatbetween participants. A second was the worder of the presented “sentences” (i.e., ObjeManner–Path vs Object–Path–Manner), amanipulated between participants. The thwas the word type that was tested in a given ttrial (i.e., object vs manner vs path), maniplated within participants. The final independevariable was the test type of a given test tr(i.e., isolated word meaning vs embedded wmeaning), also manipulated within participanThe dependent variable was the percentagecorrect responses in the test trials.

Results

The results of Experiment 1 are displayedFig. 4. An α level of .05 was adopted for aanalyses. An ANOVA with presentation condtion and word order as between-participant vables and word type and test trial type as withparticipant variables revealed a main effectpresentation condition, F(1, 108) = 8.73, p= .004,MSe = 1612.69. Participants who were presenwith words one at a time learned better than participants who heard entire sentences throuout. There was also a main effect of word tyF(2, 216) = 25.83, p < .001, MSe = 1023.75. Poshoc t tests revealed that path words were learbetter than both object words, t(111) = 2.03, p =.045, and manner words, t(111) = 7.02, p < .001,whereas object words were learned better tmanner words, t(111) = 4.94, p < .001. No othermain effects or interactions approached signcance (all ps > .10).

Discussion

The results of Experiment 1 revealed that p

l-miniature artificial language better when they

262

i

e

rd

ee,reatn

i

s

hegh aonnt a

theen-er

arethelarli-

t 2

di- in

d wordt–Path–

Manner). There were no significant effects involving either of these variables. Fifty percent correct would reflectr

were forced to learn one word at a time. In partular, participants who were initially presentewith only one word at a time and were only latpresented with additional words learned thowords better than participants who were psented with entire “sentences” of novel worthroughout learning. This finding is surprisinbecause participants who heard entire “stences” throughout could have adopted a stratof learning one word at a time. For exampleparticipant could have decided to learn the fiword in a sentence before moving on to the sond word. If a participant had adopted this stregy, he or she should have performed at leaswell as a participant who was forced to learn oword at a time by the computer. The finding thparticipants who were presented with ent“sentences” did not perform as well as the othgroup of participants is consistent with the “LeIs More” hypothesis. In particular, rather than atempting to learn the language one word at a timparticipants may have tried to learn the meaninof entire sentences. This strategy may have h

chance performance. Error bars reflect standard er

ten

KERSTEN AND EARLES

FIG. 4. Results of Experiment 1. These results are aggregated over the two types of test trials (isolatemeaning vs embedded word meaning) and over the two word orders (Object–Manner–Path vs Objec

oes.

dered their ultimate proficiency in the language

EXPERIMENT 2

The results of Experiment 1, although cons

t with the “Less Is More” hypothesis, ar

c-drsee-s

gn-gyastc-t-ase

atreerst-e,gsin-.

is-

somewhat counterintuitive. Participants in tsentence condition were provided with enouinformation that they could have executedstrategy similar to the one that was forced upparticipants in the individual word condition. Iparticular, they could have learned one word atime. The difference in performance between two conditions, however, suggests that the stence condition did not use this strategy. In ordto ensure that the results of Experiment 1 general in nature and do not merely reflect poor language learning strategies of a particuset of participants, it seemed important to repcate the results of Experiment 1. Experimenwas designed to provide such a replication.

Method

Participants

One hundred eighty undergraduates at Inana University participated in this experimentpartial fulfillment of introductory psychologycourse requirements.

Stimuli

The stimuli of Experiment 2 were identical tthose of Experiment 1 except for three chang

ors.

eFirst, for half of the participants, the two object

ninFro

neha

erei

n

odl

g

oi

toa

h

h

cbii

go

th

r-

isond

iceyedrdb-r-heda-

ntses

in

ri-sf-

dnre

fi-ns,

-

r-

er-

LESS REAL

words (ending in “-ju”) were distinguished othe basis of the body and head of the movcharacter rather than the body and legs.example, for a given participant the wo“geseju” may have always accompanied anject with the body and head depicted in Fig.whereas theappearance of the legs varied radomly. For theother half of the participants, thtwo object words were distinguished on tbasis of the body and legs of the moving charter and the head varied randomly, as in Expement 1. This manipulation was designed to twhether the advantage of the individual wocondition on manner of motion words was dpendent on prior learning of object wordswhich the parts responsible for those motio(i.e., the legs) were relevant.

Second, the manner of motion of the moviobject was made more salient. The mannermotion no longer involved just the motionsthe legs of a creature with respect to its boInstead, the creature as a whole now amoved in a manner that was consistent withleg motion. In particular, when the movincharacter angled its legs forward and backsynchrony, the character surged forward aslegs moved back, whereas it remained statiary as the legs were moved forward agaThus, the character had the appearancepulling itself forward in surges. In contraswhen the moving character angled its legs fward and back in opposite directions, the charter moved in a zig-zag, giving the appearanthat the backward pull of a leg not only movethe character forward but also pulled it to tside on which the leg was located.

