The Structural Connectivity Underpinning Language Aptitude

21 Pages·2012·0.73 MB·English

LanguageLearning ISSN0023-8333
The Structural Connectivity Underpinning
Language Aptitude, Working Memory, and
IQ in the Perisylvian Language Network
HuadongXiang
DondersInstituteforBrain,CognitionandBehaviour
DanDediu
DondersInstituteforBrain,CognitionandBehaviourandMaxPlanckInstitutefor
Psycholinguistics
LeahRoberts
UniversityofYorkandMaxPlanckInstituteforPsycholinguistics
ErikvanOort
DondersInstituteforBrain,CognitionandBehaviour
DavidG.Norris
DondersInstituteforBrain,CognitionandBehaviour,ErwinL.HahnInstituteforMagnetic
ResonanceImaging,andMIRAInstituteforBiomedicalTechnologyandTechnicalMedicine
PeterHagoort
DondersInstituteforBrain,CognitionandBehaviourandMaxPlanckInstitutefor
Psycholinguistics
Inthisarticle,wereporttheresultsofastudyontherelationshipbetweenindividual
differences in language learning aptitude and the structural connectivity of language
pathways in the adult brain, the first of its kind. We measured four components of
languageaptitude(vocabularylearning;soundrecognition;sound-symbolcorrespon-
dence;andgrammaticalinferencing)usingtheLLAMAlanguageaptitudetest.Spatial
workingmemory,verbalworkingmemoryandIQwerealsomeasuredascontrolfactors.
CorrespondenceconcerningthisarticleshouldbesenttoLeahRoberts,CentreforLanguage
LearningResearch,DepartmentofEducation,UniversityofYork,Heslington,York,YO105DD,
UK.Internet:[email protected]
LanguageLearning62:Suppl.2,September2012,pp.110–130 110
(cid:2)C 2012LanguageLearningResearchClub,UniversityofMichigan
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
Diffusion Tensor Imaging was employed to investigate the structural connectivity of
languagepathwaysintheperisylvianlanguagenetwork.Regressionanalysissuggested
significantcorrelationsbetweenmostofthesebehaviouralmeasuresandthestructural
connectivity of certain language pathways, that is, grammatical inferencing and the
BA45-andBA46-Temporalpathway,sound-symbolcorrespondenceandtheinterhemi-
sphericBA45pathway,vocabularylearningandtheBA47-Parietalpathway,IQandthe
BA44-andBA-47Parietalpathways,theBA47-Temporalpathwayandinterhemispheric
BA45pathway,spatialworkingmemoryandtheinterhemisphericBA6pathwayandthe
BA47-Parietalpathway,andverbalworkingmemoryandtheBA47-Temporalpathway.
Thesefindingsprovidefurtherinsightsintotheneuralunderpinningsofthevariationin
languageaptitudeofhumanadultsandarediscussedinrelationtorelevantfindingsin
theliterature.
Introduction
Itisuncontroversialthatadultsdifferenormouslyintheirhowsuccessfulthey
areatforeignlanguagelearningandoneofthereasonsbehindthisdifference
has been suggested to be variations in language learning aptitude (or talent)
(seeDeKeyser,2000,2012).Althoughthetopicoflanguagelearningaptitude
hasbeenresearchedanddiscussedingreatdepthinthesecondlanguage(L2)
acquisition literature (see Sparks, 2012; DeKeyser, 2012; Robinson, 2005,
for overviews), we know almost nothing about the neural underpinnings of
the variation in language learning aptitude. One recent study has revealed a
correlationbetweenthestructuralconnectivityofcertainlanguagepathwaysin
thebrain(theperisylvianlanguagepathwaysconnectingBroca’sandWernicke’s
territories) and performance in a language-learning related experimental task
in adults (theCalifornian Verbal Learning Test, a testof wordrecall) (Catani
et al., 2007). Specifically, the authors found that better performance on the
semanticassociationtaskwasassociatedwithdirectconnectionsbetweenthe
inferior frontal and the posterior temporal regions in the right hemisphere
and interestingly, with bilateral representation of language networks, rather
than extreme left-lateralization. This suggests an important role for the right
hemisphereinwordretrieval(seealsoHabib,Nyberg,&Tulving,2003),andan
advantageforthistypeofsemantictaskforbilateralrepresentation.Importantly
forthecurrentstudy,thefactthattheCatanietal.demonstratedaheterogeneous
pattern of lateralization of perisylvian pathways in their large sample (50) of
healthyadultspointstotheneedtoexamineindividualvariationinthestructural
connectivityoflanguagepathwaysinthebrain.Inthestudywereportbelow,
thefocusonindividualdifferencesinperformanceonlanguageaptitudetasks
111 LanguageLearning62:Suppl.2,September2012,pp.110–130
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
andtheirrelationshipswiththestructuralconnectivityoflanguagepathwaysin
thebrain,anunder-researchedtopic.
