Hyper-active gap filling

26 Pages·2015·0.62 MB·English

by  Akira Omaki

ORIGINALRESEARCH
published:10April2015
doi:10.3389/fpsyg.2015.00384
Hyper-active gap filling
AkiraOmaki1*,EllenF.Lau2,ImogenDavidsonWhite2,MylesL.Dakan2,AaronApple1
andColinPhillips2
1DepartmentofCognitiveScience,JohnsHopkinsUniversity,Baltimore,MD,USA,2DepartmentofLinguistics,Universityof
Maryland,CollegePark,MD,USA
Much work has demonstrated that speakers of verb-final languages are able to
construct rich syntactic representations in advance of verb information. This may
reflect general architectural properties of the language processor, or it may only reflect
a language-specific adaptation to the demands of verb-finality. The present study
addressesthisissuebyexaminingwhetherspeakersofaverb-mediallanguage(English)
wait toconsult verb transitivity information before constructingfiller-gap dependencies,
where internal arguments are fronted and hence precede the verb. This configuration
makes it possible to investigate whether the parser actively makes representational
commitmentsonthegappositionbeforeverbtransitivityinformationbecomesavailable.
A key prediction of the view that rich pre-verbal structure building is a general
architectural property is that speakers of verb-medial languages should predictively
construct dependencies in advance of verb transitivity information, and therefore that
disruptionshould beobservedwhentheverbhasintransitivesubcategorization frames
Editedby:
that are incompatible with the predicted structure. In three reading experiments (self-
ClaudiaFelser,
pacedandeye-tracking)thatmanipulatedverbtransitivity,wefoundevidenceforreading
UniversityofPotsdam,Germany
Reviewedby: disruption when the verb was intransitive, although no such reading difficulty was
ArildHestvik, observed when the critical verb was embedded inside a syntactic island structure,
UniversityofDelaware,USA
which blocks filler-gap dependency completion. These results are consistent with the
OliverBoxell,
UniversityofPotsdam,Germany hypothesis that in English, as in verb-final languages, information from preverbal noun
*Correspondence: phrases is sufficient to trigger active dependency completion without having access to
AkiraOmaki, verbtransitivityinformation.
DepartmentofCognitiveScience,
JohnsHopkinsUniversity,3400North
Keywords: filler-gap dependency, active gap filling, prediction, verb transitivity, island, plausibility mismatch
CharlesStreet,237Krieger, effects,eye-tracking
Baltimore,MD21218,USA
[email protected]
Specialtysection: Introduction
Thisarticlewassubmittedto
LanguageSciences,asectionofthe Aleadinggoalofsentenceprocessingresearchistounderstandhowtheparseradaptstoamulti-
journalFrontiersinPsychology
tudeoflinguisticdifferencesacrosslanguagestoenablesuccessfulcomprehension.Inthisregard,
Received:22December2014 comparisonsofverb-medialandverb-finallanguageshaveprovidedavaluablesourceofevidence
Accepted:18March2015 (MazukaandLust,1990;InoueandFodor,1995).Themainverbcontainsrichinformationsuch
Published:10April2015
assubcategorizationandthematicroleinformationthatiscriticalforconstructingstructuralanal-
Citation: ysesandinterpretations(e.g.,Chomsky,1965;Grimshaw,1990;PollardandSag,1994;Levinand
OmakiA,LauEF,DavidsonWhiteI,
Rappaport Hovav,1995). Much experimental evidence shows that the verb is a valuable source
DakanML,AppleAandPhillipsC
of information for parsing (e.g., Ford et al., 1982; Tanenhaus and Carlson, 1989; Boland et al.,
(2015)Hyper-activegapfilling.
1990; MacDonald et al., 1994; Spivey-Knowlton and Sedivy,1995; Garnsey et al., 1997; Mauner
Front.Psychol.6:384.
doi:10.3389/fpsyg.2015.00384 andKoenig,2000;Traxleretal.,2002;BlodgettandBoland,2004;SnedekerandTrueswell,2004).
FrontiersinPsychology|www.frontiersin.org 1 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
Theimportanceoftheinformationfromtheverbheadhasengen- associatesthefillerwiththeverbinadvanceoftheverbacrosslan-
dered theoretical claims that structure building processes do guages,regardlessofdifferencesinverbpositions. Theseresults
not even start until the parser encounters the head of a phrase suggestthattheprocedureforfiller-gapdependencycompletion
(e.g.,verbalhead)tobeconstructed,eveninverb-finallanguages may be uniform across languages, and are consistent with the
wherethiswouldbesignificantlydelayed(Abney,1989;Pritchett, viewthattheparserpredictively constructs richrepresentations
1992). attheearliestpossiblemomentinadvanceofcriticalbottom–up
However, subsequent empirical research on verb-final lan- evidence.
guageslike Japanese or Germanhasgenerated evidence against
such head-driven parsing theories in their strongest form, Background on ActiveFiller-Gap
demonstrating that the parser uses various morphological and Dependency Processing
syntactic cues to incrementally build structures and interpre- Pastresearchonfiller-gapdependencyprocessinghasestablished
tations in verb-final languages (Bader and Lasser, 1994; Koh, thattheparserpostulatesagapbeforethereissufficientbottom–
1997;ClahsenandFeatherston,1999;KamideandMitchell,1999; up evidence to confirm that analysis (Active gap filling: Fodor,
Konieczny, 2000; Bornkessel et al., 2002; Felser et al., 2003; 1978; Crain and Fodor, 1985; Stowe, 1986; Frazier and Flores
Kamideetal.,2003;Aoshimaetal.,2009;Yoshida,unpublished d’Arcais,1989).Forexample,Stowe(1986)observedtheso-called
doctoraldissertation).Thus,althoughverbinformationstrongly Filledgapeffectin(2),i.e.,slowerreadingtimesatthedirectobject
influencesparsingdecisionswhenavailable,speakersofverb-final position us in the wh-fronting condition (2a) than in a control
languagesoftenbeginbuildingsyntactic andsemanticstructure conditionthatdidnotinvolvewh-fronting(2b).Thispatternof
inadvanceoftheverb. readingtimessuggeststhattheparserhadalreadypositedagap
Thesefindingsraisethequestionofwhetherpre-verbalstruc- followingthetransitiveverb,beforecheckingwhetherthedirect
ture building reflects a language-specific adaptation to the pro- objectpositionwasoccupied.
