Post on 27-Oct-2019
MisalignmentofofflineandonlinemeasuresinRussianrelativeclause
processing
IyaKhelmPrice&JeffreyWitzel
UniversityofTexasatArlington
• Sentencecomprehensionisthecomputationofmeaning.
• Howdoesthemanipulationofsymbols(words,morphemes,phrases)producemeaningfulinterpretationsinourminds?
• Arethementalgrammarandlanguageprocessingpartsofthesamesystem?
Responsetypes
Onlineresponses:• usetime-sensitivemeasures• usuallyasthestructureunfolds• complementedbyaccuracyoncomprehensionquestionspresentedafter
thesentencedisappears
Offlineresponses:• judgmentsmadewithnotimelimits• usuallyafterthefullstructureispresented
Onlinevs.offlineresponses
• Whatstagesofcomputationdotheseofflineandonlineresponsesreflect?
• Onlineresponsesshowintermediatestepsinbuilding grammaticalrepresentations,offline judgments reflectdifferentstagesofcomputation inthesamesystem(Lewis&Phillips, 2015;Phillips&Lewis,2013).
• Attemptstoestablishhowgrammarandlanguageprocessingcouldbepartof thesamecognitivesystemhavecalledforcomparisonsofofflineandonline responsestothesameinput.
Thepresentstudy
• investigatedhowonlineandofflineresponsescomplementeachotherbyexaminingtheprocessingofRussianrelativeclauses(RCs)with
- offlinemeasures:acceptabilityjudgmenttaskcomplementedbyacorpusanalysis
- onlinemeasures:2self-pacedreadingexperimentscomplementedbyacomprehensionquestionaccuracymeasure
The housewife, who upset the old lady, lay on the couch in the living room.
1(completelyunacceptable) - 2(not fullyacceptable) -3(somewhatacceptable)- 4(acceptable)- 5(completelyacceptable)
Offlinemeasures(score1-5):AcceptabilityJudgment
Onlinemeasures(inms):Self-pacedreading
---------------------------------------------------------------------------------------------------.
7
Onlinemeasures(inms):Self-pacedreading
Housewife, ----------------------------------------------------------------------------------------.
8
Onlinemeasures(inms):Self-pacedreading
-----------who.nom --------------------------------------------------------------------------------.
9
Onlinemeasures(inms):Self-pacedreading
-------------------after walk ---------------------------------------------------------------------.
10
Onlinemeasures(inms):Self-pacedreading
------------------------------old_lady.acc --------------------------------------------------------.
11
Onlinemeasures(inms):Self-pacedreading
-------------------------------------------really -------------------------------------------------.
12
Onlinemeasures(inms):Self-pacedreading
--------------------------------------------------upset -------------------------------------------.
13
Onlinemeasures(inms):Self-pacedreading
---------------------------------------------------------with_her story,---------------------------.
14
Onlinemeasures(inms):Self-pacedreading
------------------------------------------------------------------------lay------------------------.
15
Onlinemeasures(inms):Self-pacedreading
----------------------------------------------------------------------------on couch in living_room.
16
Onlinemeasures:Comprehensionquestion
Did the housewife upset the old lady with her story?
NO YES
17
Onlinemeasures:Comprehensionquestion
CORRECT
18
RelativeclauseprocessingStudiesonrelativeclauses(RCs)inanumberoflanguageshaveshownthatobject-extractedRCs(ORCs)aremoredifficulttoprocessthansubject-extractedRCs(SRCs).
Subject-extractedRC(SRC):a.Thereporter1[thatt1attackedthesenator]admittedtheerror.Object-extractedRC(ORC):b.Thereporter1[thatthesenatorattackedt1]admittedtheerror.
AnumberofstudieshaveshownthatORCsaremoredifficulttoprocessnotjustinEnglish,butalsoinotherlanguages,includingChinese,Dutch,Hungarian,Japaneseandothers(Staub,2010;Traxler etal.,2002;Traxler etal.,2005;MacWhinney &Pleh,1998;Mak etal.,2002;Lin&Bever,2006;Miyamoto&Nakamura,2003).
