Learner corpora and Second Language Acquisition research AnkeLüdeling CLiF,Bloomington,June2016 1/189 Table of Contents 1 Corpusdesignandcollection Corpusdesign: samples,learner-specificissues Hands-onsession1 2 ErrorAnnotation 3 Otherannotationlayers 4 Quantitativeanalysis 5 Summary 2/189 Second Language Acquisition (SLA) research “Thetwomain,welldocumentedfindingsofSLAresearchofthepastfewdecadesareas follows: secondlanguageacquisitionishighlysystematic secondlanguageacquisitionishighlyvariable Althoughthesetwostatementsmightappearcontradictoryatfirstsight,theyarenot. Thefirstoneprimarilyreferstowhathasbeencalledtherouteofdevelopment(the natureofthestagesalllearnersgothroughwhenacquiringthesecondlanguage-L2). Thisrouteremainslargelyindependentofboththelearner’smothertongue(L1)andthe contextoflearning(e.g. whetherinstructedinaclassroomoracquirednaturallyby exposure). Thesecondstatementusuallyreferstoeithertherateofthelearningprocess (thespeedatwhichlearnersarelearningtheL2),ortheoutcomeofthelearningprocess (howproficientlearnersbecome),orboth. Weallknowthatbothspeedoflearningand rangeofoutcomesarehighlyvariablefromlearnertolearner: somedomuchbetter muchmorequicklythanothers.” (Myles2000) 3/189 SLA and learner corpora Corpusevidenceshouldbeusedtoinformacquisitionframeworksandmodels. Learnercorpusdata(ifcollectedandpreparedaccordingtotherelevant principles)canbeusefulforthestudyof systematicity variation. cf.,amongmany,manyothers,Bygate2009;Granger2002;Granger2009;Johansson2009; LüdelingandWalter2010;Vyatkinaetal.2015;WulffandGries2010;Römeretal.2014,and manyarticlesinGrangeretal.2015 4/189 Learner corpora Learnercorporacontaintextsfromlearnersofasecond/foreignlanguage(L2). Thesetextscanbe spoken(signaland/ortranscription),written,multimodal producedbylearnersofanyproficiencylevelandanyL1 producedfreelyortask-based longitudinalorcross-sectional “raw”orannotated Note: Althoughitishighlyrelevantforacquisitiontheory,inthissummerschoolwewillnot makeadifferencebetweenSecondLanguageAcquisition(untutored)andForeignLanguage Acquisition(tutored). ThereisalargebodyofresearchonL3etc. –hereIwilluseL2foranyLn. (Andwearenotinterestedinfirstlanguageacquisitioncorpora.) Foracomprehensivelistoflearnercorporaseethe“Corporaaroundtheworld”pagemaintainedby theCentreforEnglishCorpusLinguisticsatLouvainlaNeuve, https://www.uclouvain.be/en-cecl-lcworld.html. 5/189 Learner data Learnercorporaprovideone typeofevidenceinthestudyofacquisition. Itcan (andshould)becomplementedbyothertypesofevidence(experimental, behavioral,…). Corpusdatacanbeusedto exploreavarietyorphenomenon findhypothesesthatcanthenbetestedinothercorporaorwithother methods as“experimental”datathatisusedinhypothesistesting (careful! moreonthislater) informmodels Oncombiningdifferentmethodsinacquisitionstudies,cf. EllisandSimpson-Vlach(2009),Wulff etal.(2009),andLittré(2015). Statisticalmodellingisactuallyoneofthemostinterestingmethodsinstudyinglearnerdata. We donothavetimetogointothisherebutseeGries(2015b)andGriesandDeshors(2015). 6/189 Mantra <mantra> Corpusdesignandcorpusannotationdependprimarilyontheresearchquestion. Corpusarchitectureandformatsdependongeneralconsiderations(availability, transparency,re-usability,reproducability). </mantra> 7/189 Table of Contents 1 Corpusdesignandcollection Corpusdesign: samples,learner-specificissues Hands-onsession1 2 ErrorAnnotation 3 Otherannotationlayers 4 Quantitativeanalysis 5 Summary 8/189 Sampling | Corpus design statisticalaspects Typically,onewantstoanalyzeapopulationthatistoobigfor exhaustiveanalysisandtherefore,onehastodrawasample. Samplescanbeopportunistic,stratified,orrepresentative. The samplingstrategydeterminesthepossibilitiestoextrapolate. content Whichtextsshouldthelearnercorpuscontain? Whichsampling parametersarerelevant? Thediscussionhereisbriefandsuperficial. FormoreonthestatisticalissuesofsamplingseeBiber (1993)andOakes(1998). SeeKilgarriff(2005),Evert(2006),andGries(2010)onnon-randomness withincorpora(dispersion,burstiness). SeeGilquin(2015)onlearnercorpusdesign. 9/189 firstwehavetooperationalizewhat wemeanby“coveredbyforest” wecannottraveleverywhere–how shouldweproceed? ! extrapolationfromthese opportunisticsampleswouldlead toinvalidresults Corpus design | Sampling strategies Researchquestion: Howmuchof Germanyiscoveredbyforest? 10/189