cessingdemandsofverb-finality,orratherapropertyofageneral
parsingarchitecturethatspeakersofalllanguagesuse.Forexam- (2) a. MybrotherwantedtoknowwhoRuthwillbringushome
ple,considerlessfrequentcasesinverb-mediallanguageswhere to____atChristmas.
multiple arguments precede the verb. A classic example of this b. MybrotherwantedtoknowifRuthwillbringushometo
comesfromprocessingof‘filler-gap’dependenciesasillustrated MomatChristmas.
bytherelativeclauseconstructionshownin(1),wheretheobject
nounphrase(NP)thecity(calledthefiller)isdislocatedfromthe Convergingevidencecomesfromaneye-trackingexperimentby
post-verbal thematic position (called the gap1), and the parser TraxlerandPickering(1996),whomanipulatedthethematicfit
needs to associate the filler and the gap in order to assign a betweenthefillerandthepotentialverbhost,asin(3).
thematicinterpretation.
(3) We like the city/book that the author wrote unceasingly
(1) The city that the author visited ____ was named for an and with great dedication about _____ while waiting for a
explorer. contract.
Ithasbeenreportedthatspeakersofverb-finallanguagescom- TraxlerandPickeringfoundaplausibilitymismatcheffectat
plete filler-gap dependencies in advance of verb information, thecriticalverbin(3),i.e.,thefirstfixationtimeattheoptionally
associatingthefillerwiththeearlieststructuralpositionwherea transitive verbwrote increased when the fillerwasanimplausi-
thematicrole could beassigned(pre-verbalobject gapcreation: bleobjectoftheverb(i.e.,thecity),comparedtowhenthefiller
Nakano et al., 2002; Aoshima et al., 2004). The current study was a plausible object of the verb (i.e., the book). This suggests
examines whether this may also be the case in a verb-medial thatatleastasearlyastheverbposition,theparserpostulatesa
language like English, and whether pre-verbal gap creation is a gap and analyzes the filler as the object of the verb,even when
language-general parsing procedure rather than an adaptation the filler is a poor semantic fit to that role. In fact, there is
specifictoverb-finallanguages.Underthishypothesis,wepredict ampletimecourseevidenceforactiveobjectgapcreation,using
thatEnglishspeakersshouldpositagapirrespective ofwhether a variety of dependentmeasuressuch asreading time and gaze
theverbultimatelylicensesadirectobjectgapposition,andthat durationmeasures(CrainandFodor,1985;Frazier,1987;Frazier
signsofreadingdisruptionshouldbeobservedincaseswherethe andClifton,1989;deVincenzi,1991;PickeringandTraxler,2001,
verbdoesnotaccommodateadirectobject. 2003; Aoshima et al., 2004; Phillips, 2006; Wagers and Phillips,
We report the results of three on-line reading experiments 2009), cross-modal priming (Nicol and Swinney, 1989; Nicol,
in Englishthattestedthis prediction byexamining the effect of 1993; Nakano et al., 2002), visual world eye-tracking (Sussman
verbtransitivityonreadingtimesinfiller-gapconfigurations.The and Sedivy, 2003) as well as event-related potentials (Garnsey
resultsareconsistentwiththehypothesisthattheparseractively etal.,1989;Featherstonetal.,2000;Kaanetal.,2000;Felseretal.,
2003;Phillipsetal.,2005;Gouveaetal.,2010).
1Inthispaperweusethe‘gap-filling’or‘gap-creation’terminologyinatheory- The work summarized above may suggest that filler-gap
neutralway,asistypicalinthepsycholinguisticliterature.Thisterminologyshould dependency completion is triggered only after the parser gains
notbetakenasindicatingacommitmenttorepresentationsthatincludegapsor
accesstotheverbandconfirmsthattheverbistransitiveandis
traces;alloftheprocessingtheorieswediscussherecouldbespecifiedintermsof
representationsthatdonotincludeemptycategories. abletosyntacticallyaccommodateanobject.However,evidence
FrontiersinPsychology|www.frontiersin.org 2 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
that active dependency completion does not depend on verb Inverb-mediallanguageslikeEnglish,verbsbecomeavailable
information has been presented by studies that investigated (i) relatively earlier in the sentence, such that the average work-
subjectgapcreationinEnglish,aswellas(ii)objectgapcreation ing memory cost of waiting for the verb would be less than in
inverb-finallanguages.Forexample,Lee(2004)usedsentences verb-finallanguages.Theadvantageofwaitingfortheverbinfor-
like(4)torevealafilledgapeffectinthesubjectNPposition. mation is that the parser can reduce the likelihood of making
riskycommitments,becausetheverbmayturnouttobeintran-
(4) a. Thatisthe laboratory which, ontwo differentoccasions, sitiveanddisallowanobjectNPanalysisforthefiller.InEnglish,
Ireneusedacouriertodeliverthesamplesto___. therefore,theparsermaycreateanobjectgappositiononlyafter
b. Thatisthelaboratorytowhich,ontwodifferentoccasions, theverbisconfirmedtobetransitive.Thisstillconstitutesactive
Ireneusedacouriertodeliverthesamples___. gapfilling,intherespectthattheultimategappositionmayturn
out to be somewhere later than the object position [e.g.,after a
Here,thecontentofthewh-fillerismanipulatedinsuchaway late-arriving preposition gap, as in (2) and (3)]. Let us call this
thatthewh-fillercanplausiblybeasubject(4a)ornot(4b).The aconservativeactivegapfillingmechanism,sincethebottom–up
resultsshowed a longerreading time atthe subject NPIrenein subcategorizationinformationfromtheverbstillplaysacritical
(4a)thanin(4b),suggestingthattheparserhadpostulatedasub- role in the parser’s decision on whether to postulate an object
jectgapbeforeencounteringtheactualsubjectNP.Althoughthis gapor not. Thisviewof active gapfilling isratherstandard for
interpretationhasbeenchallenged(Staub,2010),itwouldinany explainingfiller-gapdependencycompletioninverbmediallan-
case not be surprising that the parser actively creates a subject guageslikeEnglish.Forexample,McElreeandGriffith(1998)and
gapwithouthavingaccesstoverbinformation,giventhatasub- McElreeetal.(2003)havearguedthatthe dependencycomple-
ject is present in any sentence,regardlessof verb properties. In tion process is triggered when the parser accesses information
thissense,ifverbinformationweretoplayaroleintheparser’s fromtheverbandinitiatestheretrievalprocessforthefillerthat
attempttopositagap,thecriticalempiricalevidenceshouldcome isstoredinworkingmemory(seealsoPickeringandBarry,1991;
from dependency completion at the object position, where the LewisandVasishth,2005).