Third, all participants were assigned tObject–Manner–Path sentence order. This stence order was chosen because it is most sistent with English word order, in which an oject noun typically precedes a manner of motverb, which in turn precedes a path prepositor verb particle (Talmy, 1985). This consistenwith English would be expected to most stronbenefit participants in the sentence conditiwho heard sentences with this structure throuout learning. The use of this sentence order

reflects a conservative test of the prediction the “Less Is More” hypothesis.

LY IS MORE 263

gor

db-1,-

ec-ri-std-

nns

goffy.soits

initsn-n.of

,r-c-

cede

een-on--

ononcylyn,

gh-us

Procedure

The procedure was identical to that of Expeiment 1.

Design

There were four independent variables in thexperiment. One was the presentation conditi(i.e., individual word vs sentence), manipulatebetween participants. A second was the choof attributes to distinguish object words (bodand legs vs body and head), also manipulatbetween participants. The third was the wotype that was tested in a given test trial (i.e., oject vs manner vs path), manipulated within paticipants. The final independent variable was ttest type of a given test trial (i.e., isolated wormeaning vs embedded word meaning), also mnipulated within participants. The dependevariable was the percentage of correct responin the test trials.

Results

The results of Experiment 2 are depictedFig. 5. An ANOVA with presentation conditionand object attribute as between-participant vaables and word type and test trial type awithin-participant variables revealed a main efect of presentation condition,F(1, 176)= 4.03,p = .046,MSe = 1934.00. In replication of Ex-periment 1, participants in the individual worcondition performed significantly better thaparticipants in the sentence condition. Thewas also a main effect of word type,F(2, 352)= 21.67,p < .001,MSe = 839.50. Post hoct testsrevealed that participants performed signicantly worse on tests of manner of motiowords than on tests of either object wordt(179)= 5.00,p < .001, or path words,t(179)=5.84,p < .001. Participants did not perform significantly differently on tests of object wordsand path words,t(179) = 1.02, p > .10. Therewere no other significant main effects or inteactions.

Discussion

The results of Experiment 2 again providsupport for the “Less Is More” hypothesis. Pa

ofticipants who were forced to learn one word ata time again learned those words better than

264

l

2ein

mrirhoh

tersoif

f togs.tod

better than participants who heard entire sen-

a-1n-inoi-i-

yf

t,

n

-t–

ed wordds (body

and legs vs body and head). There were no significant effects involving either of these two variables. Fifty per-

participants who heard entire sentencthroughout learning. Thus, adults who were immediately presented with the full complexity othe language had difficulty extracting the lowelevel components of the language, in particuthe meanings of the words.

Although the results of Experiments 1 andare consistent with the “Less Is More” hypothsis, they are also consistent with an alternatexplanation. In particular, participants in the idividual word condition may have learned beter not because they learned one word at a tibut rather because they learned object wofirst. Participants in the individual word condtions of Experiments 1 and 2 were always psented with object words before they had topportunity to learn words for the motions those objects. In contrast, participants in tsentence condition had the opportunity to leamotion words before object words, if they atempted to learn the words at the end of a stence first. It may be difficult, however, to leawords for the motions of unfamiliar objectKersten and Smith (manuscript submitted fpublication) have found that children have dficulty associating verbs with motions when thobjects carrying out those motions are unfam

cent correct would reflect chance performance. Erro

iar. If the adults in Experiment 1 had simila

es-

fr

ar

difficulty associating words with the motions ounfamiliar objects, they may have been ablelearn words for motions only after learninwords for the objects exhibiting those motionAs a result, participants who were forced learn object words first may have performe

r bars reflect standard errors.

KERSTEN AND EARLES

FIG. 5. Results of Experiment 2. These results are aggregated over the two types of test trials (isolatmeaning vs embedded word meaning) and over the choice of attributes to distinguish the two object wor

-ve-t-e,

ds-e-efe

rn-n-

n.r-eil-

tences at the beginning of learning.

EXPERIMENT 3

Experiment 3 was designed to test this alterntive hypothesis for the results of Experimentsand 2. Participants who were presented with etire sentences throughout learning were agacompared to participants who were forced tlearn one word at a time. In contrast to Experments 1 and 2, however, participants in the indvidual word condition were first presented withmotion words before being given the opportunitto learn object words. Thus, if the advantage othe individual word condition in Experiments 1and 2 was a result of learning object words firsthen the individual word condition should haveperformed no better than the sentence conditioin Experiment 3.