LanguageLearningAptitude
Languageaptitudeanditsassociationwithforeignlanguagelearningsuccess
hasbeeninvestigatedformanyyears,mainlyfocusingontutoredlearnersand
more recently also on learners in naturalistic settings (e.g., DeKeyser, 2000;
Robinson, 2007). In the main, language learning aptitude is assumed to be
a predominantly innate and somewhat fixed “gift” for learning languages; a
cognitivecapacitythatsomeargueislargelyindependentofIQ,motivationand
personality factors (e.g., Do¨rnyei & Skehan, 2003). Most language learning
aptitudetasksarebasedontheworkofCarroll(e.g.,Caroll,1981;Carroll&
Sapon,1959),andincludetestsofphonemecodingability(establishingsound-
symbol correspondences), grammatical sensitivity, word (semantic) learning
andinductivelearningability(e.g.,ModernLanguageAptitudeTest[MLAT];
Carroll & Sapon, 1959; Defense Language Aptitude Battery [DLAB]; Peter-
son & Al-Haik, 1976). This is also the case for the aptitude test used in the
currentstudy,theLlamaLanguageAptitudeTest(LLAMA;Meara,2005).Al-
thoughnotstandardized,thistesthasbeenusedwithsuccessinotherlanguage
learningstudies(e.g.,Abrahamsson&Hyltenstam,2008),oneofitsstrengths
being that is language neutral (being based either on nonsense or extremely
rare languages), as opposed to the majority of other aptitude tests which the
participants must undertake in their first or dominant language. As well as
measuringtheparticipants’languagelearningaptitude,wealsoexaminedthe
impactofthecontrolvariablesspatialworkingmemory,verbalworkingmem-
oryandIQ,giventhatdifferencesinsuccessfulsecondlanguagelearningmay
alsoberelatedtogeneralcognitiveabilitiesandskillsinthefirstlanguage(L1)
(seeSparks,2012).
MappingBrainConnectivity
Tostudylanguagepathwaysinthebrainandtheirstructuralconnectivity,Dif-
fusionTensorImaging(DTI)wasemployed;atechniquethatcanprovideclear
mapsofbrainconnectivity(Figure1).Morespecifically,DTIproducesimages
ofbiologicaltissuesweightedwiththelocalmicrostructuralcharacteristicsof
water diffusion. Therefore, using water molecules as a probe, the technique
allowsforthetracingofstructuralconnectionsbetweenbrainregionsinvivo,
in a non-invasive manner and provides quantitative measures of brain white
matter organization, offering a highly detailed picture of the architecture of
braintissue(LeBihanetal.,2001).
LanguageLearning62:Suppl.2,September2012,pp.110–130 112
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
Figure1 DTI(DiffusionTensorImaging)iscapableoftracingstructuralconnections
betweenbrainregionsinvivo.
Belowtheresultsofthecurrentstudyarepresented,inwhichweexamine
therelationshipbetweenindividualdifferencesinperformanceonabatterylan-
guagelearningaptitudeteststothestructuralconnectivityoflanguagepathways
inthebrain.
TheCurrentStudy
Method
Participants
Thedatareportedinthepresentstudyincludestwosamples.Sample1consisted
of15participants(3males)whohadjustgraduatedfromhighschool.Sample
2contained20participants(8males)whowerecollege/graduatestudents.The
participantsinbothsampleswerehealthy,right-handedadultsbetweentheages
of19–26years(SD=1.78).Allparticipantsprovidedwritteninformedconsent
andthestudywasconductedaccordingtoinstitutionalguidelinesofthelocal
ethicscommittee(CMOprotocolregionArnhem-Nijmegen,Netherlands).The
majority of the participants were native speakers of German or Dutch, with
Englishasasecondlanguage(seeAppendix).
BehavioralTests
Languagelearningaptitudetests
TheLLAMALanguageAptitudeTest(Meara,2005)wasusedtogatherdata
onthelanguagelearningaptitudeoftheparticipants.Itincludesfourcomputer-
basedsubtestswhicharedesignedtobeindependentofthefirstlanguage(L1)
ofthetest-taker.LLAMA_Bisavocabularylearningtask,whichmeasuresthe
113 LanguageLearning62:Suppl.2,September2012,pp.110–130
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
abilitytolearnreasonablylargeamountsofvocabularyinarelativelyshorttime.