presence or absence of anobject gap reliesonproperties of the On the other hand, pre-verbal object gap creation in verb-
verb. final languages may reflect a language-general property of the
Evidenceforpre-verbalobjectgapcreationhasbeenreported processingarchitecture,althoughevidenceforsuchmechanisms
for verb-final languages like Japanese in which the object gap maybesimplymoredifficulttoobtaininverb-mediallanguages.
position linearlyprecedestheverb.Forexample,Aoshimaetal. In the English filler-gap case, for example, in any parser that
(2004) examined processing of scrambling sentences in which adoptssomeform ofleft-corner strategy(Kimball,1975;Abney
a dative object NP was dislocated to the sentence initial posi- and Johnson, 1991; Resnik, 1992; Shieber and Johnson, 1993;
tion, and found a filled gap effect at a pre-verbal dative object Stabler,1994;Crocker, 1996;LewisandVasishth,2005;Gibson,
position for the first verb phrase (VP) in the sentence (see unpublished doctoral dissertation), the presence of the subject
also Omaki et al., 2014). Using similar sentences, Nakano et al. NPallowstheparsertopredictthepresenceofaVP.Giventhata
(2002) reported evidence for an antecedent priming effect for VPcancontainanobjectNPposition,theparsercouldprojecta
the scrambled NP at a pre-verbal gap position, although the VPwithanobjectNPslotandassignthefillertothisobjectposi-
priming effect was only found in the high working memory tionbeforeconfirmingwhethertheupcomingverbisatransitive
span group. These data indicate that the parser can in prin- verbornot.Letuscallthisahyper-activegapfillingmechanism,
ciple complete filler-gap dependencies before accessing verb because this involvesa more risky predictive structure building
information. processthanisstandardlyassumedforactiveobjectgapcreation
In verb-medial languages, no such evidence for pre-verbal inEnglish.Fillerretrievalandstructuralintegrationisstillinte-
object gap creation has been reported to date. This may reflect gral to the hyper-active gap filling mechanism, but the crucial
a real difference between languages in processing strategy, and difference is in what information triggers retrievaland integra-
pre-verbalobjectgapcreationinverb-finallanguagesmayreflect tion,andconsequently,atwhatpointinthesentencethisprocess
the parser’s adaptation to the demandsof processing these lan- isexecuted.
guages.Maintainingastructurallyunintegratedfillerinmemory Itisimportanttonotethateitherofthesetwoactivegapfill-
hasbeenarguedtoimposeaburdenonworkingmemory(King ing mechanisms is compatible with the existing data on active
and Just, 1991; Gibson, 1998; Gordon et al., 2002; Haarmann objectgapcreationreviewedabove.Afilledgapeffectonlyindi-
and Cameron,2005). Alternatively, the parser may be architec- catesthatthegaphadbeencreatedbeforetheactualobjectNPis
turally constrained to assign a thematic interpretation to the processed,andthisresultiscompatiblewithbothaccounts,given
filler as soon as possible (Pickering and Barry, 1991; Aoshima thatbothhyper-activegapfillingandconservativeactivegapfill-
etal.,2004).Onthisview,the parser shouldprioritize integrat- ing mechanisms assume that object NP gap creation happens
ing the filler into the first grammatically permissible structural before or on the verb. A plausibility mismatch effect indicates
position that canpotentially receive a thematicrole. Giventhat thatwhentheverbispotentiallytransitive,thenthesemanticfit
filler-gap dependencies are potentially unbounded, waiting for between the filler and the verb is immediately assessed.This is
the verb before constructing the ultimate object gap position also predicted by both accounts. The assessmentof the seman-
couldimposealargeprocessingburdenonspeakersofverb-final tic relation between the filler and the verb requires the parser
languages. toaccessthe contentofthe verb,bywhich pointthe object gap
FrontiersinPsychology|www.frontiersin.org 3 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
position shouldhavebeencreatedoneitheraccount.Thus,nei- Staub(2007)foundlongerfirst-fixationdurationsinthetran-
ther paradigm allows us to tease apart the two hypotheses on sitivegapcondition(5a)thaninthetransitiveno-gapcondition
what kind of information is sufficient for triggering object gap (5b), but no such difference was observed between the intran-
creation. sitive gap and no-gap conditions (5c) and (5d). This pattern of
Inthe currentstudyweaimtoteaseapartthepredictions of data supports the prediction of the conservative active gap fill-
twohypothesizedmechanismsforactiveobjectgapcreationpro- ing hypothesis, suggesting that the parser does not create an
cesses.IfEnglishspeakersconstructthegapsitebeforeencoun- object gap until it checks the transitivity information of the
tering the verb, just like speakers of verb-final languages, then verb. One concern about this design, however, is whether the
disruptionshouldbeobservedinfiller-gapconfigurations when no-gap condition was truly a neutral baseline against which a
the verbturns out to be intransitive, relative to transitive verbs transitivitymismatchcouldbemeasured,asthegapandno-gap
(e.g., The party that the student arrived/planned...). According conditions differed substantially in both the linear and struc-
totheconservativeactivegapfillingmechanismoutlinedabove, turalpositionoftheverb.AsStaub(2007)pointsout,onepiece
the parserwaitsforatransitive verbbeforepostulating the cor- of data suggesting that the control may not have been com-
responding gapstructure. Here,nodisruption isexpected atan pletelyneutralisthe factthatreadingtimesontheintransitives
intransitive verb,since the parser has not postulated a gap that werenumerically(butnon-significantly) shorterinthegapcon-
wouldrequireatransitiveverb. dition than in the no-gap condition. It is important to note
Two previous studies are relevant to the two hypotheses here that the gap conditions (5a) and (5c) contain an extra NP
about active object gap creation in English. Previous work by (i.e., the head of the relative clause) prior to the critical verb
Pickering and Traxler (2003) examined the effect of subcatego- region in comparison to the no-gap conditions (5b) and (5d).