Participants in the sentence condition of Experiment 3 heard sentences of the form Objec

rManner–Path throughout learning. Participants

b

a

eo

s

a

e

c

s

r

a

h

u”d3

e

x

e

ct-as

mea

b-nei-t in-cttedo

lv-lv-

tontstocek-

icrl-

oes.i-

ay ats.rdd

ts.cesalre

gsng1.3d

LESS REAL

in the individual word condition, on the othehand, first heard one word along with each eveThis word referred to the path of the moving oject. One-third of the way through learning,second word accompanied each event. This wreferred to the manner of motion of the movinobject. Consistent with English grammar, thmanner word was presented prior to the pword accompanying each event. Thus, these pticipants heard sentences of the form MannPath during this stage of learning. Finally, twthirds of the way through learning, a third worwas added. This word referred to the moving oject in each event. Again consistent with Engligrammar, this object word was presented priorthe other two words accompanying each eveThus, these participants heard sentences ofform Object–Manner–Path during this stagelearning, just as did participants in the sentencondition. Because participants in the individuword condition were presented with words onindividual basis before hearing entire sentenchowever, the “Less Is More” hypothesis prdicted that these participants would learn betthan participants who heard entire sententhroughout learning.

A second aim of Experiment 3 was to tewhether a focus on small segments of languaencourages not only the acquisition of womeanings but also a component of the grammof the language, in particular the inflectionmorphology of the words in the language. Ain Experiments 1 and 2, the two exampleseach type of word in Experiment 3 shared tsame word ending. For example, object worended in “-ju,” whereas manner words ended“-gop.” This is analogous to English morphology (although much simpler and more reglar), in which object words often end in “-s(when they are pluralized) and manner wor(i.e., verbs) often end in “-ing.” Experimenttested whether participants were able to leathe inflectional morphology of the miniaturartificial language they were learning.

Morphological knowledge was tested in Eperiment 3 by presenting participants with novwords after they had completed the rest of theperiment and testing whether they noticed t

presence of familiar morphemes within thos

LY IS MORE 265

rnt.-

aordgisthar-r–-

db-htont.theofcealn

es,-

teres

tgedarlsofe

dsin--

s

rn

-elx-

he

novel words. For example, after learning objewords such as “geseju” and “mogaju,” participants were presented with new words such“yuboju” and “lemuju.” If participants noticedthat the earlier object words shared the morphe“-ju,” they would be expected to assume thatnew word with this ending also referred to the oject in an event. To test this, after being given oexample of a novel word (e.g., “yuboju”), particpants were presented with a second event thavolved either the same object or a different objeand were asked whether this event also depican example of that same word. A participant whnoticed the inflectional morphology of this newword would be expected to reject an event invoing a different object and accept an event invoing the same object. Because they were forcedfocus on small segments of language, participain the individual word condition were expectedbe more likely than participants in the sentencondition to make use of morphology when maing their choices.

Method

Participants

Forty-two undergraduates at Florida AtlantUniversity participated in this experiment foextra credit in a “Psychology of Human Deveopment” class.

Stimuli

The stimuli of Experiment 3 were identical tthose of Experiment 2 except for three changFirst, participants in the individual word condtion started learning the words at the end ofsentence first. These participants heard onlpath word along with each of the first 24 evenThese participants then heard a manner wofollowed by one of the previously presentepath words for each of the next 24 evenThese participants then heard entire sentenwith the order Object–Manner–Path for the fin24 learning events. Second, object words wedifferentiated on the basis of the body and le(rather than the body and head) of the movicharacter for all participants, as in Experiment

The final change to the stimuli of Experimentwas the addition of the induction trials at the en

eof the experiment. After the embedded word

v)

b

e

n

v

s

ti

n

i

tr

eo

e

hvel

an ofrerd-

ofofult,g

atheirex-ci-ew

d

aechla-wnsreyam-in-

then

2b-ri-bleectnts-

e)

The aggregated results of the isolated word

266 KERSTEN A

meaning test trials, participants were presenwith novel words, each accompanied by a noevent. Four new object words (ending in “-ju”four new manner words (ending in “-gop”), anfour new path words (ending in “-tig”) were presented. The novel events were created by coming a novel moving character, a novel manner, aa novel path. Four novel moving characters wcreated by switching the roles played by the stionary character and the moving character inevent such that a previously stationary characnow moved with respect to a previously movincharacter, which was now stationary. Two novmanners involved motions of the legs of a movicharacter up and down or left and right with rspect to its body. Two novel paths were createdpresenting the two possible paths that werepresented to a given participant during learning

Each induction trial involved two events. Ithe first event, a participant saw an event invoing one of the new objects, one of the new maners, and one of the new paths. Each such ewas accompanied by a novel word. In the secoevent in each trial, one of these three attributethe previous event was changed, whereasother two attributes remained the same. Tsame novel word was presented along withsecond event. A participant’s task was to decwhether this second event was also an examof the presented word. Each novel word was psented twice during the induction trials. In otrial involving a given word, the attribute (e.gmanner) that corresponded to the word’s end(e.g., “-gop”) was changed from the first eventthe second event. Thus, a participant wouldcorrect to say that the second event was noexample of the presented word. In the other tinvolving a given word, one of the other two atributes (e.g., object or path) was changed. Tha participant would be correct to say that the sond event was an example of the presented w

