MainQuestion:Areexperimentaltasksagoodmeasureofnext-mentionlikelihood?1) Metacognitivejudgments2) Storycontinuations
Results
SandraA.Zerkle&JenniferE.Arnold
Discourseconstraintsonreferentialpredictability:Isthesubjectpredictable?
Background• Thegrammaticalsubjectofasentenceisatopicalposition(Ariel,1990;Givón,1983;Groszetal.,1995),andthisconstrainshowweproducepronouns(Fukumura&vanGompel,2010;Kehleretal.,2008;Stevensonetal.1994)andinterpretpronouns(Gernsbacher,1990;Gordonetal.,1993).
• Cantopicalitybedefinedintermsofpredictability?Dopeoplehaveanexpectationforthesubjecttobementionedagaininnaturaldiscoursesandexperimentaltasks?
Supportforpredictability–measuresoffrequencyinrealdiscourseshowthatsubjectsarementionedmoreoftenthanobjects/obliques:
Evidenceagainstpredictability– experimentaltasksmeasuringnext-mentionlikelihood:• Peopletendtotalkaboutthecause.
• Peopledonottendtotalkaboutthesubjectifitisn’tthecause.
Ariel,M.(1990).AccessingNoun-PhraseAntecedents.vArnold,J.E.(1998).Referenceformanddiscoursepatterns(Doctoraldissertation,StanfordUniversity).v Arnold,J.E.(2001).Theeffectofthematicroleonpronounuseandfrequencyofreferencecontinuation.DiscourseProcesses.v Brocher,A.,Chiriacescu,S.,&vonHeusinger,K.(2016).Effectsofinformationstatus...DiscourseProcesses.v Frank,M.,&Goodman,N.(2012).Predictingpragmaticreasoninginlanguagegames.Science.v Fukumura,K.,&VanGompel,R.(2010).Choosinganaphoricexpressions:Dopeopletakeintoaccountlikelihoodofreference?JML.v Gernsbacher,M.A.(2013).Languagecomprehensionasstructurebuilding.v Givón,T.(1983).Topiccontinuityindiscourse:Aquantitativecross-languagestudy(Vol.3).v Gordon,P.,Grosz,B.,&Gilliom,L.(1993).Pronouns,names,andthecenteringofattentionindiscourse.CognitiveScience.v Grosz,B.,Weinstein,S.,&Joshi,A.(1995).Centering:Aframeworkformodelingthelocalcoherenceofdiscourse.Comp.linguistics.v Kehler,A.,Kertz,L.,Rohde,H.,&Elman,J.(2008).Coherenceandcoreferencerevisited.Journalofsemantics.v Kehler,A.,&Rohde,H.(2013).Aprobabilisticreconciliationofcoherence-driven¢ering-driventheories...TheoreticalLinguistics.v Nappa,R.,&Arnold,J.E.(2014).Theroadtounderstandingispavedwiththespeaker’sintentions...Cognitivepsychology.v Rohde,H.,&Kehler,A.(2014).Grammaticalandinformation-structuralinfluencesonpronounproduction.LCN.v Stevenson,R.J.,Crawley,R.A.,&Kleinman,D.(1994).Thematicroles,focusandtherepresentationofevents.LCP.
FundedbyNSFGrant1651000toJ.Arnold
Contact:[email protected]
65%74%
0%20%40%60%80%
100%
1-sentence 3-sentence
%su
bjectm
entio
nedfirst Exp.4:pasttense/PPphrase
SubjectsAREpredictable
81%
35%
10%0%
20%
40%
60%
80%
100%
GazetoSubject GazeNeutral GazetoNon-Sub
%su
bjectcho
senne
xt
Exp.1:metacognitivejudgmentsSubjectsNOTpredictable
65%50%
26%
66% 61%44%
0%
20%
40%
60%
80%
100%
GazetoSubject GazeNeutral GazetoNon-Sub
%su
bjectm
entio
nedfirst
Exp.2:storycontinuationSubjectsNOTpredictable
1-sentence3-sentence
Methods:allparticipantsnativeEnglishM-TurkworkersExperiments1&2(N=67,N=91)• Fragmentsofvideostimuli(Nappa&Arnold,2014),‘XwithY’construction• Narrator:“ThisstoryisaboutPandaBearandPuppy.ThisisPandaBear
(placesanimal),andthisisPuppy(placesanimal).PandaBearishavinglunchwithPuppy.”
• Narratorgazedwhenutteringpronoun:Subject(A),Neutral(B),Non-Subject(C)• Here,videoendsaftergazebutbeforepronoun• Exp.1:nextmentionpreference:“Whodoyouthinkshewilltalkaboutnext?”• Exp.2:storycontinuation:“Typeinacontinuationtothestory”• 2A:onesentencelong;2B:threesentenceslong
Experiments3&4(N=22,N=27)• Writtenstimuli(nomanipulationofgaze);bothexperimentstestedstory
continuation,withlength(1-vs.3-sentence)manipulatedbyblock• Exp.3:presenttense:“AnaiscleaningupwithLiz.”• Exp.4:pasttense,PPphrase:“AnawascleaningupwithLizinthekitchen.”
32%
55%
0%20%40%60%80%
100%
1-sentence 3-sentence%su
bjectm
entio
nedfirst
Exp.3:presenttenseSubjectsNOTpredictable
42%
10%
0%
20%
40%
60%
80%
100%
subjectreferents obj-of-PPreferents
%ofa
llfirstre
ferences Arnold(2001):corpusanalysis
73%
27%
0%
20%
40%
60%
80%
100%
Subject Object/Oblique
%to
talreferen
ces
Arnold(1998):textanalyses
Q2:Doesgazeaffectpredictabilityjudgments?
Q3:Doestheexperimentaltaskchangeperformance?
Conclusions• Subject-mentioninexperimentsonlymatchescorpusfrequencyinpast-tensenarratives
• èExperimentaltasksareNOTagoodmeasureofnext-mentionpredictability
• Inmetacognitive&storycontinuationtasks,theremaybeanexpectationof“fairness”:PandaBearwasthetopicofthelastsentence,sonowPuppyshouldbementioned.
• Taskdemandsmodulateperformance:• Moresubjectsmentionedinlongerdiscourse(Brocheretal.,2016)• Moresubjectsmentionedinpasttense/PPphraseconstructions
• Raisesquestionsabouttheuseofsimpleexperimentaltaskstoestimatetheprobabilityofreference(e.g.,Frank&Goodman,2012)
Experiments1&2:Subjectandgazepredictability
Experiments3&4:Subjectpredictabilityinshortvs.longcontinuations
Aresubjectcharactersjudged
tobemorepredictable?
NO
Aregazed-atcharactersjudged
tobemorepredictable?
YESp<.0001
Aresubjectcharactersmore
likelytobementionednext?
NO
Aregazed-atcharactersmore
likelytobementionednext?
YESp=0.002
Aresubjectcharactersmore
likelytobementionedinlonger
continuations?
YESp=0.0068
Aresubjectcharactersmore
likelytobementionedinlonger
continuations?
YESp=0.028
Numericaltrend:subjectsmorelikelyinlongernarratives
Aresubjectcharactersmore
likelytobementionednext?
NO
Aresubjectcharactersmore
likelytobementionednext?
YES
73%
26%
54%
0%20%40%60%80%
100%
Subj-biasedIC Obj-biasedIC Non-IC
%su
bjectcon
tinua
tions
Rohde&Kehler(2014):sentencecontinuations(nopronounprompt)