LLAMA_Disasoundrecognitiontask,whichtestsifparticipantscanrecognize
short stretches of previously encountered spoken language. LLAMA_E is a
sound-symbolcorrespondencetask,whichrequiresparticipantstoworkoutthe
relationship between the sounds and the writing system based on previously
heard syllables and their transliterations in an unfamiliar alphabet. Finally,
LLAMA_Fisagrammaticalinferencingtestthatrequiresparticipantstowork
outthegrammaticalrulesthatoperateintheunknownlanguage.
Testsofgeneralcognitiveability.Threemoretestsofgeneralcognitiveabil-
itywerecarriedout.A20-minuteversionoftheRavenAdvancedProgressive
Matrices (APM) Test was employed to measure IQ.1 This timed version has
beenshowntobeanadequatepredictoroftheuntimedAPMscore(Hamel&
Schmittmann,2006).Anewstandardcomputerizedversionofreadingspantest
wasusedtotestverbalworkingmemory(vandenNoort,Bosch,Haverkort,&
Hugdahl,2008).Thisversionusessentencematerialsandrequiresparticipants
toreadandattempttounderstandsentenceswhilerememberingthefinalword.
A spatial working memory test which includes one forward and one back-
wardsub-testwasalsoconducted.Thistestisacomputerizedadaptationbased
on the Spatial Span subtest of the Wechsler Memory Scale—Third Edition
(Wechsler,1997).
Inordertounderstandtheinterrelationshipsbetweenthevariousbehavioral
tasks,weperformedaPrincipalComponentsAnalysis(PCA).Thisprocedure
extractsanumberoforthogonalcomponentsequaltothenumberofvariables,
withthefirstcomponent(PC1)explainingmostofthevarianceinthedataset,
followedindecreasingorderoftheamountofvarianceexplainedbyPC2,PC3,
etc.Thus,byanalysingtheloadingsofeachofourtestsonPC1,PC2,etc,we
canbetterunderstandthenatureofourtasks.
DTIparameters
ParticipantsinbothsampleswerescannedonaSiemens3TTrioscannerwith
thefollowingparameters:
Sample 1: echo EPI (echo-planar imaging) sequence, repetition time
(TR) = 10500 ms, echo time (TE) = 94 ms, 110 diffusion directions de-
finedevenlyacrossthespherewithadiffusionweightingofb=1000s/mm2,
70slices,matrix=110×110,FOV(fieldofview)=220×220mm2,isotropic
voxelsof2.0mm.3
Sample2:echoEPIsequence,TR=13000ms,TE=101ms,256directions
atb=1500and24directionsatb=0,70slices,matrix=256×256,FOV=
256×256mm2,isotropicvoxelsof2.0mm3.
LanguageLearning62:Suppl.2,September2012,pp.110–130 114
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
DTIdataanalysis
DTIdatasetswerepreprocessedbytheDTItoolboxdevelopedattheDonders
InstituteforBrain,CognitionandBehaviour(Zweieretal.,2009).Theproba-
bilistictractographytechniqueoftheFMRIBSoftwareLibrary(FSL)(Behrens
etal.,2003)wasemployedtotracethepossiblepathwaysbetweentheseedre-
gionintheleftparietallobeandeachofthefoursubregionsofBroca’scomplex
(i.e.,leftBA6,BA44,BA45andBA47,asthetargetregion),respectively.To
investigatethelateralityofthesepathways,thesametractographywasapplied
totherighthomologousregions.Thepossiblepathwaysbetweenposteriortem-
porallobe(seed)andeachofthefoursubregionsofBroca’scomplex(target)
weretracedinthesameway.Seedregionsweredrawnbasedonourprevious
study on the topographical functional connectivity pattern of the perisylvian
languagenetwork(Xiang,Fonteijn,Norris,&Hagoort,2010).Brainregionsin
theparietallobeorposteriortemporallobethatshowedsignificantfunctional
connectivitytoBroca’scomplexweretakenasregionsofinterest(ROI).Voxels
intheseROIswerethendilated(i.e.,extended)witharadiusof4mmunderthe
restrictionofnotexceedingtheboundariesoftheparietalorposteriortemporal
lobe. The purpose of the dilation is to better accommodate the endpoints of
fiberbundles.TargetregionsweredefinedaccordingtotheAAL(Automated
AnatomicalLabeling)template(Tzourio-Mazoyeretal.,2002).Anexclusion
mask of the sagittal midline was also implemented in the tractography to re-
movepathwaysthatcrossintotheotherhemisphere.Thepathwayswerevisual-
izedwiththehelpofMRIcron(http://www.cabiatl.com/mricro/mricron/index.
html).