Procedure

The procedure was identical to that of Expiments 1 and 2 except for the addition of theduction trials. After the embedded word meaing test trials, participants were instructed tthey would be presented with examples of no

words. They would then be asked to decide,

ND EARLES

tedel,d-in-ndre

ta-antergelg

e-by

not.nlv-n-entndof

thehehedeplere-e

.,ngtobean

ialt-us,c-rd.

r-in-n-at

each word, whether a second event was alsoexample of the same word. As an examplehow they would make this decision, they weinstructed that if they were to hear a novel woin English such as “morping,” they would probably have expectations about the meaningthat word on the basis of their knowledge English. They were instructed that as a resthey would probably accept an event involvina different object but not an event involvingdifferent action as being another example of word “morping.” They were instructed that thetask would essentially be the same as this, cept that they would have to make their desions on the basis of their knowledge of the nlanguage they had learned.

Participants clicked on a button labele“Begin” to start the induction trials. After thefirst event in each trial, participants clicked onbutton labeled “Next Event” to proceed to thsecond event. After the second event in eatrial, three buttons appeared. One button, beled “Repeat,” allowed a participant to viethe two events again. The other two buttowere labeled “Yes” and “No.” Participants weinstructed to click on the “Yes” button if thethought that the second event was also an exple of the novel word, whereas they were structed to click on the “No” button if theythought that the event was not an example of novel word. There were a total of 24 inductiotrials, 8 for each type of word.

Design

The design differed from that of Experimentin that the choice of attributes to distinguish oject words was no longer an independent vaable. In addition, a second dependent variawas added. This was the percentage of corrchoices in the induction trials. The independevariables for this dependent variable were preentation condition (individual word vs sentencand word type (object vs manner vs path).

Results

Isolated Word Meaning and Embedded WordMeaning Test Events

formeaning and embedded word meaning test

L

aneo

dno

h

r-n

n

t

the-sf-tedd,ctt-

he,i-thei-

re-e-d

ntn-

LESS REAL

events of Experiment 3 are depicted in Fig.An ANOVA with presentation condition asbetween-participant variable and word type atest trial type as within-participant variables rvealed a significant main effect of presentaticondition, F(1, 40) = 4.42, p = .042, MSe =1403.52. Participants learned better in the invidual word condition than in the sentence codition. There was also a significant effectword type,F(2, 80) = 14.60,p < .001, MSe =775.17. Post hoct tests revealed that patwords were learned better than both objewords, t(41) = 5.24, p < .001, and mannewords,t(41) = 3.81,p < .001. There was no significant difference between object words amanner words,t(41) = 0.79, p > .10.

The only other significant effect was the iteraction of word type and test trial type, F(2,80) = 4.95, p = .009, MSe = 457.22. Post-hoct tests revealed that participants performed bter on tests of object words in the embeddword meaning test events than in the isolaword meaning test events, t(41) = 2.10, p = .042.Participants averaged 68% (SD = 25%) in theembedded word meaning test events, wherthey averaged 56% (SD = 35%) in the isolated

word meaning test events. There were no signic

d

performance. Error bars reflect standard errors.

Y IS MORE 267

6.

d-n

i--f

ct

d

-

et-eded

eas

if-

for path words, t(41) = 1.31, p > .10, or for man-ner words, t(41) = 1.26, p > .10.

Induction Trials

The percentages of correct responses ininduction trials were computed via the following steps. First, for each word type (e.g., wordending in “-gop”) we computed the number ocorrect rejections of events in which an attribute (e.g., manner) was changed that was relato the morphology of that word type. Seconwe added this number to the number of correacceptances of events in which a different atribute (e.g., object or path) was changed. Tmaximum score for a given word type was 8with 4 possible correct rejections and 4 possble correct acceptances. Thus, we computedpercentage correct for a given word type by dviding the total number of trials correct by 8and multiplying by 100. These percentages adepicted in Fig. 7. An ANOVA on these percentages with presentation condition as a btween-participant independent variable anword type as a within-participant independevariable revealed a significant effect of presetation condition,F(1, 40)= 6.47,p = .015,MSe

= 405.26. Participants in the individual wor

phol-

3. Thet chance

ant differences between the two test trial typescondition were more sensitive to the mor

FIG. 6. Results of the isolated word meaning and embedded word meaning test events in Experimentresults depicted here are aggregated over these two types of test trials. Fifty percent correct would reflec