rization frequency in optionally transitive verbs (e.g., Those are This may have led to a difference in the amount of contex-
thelines/propsthattheauthorspoke[about]...).Itwasfoundthat tualinformationavailablepriortotheverb.Increasedcontextual
readers did not take subcategorization frequency into account informationcanfacilitateprocessingforsubsequentlexicalitems
in deciding where to posit a gap, as there was a strong prefer- (Stanovich and West, 1983; Van Petten and Kutas, 1990; Kutas
encetopositagapintheverbobjectposition(NPcomplement) and Federmeier, 2000), and for this reason, lexical access for
even with verbs that more frequently take a PP complement. the intransitive verb in the gap condition may have become
Theabsenceofsubcategorizationfrequencyeffectinactiveobject faster and masked the potential reading time slowdown associ-
gapcreationcouldbetakentoindicatethatverbinformation is atedwiththestructuralmanipulation.Inanattempttoprovidea
not relevant for object gap creation processes. However, all of better test of the predictions of the hyper-active and conserva-
the verbs in Pickering and Traxler’s study could grammatically tive active gap filling accounts, the current study used relative
accommodateanNPcomplement,andtheparsermaytherefore clause islands as a control condition, which allowed the target
havereliedonthetransitivityinformationoftheverbtocreatean sentences to more closely match in informational content and
object gap.Therefore,thisfindingdoesnotdistinguishthe pre- wordposition.
dictions of the two proposed mechanisms for active object gap
creation.
Toour knowledge,the onlyprevioustestofthese twoactive Experiment 1
objectgapcreationhypothesesisinExperiment3ofStaub(2007).
The test sentences in this experiment (5a–d) manipulated the Experiment1wasaself-pacedreadingstudythatwasdesignedto
transitivityoftheverb(calledvs.arrived)andsentencestructure test the predictions of the hyper-active and conservative active
(relativeclausewithagapvs.simpledeclarativewithnogap).The gap filling hypotheses, while addressing methodological con-
fillerwasmanipulatedtobeanimplausibleobjectofthetransitive cerns about previous work. We employed the transitivity mis-
verb(gadget-called).Underthehyper-activegapfillinghypothe- match paradigm used in Staub (2007) in order to test whether
sis,theparserineffectpredictsthepresenceofatransitiveverb, averbtransitivitymanipulationaffectsreadingtimeattheverb.
andthereforethereadingprocessesinthegapconditionsshould Critically,inthebaselineconditionsthecriticalverbwasembed-
be disrupted in either intransitive or transitive condition, but dedinsidearelativeclausestructure,asyntactic‘island’domain
fordifferentreasons:whentheverbturnsouttobeintransitive, that prohibits filler-gap dependency formation (Ross, unpub-
and processing should also be disrupted when the verb is tran- lished doctoral dissertation; for a review, see Szabolcsi and den
sitive because of the plausibility mismatch effect. On the other Dikken,2003).AsamplesetofstimuliisshowninTable1.
hand,theconservativeactivegapfillingmechanismpostulatesa A number of previous studies have shown that the parser
gap only after checking whether the verb is capable of hosting respects island constraints in real-time syntactic processing,
anobjectNP,andthereforereadingdisruptionispredictedonly such that it avoidsactively constructing filler-gap dependencies
in the transitive gapcondition due tothe plausibility mismatch that span syntactic island boundaries (Stowe, 1986; Kluender
effect. and Kutas, 1993; McKinnon and Osterhout, 1996; Traxler and
Pickering,1996;McElreeandGriffith,1998;WagersandPhillips,
(5) a. Thegadgetthatthemanagercalledoccasionallyabout... 2009; Omaki and Schulz, 2011; Yoshida, unpublished doctoral
b. Themanagercalledoccasionallyaboutthegadget... dissertation).Therelativeclauseislandcondition thusprovided
c. Thepartythatthestudentarrivedpromptlyfor... a baseline measure of reading times for the critical transitive
d. Thestudentarrivedpromptlyfortheparty... and intransitive verbs, independent of processes of filler-gap
FrontiersinPsychology|www.frontiersin.org 4 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
TABLE1|SamplematerialsandconditionsforExperiment1.