Two DTI measures were adopted to describe the structural connectivity
properties of the traced pathways: the number of streamlines (Nstr) and the
fractionalanisotropy(FA).Theprobabilistictractographyrepetitivelysamples
from the distributions of voxel-wise principal diffusion directions, each time
computing a streamline from the distribution on the location of the true fiber
connections.Nstrcountshowmanysuchstreamlinesrunfromthepredefined
seed region to the predefined target region (Behrens et al., 2003). Nstr is an
estimationofthenumberofrealfibersbasedontheartificialreconstructionof
these fibers. It should be kept in mind that Nstr (represented by a parameter
called “waytotal” in FSL) is not a straightforward measure of tract integrity.
It has been suggested that normalized values of waytotal should be used to
avoidapossibletractabilitybiascausedbysubjectmotionorscannernoise.FA
is a scalar measure of anisotropy. It is thought to reflect fiber density, axonal
diameter,andmyelinationinwhitematter(Basser&Pierpaoli,1996).Toavoid
subjectivethresholding,aprobability-weightedaverageFAwascalculatedfor
115 LanguageLearning62:Suppl.2,September2012,pp.110–130
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
each pathway and subject using the following equation (derived from: Hua
etal.,2008):
(cid:2) (cid:3)(cid:2)
MeanFA= (Pr∗FA) Pr (1)
i i i
WherePr istheprobabilityoftheithvoxeltobepartofthereconstructed
i
tract, empirically decided by the number of streamlines that pass through the
ithvoxeldividedbythetotalnumberofstreamlines.
Toassessthestructurallateralityofeachconnection,lateralizationindices
(LI)werecalculatedforbothNstrandFAusingthefollowingequation:
LI=(Left−Right)/(Left+Right) (2)
Toseewhethertheabsolutedifferencebetweenleftandrightconnections
also influence the behavioral measures, an absolute lateralization index (LA)
wasalsocalculatedusingthefollowingequation:
LA=Left−Right (3)
Thus for both LI and LA, a more positive value indicates “more left-
lateralized”andamorenegativevaluemeans“moreright-lateralized.”
Regressionanalysisontherelationshipbetweenbehavioralmeasures
andstructuralconnectivityprofiles
StepwiseregressionanalysisimplementedinSPSS16.0(SPSS,Inc.,Chicago
IL)wasusedtoinvestigatewhethereachofthebehavioralmeasuresthatwere
gathered could be explained by the connectivity profiles of certain pathways.
Behavioralmeasureswereregressedontothefollowing:
- Numberofstreamlines(Nstr)
- LateralisationindecesofNstr(LI_Nstr)
- LateralisationindexofNstr(LA_Nstr)
- thesummationofleftandrightNstr(Sum_Nstr).
Sum_Nstrwasincludedasaregressortocontrolforbrainsizeandthebias
oftractabilitycausedbysubjectmotionandscannernoise.Thegenderofthe
LanguageLearning62:Suppl.2,September2012,pp.110–130 116
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
Table1 Loadingsofthebehavioralmeasuresontheprincipalcomponents
Component
1 2 3
Vocabularylearning .657 .214 −.273
Soundrecognition .438 −.712 −.108
Sound-symbolcorrespondence .618 .386 −.111
Grammaticalinferrencing .832 −.094 −.328
IQ .341 .667 .119
Spatialworkingmemory .349 .116 .816
Verbalworkingmemory .587 −.436 .413
subject (gender) and the sample to which the subject belongs (sample) were
enteredascontrolregressorsaswell.
Thesamestepwiseregressionanalyseswerecarriedouttoinvestigatethe
relationship between the behavioral measures and the fractional anisotropy
(FA).
Results
PCAAnalysisoftheBehavioralMeasures
ThefirstthreePrincipalComponents(PCs)accountedfor67%ofthevariance.
The first PC (PC1) explains 32.6% of the variance. The second PC (PC2)
explains 19.4% and the third (PC3) 15% of the variance. The corresponding
loadingsareshowninTable1.
PC1haspositiveloadingsonall7behavioralmeasuresindicatingthatall
thesebehavioralmeasuresmaybeimportantforforeignlanguagelearning.PC2
hadpositiveloadingsonvocabularylearning,sound-symbolcorrespondence,
spatial working memory, and IQ (the largest loading) and negative loadings
on sound recognition, verbal working memory and grammatical inferencing.