268

ed

lc

e

s

,sn

np

nei

n-r-nd

seyn- in-ft 3u-le-dsntstee

edaltson

f

ogy of the presented words when making thchoices. The choices of participants in the invidual word condition were consistent with thmorphology of the presented word significantmore often than would be expected by chant(20) = 2.50, p < .05. Participants in the sentence condition, on the other hand, performno differently than would be expected bchance,t(20) = −0.77,p > .10. There was also amain effect of word type,F(2, 80)= 16.08,p <.001, MSe = 586.25. Post hoct tests revealedthat participants made more correct responwith path words than with either mannewords,t(41) = 3.52,p = .001, or object wordst(41) = 5.58, p < .001, whereas participantmade more correct responses with manwords than with object words,t(41) = 2.20,p <.05. Path was apparently more salient thanther the manner or the object in an event, athus participants were less willing to accepath changes than either manner changes orject changes. Similarly, manner was apparenmore salient than objects, and thus participawere less willing to accept manner changthan to accept object changes. The interact

FIG. 7. Results of the induction trials o

of presentation condition and word type did noapproach significance,F(2, 80)= 0.87,p > .10,MSe = 586.25.

d

iri-eye,

-d

y

esr

er

ei-dt

ob-tlytss

ont

Discussion

The results of Experiment 3 were again cosistent with the “Less Is More” hypothesis. Paticipants learned both the word meanings amorphology of a miniature artificial languagebetter when they were initially forced to focuon small pieces of that language than when thwere allowed to process larger segments of laguage. This was true even though participantsthe individual word condition were not presented with object words until the final set olearning events. Thus, the results of Experimenare inconsistent with the hypothesis that the sperior learning of participants in the individuaword conditions of Experiments 1 and 2 was bcause they were forced to learn object worfirst. Instead, the presentation of small segmeof language seems to result in superior ultimalearning of that language, regardless of thmeanings those segments convey.

Participants in the individual word conditionnot only learned the meanings of the presentwords better, they also learned the inflectionmorphology of those words better. Participanin the present task were not forced to focus individual morphemes, as children may be force

Experiment 3. Error bars reflect standard errors.

KERSTEN AND EARLES

to do as a result of internal processing limita-tions (Newport, 1988, 1990). Given the simplic-

s

ar-e

ina-e-nt

LESS REAL

ity of the morphology of the miniature artificialanguage, however, participants who focuson individual words were apparently able discern the underlying morphology of thowords. Participants who were presented w

entire sentences throughout learning, howev

tsn

e

ioe

o f

i

asisin-

henglern-bu-n,

-l

ticm-en--

heer-xdr-”inar-n-

n-n

dce

al.rse’s-rd

were apparently less able to discern the undeing morphology.

GENERAL DISCUSSION

The present research provides evidence is consistent with Newport’s (1988, 1990) “LeIs More” hypothesis for the critical period ilanguage learning. Adult participants learnedminiature artificial language better when thwere initially presented with only small segments of language rather than hearing the lguage in its full complexity from the beginninof learning. This may be analogous to the wachild shows better ultimate competence of a laguage than does an adult because the chprocessing limitations force him or her to prcess small segments of language. This procing of small segments of language may encoage a language learner to acquire the lowlevel elements of a language that convey meings. In addition, it may foster the learning rules for combining those meaning elements,lowing a language learner access to the combinatorial richness of a language.

The language employed in these experimewas much simpler than real-world natural laguages. In particular, its vocabulary consistedonly six words, its word order was very rigidand its morphology was entirely redundant wword order in marking the thematic and syntatic role of a word. There is reason to believe ththe present results will generalize to more naral languages, however, at least with regardthe acquisition of word meaning and morphlogy. With regard to the acquisition of wormeaning, the present learning task was simto word learning in natural languages in ththere was a very large hypothesis space to csider when trying to determine the referent o

word. In particular, because there was indepeent variation on eight different attributes of thevents (not counting random variation on t

a

LY IS MORE 269

ledtoeither,rly-

hats

ay-

an-gy an-ld’s-ss-

ur-estan-f

al-ull

ntsn- of,thc-at

tu- too-dilaraton-f and-

starting positions of the two characters), a pticipant’s task was to determine which of th8192 events generated by the possible combtions of these attributes fit into the category dscribed by a given word. Because a participasaw at most 36 examples of a given word, it wnot possible to narrow down the hypothesspace regarding the meaning of a word to a sgle hypothesis that was consistent with tgiven information. Instead, learners had to bribiases into the learning situation in order to ruout certain classes of hypotheses, just as leaing biases are necessary to acquire the vocalary of natural languages (see, e.g., Markma1990; Mitchell, 1980; Quine, 1960).