Analysisregions
1 2 3 4 5 6 7 8 9 10 11
Transitive,non-island The city that the author wrote regularly about was namedforanexplorer
Transitive,island The city that the author who wrote regularly saw was namedforanexplorer
Intransitive,non-island The city that the author chatted regularly about was namedforanexplorer
Intransitive,island The city that the author who chatted regularly saw was namedforanexplorer
Examplequestion Wasthecitynamedforanexplorer?
dependencycompletion.Theuseofislandconfigurationsallowed madeavailableinSupplementary Materials.Thetransitivenon-
ustoaddressthemethodologicalconcernswithpreviouswork. island and island conditions were taken from the implausible
First, this design allowed the baseline condition to present a semantic fit conditions in Omaki and Schulz (2011), who used
fillerNPpriortothecriticalregion,suchthatthesameamount a modified version of the plausibility manipulation materials
of contextual information from the lexical itemswaspresentin fromTraxlerandPickering(1996).OmakiandSchulzreplicated
advance of the critical verb region across the four conditions. Traxler and Pickering’s plausibility mismatch effect with native
Second,thewordpositionforthecriticalregions(Regions7and and non-native speakers alike, confirming that the semantic fit
8inTable1)wascloselymatchedacrossconditions(wordposi- between the filler and the verb affects the reading time for the
tions6and7inthenon-islandconditions,wordpositions7and verbwhenthe verbisina gapfilling (i.e.,non-island) environ-
8intheislandconditions),anditwasalsoplacedawayfromthe ment, but not when the verb is inside a relative clause island.
earlyportionofthesentence. Critically,itwasalsofoundthattheimplausibleverb-fillercom-
Furthermore, following Staub’s design, we selected transi- bination in a non-island environment (e.g., city-wrote) led to a
tive verbs that are implausible hosts for the filler. Under this significant slow down at the verb compared to its island coun-
design,the hyper-active gapfilling hypothesis predicted aread- terpartwiththesameimplausibleverb-fillercombination.Thus,
ing time slowdown in both the non-island transitive and the even though the current experiment did not include a plausi-
non-island intransitive conditions relative to their island coun- ble counterpart of the implausible transitive verbcondition, we
terparts, but for a different reason in the two cases. In the couldbeconfidentthatareadingtimecontrastbetweenthetran-
transitive condition, the slowdown would reflect a plausibil- sitivenon-islandandislandconditionsresultsfromthesemantic
ity mismatch effect triggered by the poor semantic fit between misfit betweenthe fillerand the verb. Inother words, the find-
the filler and the verb. In the intransitive condition, the slow- inginOmakiandSchulz’sstudysupportsthenotionthatisland
downwouldresultfromatransitivitymismatcheffectduetothe conditions in general can be used as baseline conditions for a
mismatch between the expected subcategorization property of readingdisruptionassociatedwithactiveobjectgapcreation.The
the verb(i.e.,transitive) and the actualsubcategorization prop- intransitive conditions were modeled after the transitive condi-
erty of the verb. On the other hand, the conservative active tionsbyreplacingtheoptionally transitiveverbwithunergative
gap filling hypothesis predicted an interaction. A reading time orunaccusativeintransitiveverbs(LevinandRappaportHovav,
contrast should be observed between the non-island transitive 1995).
condition and the island transitive condition due to the plausi- The non-island and island conditions differed in the num-
bilitymismatcheffect,butnocorresponding contrast shouldbe ber of relative clauses. The non-island condition had only one
observedbetweenthetwointransitiveconditions,giventhatthe relative clause (the city that the author wrote/chatted regularly
parser should not actively create an object gap in either condi- about), such that the object position of the verb wrote/chatted
tion. Note that the lexical difference in the critical verb region was the first potential gap position after the embedded sub-
acrossconditionswasnotproblematic,sincethecriticalcontrast ject was encountered. In the island conditions, the critical verb
wasbetweennon-island andislandconditions within eachverb was embedded inside another relative clause the author who
type. wrote/chatted regularly, such that linearly this was still the first
verb but grammatically the filler should not be accessible to
Method the verb due to the relative clause island constraint. Thus, the
Participants first verb served as the critical region for testing the plausibil-
We recruited 32 native speakers of American English from the ityandtransitivitymismatcheffects.Allthetransitiveverbswere
UniversityofMarylandcommunity.Theyreceivedacoursecredit optionally transitive, such that the sentences in the island con-
or were paid $10 for their participation and were naïve to the ditions were all ultimately grammatical. The subcategorization
purposeoftheexperiment. frequency of the optionally transitive verbs was not controlled,
sincePickeringandTraxler(2003)havedemonstratedthatplau-
Materials sibility mismatch effects are attested for optionally transitive
We used 28 setsof four sentences like those shown inTable 1. verbsregardlessofsubcategorizationfrequency.Inallfourcon-
All of the stimuli from experiments reported in this paper are ditionsthesameadverbimmediatelyfollowedtheverb,making
FrontiersinPsychology|www.frontiersin.org 5 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
itpossibletoobservepotentialspill-overeffects.The28sentence for R (Bates et al., 2014). The fixed effects of island structure
setswerecounter-balanced acrossfourlistssothateachpartici- type (non-island vs. island) and verb transitivity (transitive vs.