ModelsoftheSRC-ORCprocessingasymmetry:• Expectation-basedaccounts(Hale,2001;Levy,2008;Reali &Christiansen,
2007;MacDonald&Christiansen,2002)– NOMINALRCs(withadescriptiveNPinsidetheRC): ORCsMOREdifficultthanSRCs– PRONIMINALRCs(withapronouninsidetheRC): ORCsMOREdifficultthanSRCs
• Memory-basedaccounts(King&Just,1991;Gibson, 1998,2000;Gordonetal.,2001;Johnson etal.,2011)
– integrationcostsdependondistance,numberandtypesofNPstobeheldinworkingmemory,NPsimilarity
• Structure-basedaccounts(MacWhinney &Pleh,1998;Lin&Bever,2006;Clifton&Frazier,1989)
(Staub,2010;Levyetal.,2013)– acombinationofthesefactors
Relativeclauseprocessing
Thereporter1 [thatthesenatorattackedt1]admittedtheerror.Thereporter1 [thatyouattackedt1]admittedtheerror.
20
Thepresentstudy
• Inmanylanguages,wordorderdifferencesbetweenSRCsandORCsmakeitdifficulttotestamongtheseaccounts.
SRC:a.Thereporter1 [thatt1attacked thesenator]admittedtheerror.ORC:b.Thereporter1 [thatthesenator attacked t1]admittedtheerror.
• RussianallowsSRCsandORCstohavethesamelexicalmaterialinthesamelinearorder(withcase-markingdistinguishingbetweenRCtypes).
21
Thesentencesofinterest(seeHandout)
Thesentencesofinterest(seeHandout)
23
SRC
ORC
Thesentencesofinterest(seeHandout)
24
• anNPargument intervenedbetweenthemodified nounandtheRCverb:NP-O/S+V
Thesentencesofinterest(seeHandout)
25
• thesamenumberofNPargumentswasavailableforintegrationattheRCverb,acrossthesamelineardistance,inbothSRCsandORCs
Thesentencesofinterest(seeHandout)
26
• theinfluenceofstructuralexpectationswasinvestigatedbyusingdifferentNPtypes-- descriptiveNPsandpronouns -- insidetheembeddedclause
RussiancorpusanalysisPercentages of SRCs and ORCs with different word orders depending on
embedded NP type
RussiancorpusanalysisPercentages of SRCs and ORCs with different word orders depending on
embedded NP type
RussiancorpusanalysisPercentages of SRCs and ORCs with different word orders depending on
embedded NP type
RussiancorpusanalysisPercentages of SRCs and ORCs with different word orders depending on
embedded NP type
RussiancorpusanalysisPercentages of SRCs and ORCs with different word orders depending on
embedded NP type
Anofflineacceptabilityjudgmentstudy
Mean rating scores for RCs with different word orders inside the embedded clause
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Mea
n ra
ting
scor
es
Anofflineacceptabilityjudgmentstudy
Mean rating scores for RCs with different word orders inside the embedded clause
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Mea
n ra
ting
scor
es
Anofflineacceptabilityjudgmentstudy
Mean rating scores for RCs with different word orders inside the embedded clause
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Mea
n ra
ting
scor
es
Anofflineacceptabilityjudgmentstudy
Mean rating scores for RCs with different word orders inside the embedded clause
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Mea
n ra
ting
scor
es
Anofflineacceptabilityjudgmentstudy
Mean rating scores for RCs with different word orders inside the embedded clause
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Mea
n ra
ting
scor
es
Corpusfrequenciesvs.Acceptabilityscores
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Anofflineacceptabilityjudgmentstudy
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRCControl CC ORCControlCC
Default(SVO) Scrambled(OVS)
Mean rating scores for CCs with different word orders inside the embedded clause.
Mea
n ra
ting
scor
es
• Howdoestheofflinedatacorrespondtotheonlinemeasures?