With regard to morphology, the present miniature artificial language was similar to naturalanguages in that the words of a given syntaccategory shared the same inflection. For exaple, just as manner-of-motion words in thpresent artificial language always had the iflection “-gop” attached to them, verbs in English often occur with the inflection “-ing” at-tached to them. The key difference between tmorphology of the present artificial languagand that of natural languages was that mophology was perfectly redundant with syntain predicting the syntactic category of a worin the artificial language. For example, manneof-motion words not only always ended in “-gop,but also always occurred in a given positiona sentence (e.g., the second position for a pticipant assigned the Object–Manner–Path setence order). In contrast, morphology and sytax are not typically so perfectly redundant inatural languages.

This redundancy between morphology ansyntax may be expected to benefit the sentencondition to a greater extent than the individuword condition in the learning of morphologyBillman (1989) has demonstrated that learneof an artificial language are more likely to noticthe predictiveness of a particular cue to a wordsyntactic category if it covaries with other predictive cues. In the sentence condition, woorder cues as well as morphological cues to

e

heword’s syntactic category were present through-out learning. Thus, there were multiple predictive

ne

a

i

o

l

r

a

m

r

el

e- in-

p-

h2).iritech toch didL

an-endy,e-

h aed,ofr-tsageer-

es ininee,ior

icbe-asen-nd-

llytonte

-t al.

270 KERSTEN A

cues to a word’s syntactic category, potentiafacilitating the learning of the predictivenesseach cue. In contrast, only morphological cuwere initially present for the individual wordcondition. Thus, there were no other cues itially available to facilitate the learning of thpredictiveness of the morphology of a word. Itthus possible that the advantage of processsmall segments of language for the learningmorphology may be even larger with a more nural language in which syntactic cues are notperfectly redundant with morphology.

Implications for Adult Second LanguageLearning

Although it would be premature to makstrong recommendations for educational currula, the present research does have implicatfor the optimal conditions under which to leaa second language as an adult. It suggeststotal immersion in a language without pristudy of the components of the language is an effective strategy for learning a languaduring adulthood. Although children clearlearn language through total immersion, thmay be effective only because children’s pcessing limitations force them to focus on smsegments of language. The present researchgests that if adults are immersed in a languwithout prior instruction, they may have difficulty learning the meanings of the lower levcomponents of language such as morpheand words. This may in turn hinder their learing of the rules for combining these componeto create more complex meanings. Immersmay be more useful for language learning flowing training on the lower level componenof language, just as participants in the individuword conditions of the present experimenwere given training on individual words befobeing confronted with the full complexity of thlanguage they were learning.

Current educational practice appears to largely consistent with this suggestion. In Canawhere French immersion programs have bimplemented on a massive scale, nearly all (i.e., grade 6 or later) immersion programs volve prior experience in traditional elementa

French as a second language (EFSL) cour

ND EARLES

llyofes

i-

isingoft-so

eic-onsrnthatr

notgeyiso-allsug-ge

-eles

n-ntsionol-tsaltsee

beda,en

atein-ry

often starting as early as kindergarten (Gensee, 1983). In contrast, early (i.e., beginningkindergarten) immersion programs typically involve no prior instruction in French. One excetion to this rule of providing EFSL instructionprior to late immersion is a program in BritisColumbia described in Shapson and Day (198Children in this program received 60% of theinstruction in grades 6 and 7 in French, despthe fact that they had had no prior Frenclasses. Children in this program were foundultimately perform just as well on tests of Frencomprehension and French language arts aschildren who had received 1 to 2 years of EFSinstruction prior to immersion.

There are a couple of reasons why no advtage to prior French instruction may have beobserved in the Shapson and Day (1982) stuhowever. First, 1 to 2 years is not a very long priod of time for EFSL study. Even children witthis EFSL experience may not have attainedmastery of the lower level components of thFrench language prior to immersion. Indeelate immersion students with this amount prior EFSL instruction have been found to peform more poorly than early immersion studenon tests of French comprehension and languarts (Day & Shapson, 1988), whereas the diffences between the two groups are sometimsmaller or even nonexistent in school districtswhich late immersion students are enrolled EFSL courses starting in kindergarten (Genes1981). Second, children who were given no prexperience in EFSL were screened for academachievement and interest in learning French fore being allowed to enter the program, wherethere was no screening procedure for childrwho were given prior EFSL instruction. Children who were given no prior EFSL instructiomay thus have overcome this deficit through aditional motivation and/or ability, resulting inno differences between groups. More carefucontrolled experiments are clearly needed better understand the effects of prior foreiglanguage instruction on performance in a laimmersion setting.