pantsawonly oneversion ofthe targetitemsand consequently intransitive) were coded using sum contrasts, with one levelof
readseventokensofeachcondition.Inaddition,72fillersofsim- thefactorcodedas−0.5,andtheotheras0.5.Thissumcontrast
ilar length and complexity were constructed and addedto each coding makes the mixed effect model estimates roughly com-
list. parabletotheactualaveragereadingtimecontrasts. Themodel
includedrandominterceptsforparticipantsanditems.Forran-
Procedure
dom slopes, we used the following procedure to determine the
The self-paced reading task was implemented on the Linger optimal random effect structure (for discussions: Jaeger, 2011;
software developed by Doug Rohde (http://tedlab.mit.edu/∼dr/ Barr et al., 2013). First, we constructed a fully crossed model
Linger/).Weusedaword-by-word,non-cumulativemovingwin- that included the fixed effects and an interaction term as ran-
dowpresentation(Justetal.,1982).Inthisdesign,eachsentence dom slopes for both participants and items.Thisfully specified
initially appears as a series of dashes, and these dashes are model failed to converge, plausibly due to the complexity of
replaced byaword from lefttoright everytimethe participant the model and missing data points in some of the trials (Barr
presses the space bar. In order to ensure that the participants etal.,2013).Next,wesimplified the random effect structure by
werepayingattentionwhile readingthesentences,allsentences only keeping the verb transitivity factor as a random slope for
were followed by yes-no comprehension questions, and feed- participants and items. In our experimental design, the island
back was provided if the questions were answered incorrectly. structureisinvariantacrossallitems,anditisalsoknowntobe
Comprehensionquestionsneveraddressedthecriticalfiller-gap robustacrossindividuals,regardlessofworkingmemorycapacity
portion of the sentence. At the beginning of the experiment, (see Sprouse et al., 2012). On the other hand, the verbs dif-
participants were instructed to read at a natural pace and to fered across items, and it is possible that the subcategorization
answer the questions as accurately as possible. Seven practice bias differs across participants. This mixed effects model con-
itemsprecededtheself-pacedreadingexperiment,andtheorder verged for all regions. We computed p values for linear mixed
ofpresentationwasrandomizedforeachparticipant.Theexper- effectsmodelsusingthe lmerTestRpackage(Kuznetsova etal.,
imenttook∼30min.Theexperimentprotocolforthisstudywas 2014).
approvedbytheInstitutionalReviewBoardattheUniversityof
Maryland. Results
Comprehensionaccuracy
DataAnalysis
The mean comprehension question accuracy for experimental
Thedatafromtwoitemswereexcludedfromanalysesduetocod- itemsacrossparticipantsanditemswas93.0%.Forthenon-island
ingerrors.Onlytrialsinwhichthecomprehensionquestionwas conditions,thetransitiveitemswereansweredwithanaccuracy
answeredaccuratelywereincludedintheanalysis,whichaffected of93.7%(SE=1.9),andtheintransitiveitemswithanaccuracy
5.7%ofthetrials.Wealsoanalyzedthedatawithoutexcludingthe of94.6%(SE=1.4).Fortheislandconditions,thetransitiveitems
trialsbasedoncomprehensionaccuracy,buttheoverallpatternof were answered with an accuracy of 91.5% (SE = 1.7), and the
resultsdidnotchange. intransitiveitemswithanaccuracyof92.0%(SE=2.2).Themean
Self-pacedreadingtimesforthetargetsentenceswereexam- accuracy did not differ reliably across conditions, although the
ined for each successive region, although the words after the factthatthemeanaccuracyforislandconditionswasnumerically
auxiliary waswere combined into a single region because these lowermayreflectthe complexity difference betweennon-island
laybeyondthecriticalregionsandwereunlikelytoshoweffects andislandconditions.
relevantforthecriticalmanipulation.Thecriticalregionswherea
potentialplausibilityortransitivitymismatcheffectwasexpected Readingtimedata
consist of Region 7 (i.e., the verb wrote/chatted) and the fol- Theregion-by-region meanreadingtimeforthe transitivecon-
lowing Region 8 (i.e., the adverb regularly), in which spill-over ditionsispresentedinFigure1,andthemeanregion-by-region
effects could be observed. Regions 1 through 6 were predicted reading time for the intransitive conditions is presented in
to show no difference across conditions, since they were lexi- Figure2.
cally matched. Regions 9 through 11 could reveal reading time Inthenon-criticalRegions1–6,therewerenosignificantdif-
differencesafterthefiller-gapdependencyiscompleted(Region ferencesinRegions1,2,4–6(ps>0.06).InRegion3therewasa
9 hosts the true gap site), and with a possible additional dif- maineffectofverbtype(Estimate=−17.3,SE=7.6,t=−2.27,
ference in the island conditions due to the structural com- p<0.05),duetoslowerreadingtimesinthetransitiveconditions
plexity associated with the extra relative clause in these condi- than in the intransitive conditions (381 vs. 358 ms). Since this
tions. regionwaslexicallymatchedacrossconditions,weconcludethat
Reading time data that exceeded three standard deviations this is a spurious effect. Butgiventhat the effect wassmalland
fromthegroupmeanateachregionandineachconditionwere occurredwellaheadofthecriticalregions,thisunexpectedeffect
excluded,affecting1.7%ofthedata.Theremainingreadingtime was unlikely to have impacted the observations in the critical
data were analyzed using linear mixed effects models (Baayen regions.
et al., 2008). These analyses were conducted in the R environ- At the critical verb in Region 7 there were no signifi-
ment(RDevelopmentCoreTeam,2011),usingthelme4package cant differences (ps > 0.1). The following spill-over region
FrontiersinPsychology|www.frontiersin.org 6 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
Discussion
In Experiment 1, we tested the predictions of two hypotheses
about active object gap creation. The hyper-active gap filling
hypothesispredictedthepresenceofreadingdisruptionatintran-
sitiveverbs,becauseencounteringanintransitiveverbinafiller-
gap context would be incompatible with the object gap struc-
ture constructed earlier. On the other hand, the conservative
activegapfillinghypothesispredictednosuchreadingdisruption,
because the parser should first consult the transitivity informa-
tionoftheverbtodecidewhethertopositanobjectgapornot.As
abaselineforestimatingthedegreeofdisruptionattheverb,we
usedrelativeclauseislandconstructions,whichblocktheassocia-
tionofthefillerwiththecriticalverb.Theresultswereconsistent
with the predictions of the hyper-active account: in the region
followingtheverb,weobservedslowerreadingtimesforintran-
sitiveverbsinnon-islandconditionsthanincorrespondingisland
conditions.