Thissuggeststhatvocabularylearning,establishingsound-symbolcorrespon-
dencesandspatialworkingmemoryaremorerelatedtoIQthanverbalworking
memory, the ability to recognise sounds, and deducing grammatical patterns
in the input. PC3 distinguished those abilities assumed to relate specifically
tolanguageaptitudefrommoregeneralcognitiveabilities,withnegativeload-
ingsonvocabularylearning,soundrecognition,sound-symbolcorrespondence,
grammaticalinferencingandverbalworkingmemoryandpositiveloadingson
IQandspatialworkingmemory.
117 LanguageLearning62:Suppl.2,September2012,pp.110–130
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
Figure2 (a)Sagittalviewoftracedpathwaysbetweenparietal/posteriortemporallobe
and 3 subregions (BA44, 45 and 47) of IFG (Broca’s complex). (b) Left or Right
lateralized BA6-Temporal pathways projected on the MRICRON brain template. L:
Lefthemisphere;R:righthemisphere;A:anteriorbrain;P:posteriorbrain.
LanguagePathways
DTI fiber tractography discovered robust structural connections for all seed-
targetpairsin97%oftheparticipants(Figure2).Whileconnectionsbetween
theparietal/posteriortemporallobeandBA6/BA44areobservedtoliemainly
inthedorsalpartofthebrain,connectionsbetweenthetwolobesandBA47go
throughtheventralpart.ConnectionsbetweenthetwolobesandBA45areseen
inbothdorsalandventralpartsofthebrain.Interhemisphericpathwayswere
showninFigure3;allofthesepathwaysgothroughthecorpuscallosum.
TheRelationshipBetweenBehavioralMeasuresandStructural
ConnectivityintheLanguagePathways
The step-wise regression analyses found the most optimal models for vocab-
ulary learning (for Nstr, adjusted R2 = 0.14, p = 0.03), sound recognition
(forNstr,adjustedR2=0.37,p=.04),sound-symbolcorrespondence(forFA,
adjusted R2 = 0.25, p = .003), grammatical inferencing (for Nstr, adjusted
R2 = 0.36, p = 0.002), IQ (for Nstr, adjusted R2 = 0.76, p < .0001; for FA,
LanguageLearning62:Suppl.2,September2012,pp.110–130 118
Xiangetal. LanguageAptitudeandLanguagePathwaysintheBrain
Figure 3 Interhemispheric pathways of BA6, BA44, BA45, BA47, parietal lobe and
posteriortemporallobeprojectedontheMRICRONbraintemplate.L:lefthemisphere;
R:righthemisphere;A:anteriorbrain;P:posteriorbrain.
adjusted R2 = 0.24, p = .005), spatial working memory (for Nstr, adjusted
R2 = 0.32, p = .002; for FA, adjusted R2 = 0.23, p = 0.006), and verbal
working memory (for Nstr,adjusted R2 = 0.34, p = .003). Table 2 shows the
detailsofthebestpredictorssuggestedbyeachofthesemodels.
ItshouldbenotedthatthesignificanceofthepredictorofNstrLA47P(ab-
solutelateralizationindexofBA47-parietalpathway)forvocabularylearning
did not survive a Bonferroni correction. We still report this result because it
isrelevanttopreviousfindings(seeDiscussionformoredetails).Allthesig-
nificant results reported are Bonferroni corrected unless otherwise specified.
Figure 4 shows the correlations between the four language aptitude tests and
theirbestpredictors.
Discussion
BehavioralMeasures
Forthe7behavioralmeasures,thefirstthreePrincipalComponents(PCs)ex-
plained most of the variance. PC1 was the largest component (32.6%) and
represents the agreement between all these behavioral measures, suggesting
thatageneralabilitymightbeimportanttoL2acquisition.PC2explains19.4%
andseemstosuggestthatvocabularylearning,sound-symbolcorrespondence
andspatialworkingmemoryaremorecloselyrelatedtogeneralIQthansound
recognition and verbal working memory. Finally, PC3 explains 15% of the
varianceanddistinguishesthetasksspecificallydesignedtotapintolanguage
aptitude (vocabulary learning, sound recognition, sound-symbol correspon-
denceandgrammaticalinferencing)frommoregenericmeasures(IQ,spatial
andverbalworkingmemory).
AnatomicalConnectionsinthePerisylvianLanguageNetwork
Our fiber tracking results are generally in agreement with previous findings
on language pathways (Catani, Howard, Pajevic, & Jones, 2002; Friederici,
119 LanguageLearning62:Suppl.2,September2012,pp.110–130