In addition to prior instruction in the components of a language, the results of Cochran e

ses,(1999) suggest an alternative manipulation that

an

a

c

o

ta

rnn

u

t

t

m

eee

d ofeeasren-

n-si-onnty

enr-n-e

ts,er-il-

-reodl-ntss-n-

bere-tsk.

her-

atyee.-tse-intswr-

LESS REAL

may also result in successful second langulearning. In particular, immersion in a secolanguage may be a useful strategy if processlimitations are imposed on adults through tadministration of a concurrent task. Cochret al. had participants perform a secondary tof counting tones while they performed the pmary task of learning ASL verbs. These partipants showed better learning of certain aspeof ASL than did participants who were nasked to perform a secondary task. One shobe cautious, however, in interpreting these sults. As mentioned above, the overall perforance of participants who were assigned a secoary task was no better than that of participawho were not assigned a secondary task; in fthe performance of participants assigned a ccurrent task was worse when the learning perwas brief. Thus, it has yet to be demonstrated a concurrent task results in superior ultimcompetence with a language.

Second, it is not clear whether Cochran et aresults would generalize to spoken languagespossible explanation for why participants weable to learn ASL while simultaneously peforming a concurrent task is that the tasks volved separate modalities. In particular, leaing ASL was a visual task whereas countitones was an auditory task. Given the evidethat two auditory tasks interfere with one aother to a greater extent than do one auditand one visual task (e.g., Allport, Antonis, Reynolds, 1972), it may be much more difficto learn a spoken language given the same tractor task. Of course, visual distractor tascould be devised, but it is unclear whether effects of a visual distractor task on an auditoprimary task would be identical to the effectsan auditory distractor task on a visual primatask. Further research is needed to understhe generalizability of this finding.

Predictions for Child Language Learning

Adults in the present research who were imersed in the full complexity of a language hdifficulty acquiring the words of that languagThis task is similar to the one faced by childrlearning their first language. Children are giv

no training on the language they are to learn,

LY IS MORE 271

gedingheansk

ri-i-ctstuldre-m-nd-

ntsact,on-iodhatte

l.’s. Arer-in-n-gce

n-ory&lt

dis-ksheryofryand

-ad.nn

rather they hear it in all of its complexity ansomehow manage to acquire the meaningsthe words and the rules for combining thoswords to form more complex meanings. Thfact that children succeed at this task, whereadults have difficulty, suggests that children aoperating under a different set of internal costraints than are adults.

Because of these differences in internal costraints between children and adults, it is posble that the external constraints imposed adults in the present task would have differeeffects when imposed on children. Learning badults in the present task was improved whthere were limitations on the amount of infomation that was presented. Thus, external costraints functioned to improve the performancof adults. Because of their internal constrainhowever, these external constraints would phaps have no impact on the performance of chdren. In particular, if children were initially presented with small segments of language befobeing totally immersed, they may perform ndifferently than children who were immersefrom the beginning of language learning. Chidren may be forced to focus on small segmeof language because of limitations on procesing resources, and as a result external costraints on the amount of information that can processed may have no further impact. This pdiction could be tested with children of differenages using simpler variants of the present taIf Newport’s (1988, 1990) “Less Is More” hy-pothesis is correct, one would expect that tolder children were, the more closely their peformance would approximate that of adults.

Conclusions

The present research provides evidence thadults who are immersed in the full complexitof a language have difficulty extracting thbasic meaning elements of that languagAdults who are forced through external constraints to initially process only small segmenof the language show better ultimate comptence at the language. These external constramay resemble the internal constraints that allochildren to learn a language. A better unde

butstanding of how these internal constraints

N

o ate

272 KERSTEN A

change with age may allow for the creation

ng

-

-

ne

d

ns

oa-

a

p

f

s

)i

n

a-

ge

f.-

d

f-of a

s

nif-

rd

--i-

-

iri-

ge

in

i-w

of

l

learning environments that are optimally cofigured for the acquisition of a second languain adulthood.

REFERENCES

Allport, D. A., Antonis, B., & Reynolds, P. (1972). On the division of attention: A disproof of the single channel hypothesis.Quarterly Journal of Experimental Psychology, 24,225–235.

Bialystok, E., & Hakuta, K. (1999). Confounded age: Liguistic and cognitive factors in age differences for sond language acquisition. In D. P. Birdsong (Ed.), Sec-ond language acquisition and the critical periohypothesis(pp. 161–181). Mahwah, NJ: Erlbaum.

Billman, D. (1989). Systems of correlations in rule and cagory learning: Use of structured input in learning sytactic categories. Language and Cognitive Processe,4, 127–155.

Birdsong, D. (1999). Introduction: Whys and why nots the critical period hypothesis for second language quisition. In D.P. Birdsong (Ed.), Second language acquisition and the critical period hypothesis(pp. 1–22).Mahwah, NJ: Erlbaum.

Birdsong, D., & Molis, M. On the evidence for maturationconstraints in second-language acquisition. Journal ofMemory and Language, 44,235–249.

Cochran, B. P., McDonald, J.L., & Parault, S.J. (1999). Tsmart for their own good: The disadvantage of a surior processing capacity for adult language learneJournal of Memory and Language, 41,30–58.