Onlinevs. Offlinemeasures(NominalRCs)
• Someindicationsofonlineprocessingdifficultypatternedwiththeofflinemeasures,butothersdidnot.
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRCControl CC ORCControlCC
Default(SVO) Scrambled(OVS)
RCsentences- AcceptabilityJudgment CCcontrolsentences- AcceptabilityJudgment
OnlineSelf-pacedReadingExperiment1(NominalRCs)
Expectationeffects=>alignment
Memoryeffects=>misalignment
Onlinevs. Offlinemeasures(PronominalRCs)
• Someindicationsofonlineprocessingdifficultypatternedwiththeofflinemeasures,butothersdidnot.
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRCControl CC ORCControlCC
Default(SVO) Scrambled(OVS)
RCsentences- AcceptabilityJudgment CCcontrolsentences- AcceptabilityJudgment
OnlineSelf-pacedReadingExperiment2(PronominalRCs)
Expectationeffects=>alignment
Memoryeffects=>misalignment
Onlinevs. Offlinemeasures
Online= Offline=> alignmentForsentencesthatweredispreferredinofflinejudgmentsandlessfrequentinthecorpus,longerreadingtimeswererevealedatthefirstunexpectedword-- theembedded-clauseNP.
Online≠ Offline=> misalignmentTheRCvs.CCpatternofexpectationeffectsdoesnotcontinueonthroughoutthesentence.ComparableintegrationcostsforSRCsandORCsat/aftertheRCverbwhendistanceandthetypesofintegratedelementswereheldconstant.
• AlthoughORCswithdescriptiveNPswerejudgedofflineasmoreacceptablethanSRCs,andwerefoundtobemorefrequentinthecorpus…
2.5
3
3.5
4
4.5
Nominal Pronominal Nominal Pronominal
SRC ORC
Default(VO/SV) Scrambled(OV/VS)
Comprehensionquestionmeasure=>misalignment
RCsentences- AcceptabilityJudgment RCsentences- Corpus Analysis
Comprehensionquestionmeasure=>misalignment
• Late-stagecomprehensiondifficultywasrevealedfornominalORCsentencesinparticular
• Thissuggeststhatsimilarity-basedinterference(Gordonetal.,2001,2002,2004,2006),combinedwithORCstructuralprocessingdifficulty,alsoinfluencesprocessesrelatedtoretrievingandassigningthematicrolestoNPsduringRCprocessing.
Interpretationofresults
• Theseresultsthussuggestthat– intermediatestepsinonlinestructurebuildingrelatedtoexpectation-
based processingcorrespondtoofflinemeasures,– whereasonlineprocessingdisruptionsandcomprehensiondifficulty
thatappeartorelatetomemory demandsdonot.
• Onlinemeasuresreflectintermediatestagesofcomputationthatofflinemeasuresarenotabletoshow.
• Thesedifferencesbetweentheonlineandofflineresultsmightbetakentoreflectdifferentstagesofcomputationinasinglecognitivesystemforlanguageprocessing.
Thankyou!