Previous work has shown a filler-gap plausibility mismatch
FIGURE1|Meanreadingtime(ms)forthetransitivenon-islandand
effect at the verb such that mismatched transitive verbs in a
islandconditions.Errorbarsindicatestandarderrorofthemean.
non-island environment elicit longer reading times than their
plausiblenon-islandorplausible/implausibleislandcounterparts
(Traxler and Pickering, 1996; Omaki and Schulz, 2011), and
herewereplicatedthisfinding.Thiseffectcanbeinterpretedas
the result of active association of the filler with the transitive
verb, which in these stimuli resulted in a verb–object plausi-
bility mismatch. On the other hand,the slowdown observed in
the intransitive non-island condition relative to the intransitive
island condition can be interpreted as a transitivity mismatch.
This suggests that the parser does not wait for bottom–up evi-
dence from the verb that the verb can syntactically license a
gap,butratherattemptstoconstructthedependencybeforethis
information is available. This slowdown cannot reflect the cost
of maintaining the filler in working memory, because a filler
is also being maintained at this position in the baseline island
condition.
Itisalsoimportanttonotethattheshorterreadingtimesinthe
criticalregionsoftheislandconditionsaretheoreticallyinforma-
tive.Thesefindingssuggestthatthereadingtimeincreaseinthe
FIGURE2|Meanreadingtime(ms)fortheintransitivenon-islandand non-island conditions is specifically due to an expectation vio-
islandconditions.Errorbarsindicatestandarderrorofthemean. lation following premature gapcreation. Aplausible alternative
explanation of the reading disruption in the non-island condi-
tionsisthatitreflectsamoregeneralcostassociatedwithdelaying
(Region 8) revealed no main effect of verb type, but there gapcreationdecisions.Underthisalternativeaccount,weshould
was a main effect of structure type (Estimate = −92.0, expecttoobservereadingdisruptionintheislandconditionsas
SE = 16.4, t = −5.61, p < 0.001), reflecting the fact that well,because gapcreation mustwaituntil the verbthat follows
the non-island conditions produced significantly slower read- therelativeclauseislandregion(e.g.,sawinRegion9).However,
ing times than the island conditions (529 vs. 435 ms). There thispredictionisnotsupportedbythedata,asthereadingtimein
was no significant interaction of verb type and structure type theadverbregion(Region8)oftheislandconditionswasreliably
(p>0.1). shorterthaninnon-islandconditions.
Region9 consisted of asecond verbinthe island conditions In Regions 9 and 10, the island conditions were read more
and a preposition in the non-island conditions. We observed slowlyfor bothlevelsofverbtype.Region9corresponds tothe
a main effect of structure type in Region 9 (Estimate = 63.7, word that licensed the true gap site across all conditions, and
SE = 15.9, t = 4.01, p < 0.001), as well as in Region 10 hence this slowdown could reflect a difference in the so-called
(Estimate =46.1,SE=11.5,t =4.0,p<0.001),inthesecases integration cost (Gibson, 1998, 2000) between non-island and
due to slower readingtimesin the island conditions (Region 9: island conditions. Previouswork on filler-gap dependency pro-
519vs.451ms,Region10:451vs.406ms).Region11revealedno cessing has demonstrated that increased complexity and length
significantdifferences(ps>0.09). differences result in increased processing difficulties at the gap
FrontiersinPsychology|www.frontiersin.org 7 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
site, as measured by reading time (Gibson and Warren, 2004; rather to a word category expectation mismatch in the adverb
Wagers and Phillips, 2014) and reduced accuracy in speeded regionthatwastriggeredbytheverbitself.Thisaccountiscon-
acceptabilityjudgmenttasks(McElreeetal.,2003).However,the sistent with the conservative active gap filling hypothesis, since
readingtimedifferenceinRegion9maysimplybeduetolexical theparser’sexpectationregardingfiller-gapdependencycomple-
differences(prepositionsinthenon-islandconditionsvs.verbsin tionisbasedontheinformationfromtheverb.Incidentally,the
theislandconditions), sothereadingtimecontrastbetweenthe readingdisruptionobservedinthetransitiveconditionsofStaub
islandandnon-island conditions maynotreflectanintegration (2007) wasatthe verbregion. One possible reason for this dis-
costdifference. crepancyisthedifferenceinthedependentmeasure:Staub(2007)
Notethatitisunlikelythatthereadingtimecontrastbetween usedaneye-trackingduringreadingmethodwhileweusedself-
non-island and island conditions in Region 8 is related to the pacedreadinginExperiment1.Aneye-trackingduringreading
overallcomplexityoftheconstructionsusedinourstimuli,given method generally provides better temporal precision than the
that on all accounts that we are aware of, island domains have self-pacedreadingmethod(Rayner,1998;RaynerandPollatsek,
beenarguedtobesyntacticallymorecomplexandmoretaxingfor 2006). Thus, an eye-tracking replication of Experiment 1 may
working memory resources (Deane, 1991; Kluender and Kutas, yieldatransitivitymismatcheffectontheverbregion,andpro-
1993;Kluender,1998,2004;HofmeisterandSag,2010).Thefact videstrongerevidenceforthehyper-activegapfillinghypothesis.
thattheputativelylesscomplexnon-islandconditionswereread ThisisaddressedinExperiment2.
moreslowlyallowsustoattributetheslowdowntoprocessesthat
uniquely occur in the non-island conditions, namely filler-verb
Experiment 2
association.