Curtiss, S. (1977).Genie: A psycholinguistic study oa modern day “wild child.” New York: AcademicPress.

Day, E. M., & Shapson, S. (1988). A comparison studyearly and late French immersion programs in BritiColumbia. Canadian Journal of Education, 13, 290–305.

Elman, J. L. (1993). Learning and development in neunetworks: The importance of starting small. Cognition,48,71–99.

Emmorey, K., Bellugi, U., Friederici, A., & Horn, P. (1995Effects of age of acquisition on grammatical sensitity: Evidence from on-line and off-line tasks. AppliedPsycholinguistics, 16,1–23.

Emmorey, K., & Corina, D. (1990). Lexical recognitionin sign language: Effects of phonetic structure amorphology. Perceptual and Motor Skills, 71,1227–1252.

Flege, J. E. (1999). Age of learning and second-languspeech. In D. P. Birdsong (Ed.), Second language acquisition and the critical period hypothesis(pp. 101–131). Mahwah, NJ: Erlbaum.

Flege, J. E., Yeni-Komshian, G. H., & Liu, S. (1999). Agconstraints on second-language acquisition. Journal ofMemory and Language, 41,78–104.

D EARLES

f-e

-

-c-

te--

fc-

l

ooe-

rs.

ofh

ral

.v-

d

ge

Genesee, F. (1981). A comparison study of early and lsecond language learning. Canadian Journal of Behav-ioral Sciences, 13,115–128.

Genesee, F. (1983). Bilingual education of majority-languachildren: The immersion experiments in review. AppliedPsycholinguistics, 4, 1–46.

Goldowsky, B. N., & Newport, E. L. (1993). Modeling theeffects of processing limitations on the acquisition omorphology: The less is more hypothesis. In E. VClark (Ed.), The Proceedings of the Twenty-Fourth Annual Child Language Research Forum(pp. 124–138).Stanford, CA: Center for the Study of Language anInformation.

Johnson, J. S., & Newport, E. L. (1989). Critical period efects in second language learning: The influence maturational state on the acquisition of English assecond language. Cognitive Psychology, 21,60–99.

Kersten, A. W. (1998a). A division of labor between nounand verbs in the representation of motion. Journal ofExperimental Psychology: General, 127,34–54.

Kersten, A. W. (1998b). An examination of the distinctiobetween nouns and verbs: Associations with two dferent kinds of motion. Memory and Cognition, 26,1214–1232.

Kersten, A. W., & Smith, L. B. (manuscript submitted forpublication). Attention to novel objects during verblearning.

Markman, E. M. (1990). Constraints children place on womeanings. Cognitive Science, 14,57–77.

Mayberry, R. I., & Eichen, E. (1991). The long-lasting advantage of learning sign language in childhood: Another look at the critical period for language acquistion. Journal of Memory and Language, 30,486–512.

Mitchell, T. M. (1990). The need for biases in learning generalizations (Technical Report CBM-TR-117). NewBrunswick, NJ: Rutgers Univ. Press.

Newport, E. L. (1988). Constraints on learning and therole in language acquisition: Studies of the acquistion of American Sign Language. Language Sciences,10, 147–172.

Newport, E. L. (1990). Maturational constraints on langualearning. Cognitive Science, 14,11–28.

Oyama, S. (1979). The concept of the sensitive perioddevelopmental studies. Merrill-Palmer Quarterly, 25,83–102.

Quine, W. V. O. (1960). Word and object. Cambridge, MA:MIT Press.

Rohde, D. L. T., & Plaut, D. C. (1999). Language acquistion in the absence of explicit negative evidence: Hoimportant is starting small? Cognition, 72,67–109.

Shapson, S. M., & Day, E. M. (1982). A comparison study three late immersion programs. Alberta Journal of Ed-ucational Research, 28,135–148.

Snow, C. E., & Hoefnagel-Höhle, M. (1978). The critica

e period for language acquisition: Evidence from

second language learning.Child Development, 49,1114–1128.

LESS REALL

Talmy, L. (1985). Lexicalization patterns: Semantic struture in lexical forms. In T. Shopen (Ed.), Language typo-logy and linguistic description. Vol. 3: Grammaticacategories and the lexicon(pp. 57–149). Cambridge,UK: Cambridge Univ. Press.

Weber-Fox, C. M., & Neville, H. J. (1999). Functional neural subsystems are differentially affected by delays

Y IS MORE 273

c-

l

-in

second language immersion: ERP and behavioral evi-dence in bilinguals. In D. P. Birdsong (Ed.), Secondlanguage acquisition and the critical period hypothesis(pp. 23–38). Mahwah, NJ: Erlbaum.

(Received October 5, 1999)(Revision received July 5, 2000)