48
ReferencesClifton, C.,Jr.,&Frazier,L.(1989).Comprehending sentenceswithlongdistancedependencies. InG.Carlson&M.Tanenhaus (Eds.), Linguisticstructureinlanguage processing (pp.273–317).Dordrecht,TheNetherlands:Kluwer.Gordon, P.C.,Hendrick, R.,&Johnson, M.(2001).Memoryinterferenceduringlanguageprocessing.JournalofExperimentalPsychology: Learning,Memory,&Cognition, 27,1411–1423.Gordon, P.C.,Hendrick, R.,&Johnson, M. (2004). Effectsofnounphrasetypeonsentencecomplexity. Journal ofMemoryandLanguage,51,97-114.Gordon, P.C.,Hendrick, R.,Johnson, M.,&Lee,Y.(2006).Similarity-based interferenceduringlanguagecomprehension:Evidencefromeyetrackingduringreading.Journal ofExperimentalPsychology: Learning,Memory,andCognition,32(6), 1304-1321.Gordon, P.C.,Hendrick, R.,&Levine,W.H.(2002).Memory-loadinterferenceinsyntacticprocessing.Psychologicalscience,13(5),425-430.Forster,K.I.,&Forster,J.C.(2003).DMDX:AWindows displayprogramwithmillisecond accuracy.BehaviorResearchMethods,Instruments,&Computers, 35,116-124.Hale,J.(2001).Aprobabilistic earlyparserasapsycholinguistic model.In ProceedingsoftheNorthAmericanChapteroftheAssociation forComputational Linguistics, (pp.159–166).Pittsburgh,PA:Association forComputational Linguistics.Johnson, M.L.,Lowder,M.W.,&Gordon, P.C.(2011).Thesentencecomposition effect:Processingofcomplexsentencesdepends ontheconfigurationofcommonnounphrasesversus unusual nounphrases.Journal ofExperimentalPsychology:General,4, 707-724.Just,M.A.,Carpenter,P.A.,&Woolley, J.D.(1982).Paradigmsandprocesses inreadingcomprehension. JournalofExperimentalPsychology:General,3,228–238.King,J.,&Just,M.A.(1991).Individual differencesinsyntacticprocessing: Theroleofworkingmemory.JournalofMemoryandLanguage, 30,580–602.
ReferencesLewis,S.&Phillips, C.(2015).Aligninggrammaticaltheoriesandlanguageprocessingmodels.JournalofPsycholinguisticResearch44.27–46.Levy,R.(2008).Expectation-based syntacticcomprehension.Cognition,106, 1126–1177.Levy,R.,Fedorenko, E.,&Gibson, E.(2013).Thesyntacticcomplexity ofRussian relativeclauses.JournalofMemoryandLanguage, 69, 461-495.Lin,C.C.,&Bever,T.G.(2006).Subjectpreferenceintheprocessing ofrelativeclauses inChinese.InWCCFL 25Proceedings,ed.D. Baumer,D.Montero,andM.Scanlon, pp.254-260.Somerville, MA:Cascadilla ProceedingsProject.MacWhinney, B.,&Pléh,C.(1998).Theprocessing ofrestrictiverelativeclausesinHungarian.Cognition, 29, 95–141.Mak,W.M.,Vonk, W.,&Schriefers,H.(2002).Theinfluenceofanimacyonrelativeclauseprocessing.Journal ofMemoryandLanguage, 47, 50-68.Miller,G.A.,&Chomsky, N.(1963).Finitary models oflanguageusers.InR.Duncan,R.Bush, andE.Galanter (Eds.),Handbook ofMathematicalPsychology, 419-492.NewYork:Wiley.Miyamoto,E.T.,&Nakamura,M.(2003).Subject/object asymmetriesintheprocessing ofrelativeclausesinJapanese.InWCCFL22Proceedings,ed.G.Garding andM.Tsujimura, pp.342-355.Somerville,MA:Cascadilla Press.Reali F.,&ChristiansenM.H.(2007).Processingofrelativeclausesismadeeasierbyfrequencyofoccurrence.JournalofMemoryandLanguage, 57,1–23.RussianNationalCorpus. (2003-2012).300millionwords.1800s-2000s. Availableonline athttp://www.ruscorpora.ruPhillips, C.&Lewis,S.(2013) Derivationalorderinsyntax:Evidenceandarchitecturalconsequences. StudiesinLinguistics6.11-47.Slabakova,R.(2008).Meaninginthesecondlanguage (Vol.34).WalterdeGruyter.Staub,A.(2010).Eyemovements andprocessing difficulty inobjectrelativeclauses.Cognition, 116,71–86.Traxler,M.J.,Morris, R.K.,&Seely,R.E.(2002).Processingsubject andobjectrelativeclauses:Evidencefromeyemovements.JournalofMemoryandLanguage, 47(1), 69-90.Taxler,M.J.,WilliamsR.S.,Blozis S.A.,&MorrisR.K.(2005).Workingmemory,animacy, andverbclassintheprocessingofrelativeclauses.JournalofMemoryandLanguage, 53,204-224.