In summary, the presence of both a plausibility mismatch
Experiment 2 addressedtwo methodological concerns raised in
effect and a transitivity mismatch effect lends support to the
Experiment 1 by removing sources of slowdown in the transi-
hyper-active gap filling hypothesis, and argues against a con-
tiveconditions,andalsobyusingtheeye-trackingduringreading
servative active gap filling hypothesis under which transitivity
method.
information is consulted before attempting to create an object
gap.ThisfindingdirectlycontrastswiththatofStaub(2007),who Method
didnotfindevidenceforatransitivitymismatcheffect. Participants
However,thisconclusionisnotwarranteduntiltwomethod-
We recruited 33 native speakers of American English from the
ologicalconcernsareaddressed.First,thedesigninExperiment1
Johns Hopkins University community, but data from one par-
wasmodeledafterStaub(2007),whousedaplausibilitymismatch
ticipant were removed due to calibration errors. Participants
design for transitive verb conditions, and transitivity mismatch
received course credit or $10 for their participation. They were
design for intransitive verb conditions. Our findings differed
allnaïvetothepurposeoftheexperiment.
fromStaub’saswefoundmismatcheffectsforbothtransitiveand
intransitive non-island conditions, but it is possible that some Materials
nuisance factor common to both non-island conditions led to Weused28setsoffoursentencesasshowninTable2.Thisexper-
aslow-downacrosstheboard.Strongerevidenceforthehyper- imentusedthesametransitivitymismatchlogicasExperiment1
active gap filling hypothesis can be obtained if we replicate the andmanipulatedtheverbtransitivitytype(intransitivevs.tran-
transitivity mismatch slowdown in the intransitive non-island
sitive). However, in this experiment the semantic fit between
condition, while at the same time observing no reading dis- the filler and the transitive verb wasalways plausible,such that
ruption in the transitive non-island condition. Experiment 2
noreadingdisruption wasexpected atthetransitive verbinthe
accomplishedthisbymakingthefillerandtheverbsemantically non-islandcondition.AsinExperiment1wemanipulatedstruc-
fitinthetransitiveconditions.Theabsenceofreadingdisruption
ture type (non-island vs. island), using conditions with relative
inthetransitiveconditionswouldsuggestthatthedisruptionin clause island structures as baseline conditions. Relative clause
thenon-island,intransitiveconditionisduetotheintransitivity
islandsprovideaneffectivebaseline,sincetheyincludethesame
oftheverb. fillerNPandotherlexicalmaterialasthenon-island condition,
Second,itisimportanttonotethatourevidenceforreading
whilepreventingdependencycompletionatthecriticalverb.As
disruptionfortransitiveandintransitiveverbs(i.e.,theslowdown in Experiment 1, the transitive verbs were optionally transitive
innon-islandconditionscomparedtoislandconditions)wasnot
and the true gap position occurred outside the island domain,
observeduntilthespill-overadverbregion.Spill-overeffectsare allowingthesentencetocontinuegrammatically.
widelyobservedinself-pacedreadingexperiments,anditisthus
The28sentencesetswerecounter-balancedacrossfourlistsso
commontoattributespill-overeffectstoprocessestriggeredina thateachparticipantsawonlyoneversionofthetargetitemsand
precedingregion.However,inourexperimentthereisanalterna-
consequently read seven tokens of each condition. In addition,
tiveexplanationfortheeffectintheadverbregionthatwouldnot 76fillersofsimilarlengthandcomplexitywereconstructedand
requirehyper-activegapfilling.Fortheintransitivecondition,the
addedtoeachlist.
slowdownintheadverbregioncouldindicatethattheparserhad
Procedure
expectedthepresenceofapreposition,whichwouldallowstruc-
turalintegrationofthefiller.Underthisalternativeaccount,the An Eyelink 1000 eye-tracker (SR Research: Mississauga, ON,
slowdownisnotduetoatransitivitymismatchontheverb,but Canada) was used to record eye movements. The participant’s
FrontiersinPsychology|www.frontiersin.org 8 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
TABLE2|SamplematerialsandconditionsforExperiment2.
Analysisregions
Sentenceinitial Pre-verb Verb Post-verb Sentencefinal
Transitive,non-island Thebookthat theauthor wrote regularly aboutwasnamedforanexplorer
Transitive,island Thebookthat theauthorwho wrote regularly sawwasnamedforanexplorer
Intransitive,non-island Thebookthat theauthor chatted regularly aboutwasnamedforanexplorer
Intransitive,island Thebookthat theauthorwho chatted regularly sawwasnamedforanexplorer
Examplequestion Wasthebooknamedforanexplorer?
headwasstabilizedbyachinrestandaforeheadrest.Theposition author who in island conditions), in order to ensure that there
oftherighteyeonlywasmonitoredatasamplingrateof1000Hz. werenounexpectedreadingbehaviordifferencesthatmightcom-
The eye-tracker display allowed a maximum of 120 characters promise the interpretation of the data from the critical region,
per line, in 10 pt Monaco font. Some filler sentences were dis- (b) the verb region, which is the critical region where poten-
playedontwolines,butalltargetsentencesweredisplayedonone tialtransitivity mismatcheffectsmight beobserved,and(c) the
line.Stimuliweredisplayedona21.5-inchSamsungSyncMaster post-verb region, which corresponds to the post-verbal adverb
monitor,andparticipantswereseated65cmfromthecomputer and could be used to probe for potential spill-over effects. The
screen.Beforetheexperimentstarted,participantswereseatedin data in the remaining regions are not reported, because read-
front oftheeye-trackerandreceivedinstructions forthe exper- ing timesat these regions are not critical for distinguishing the
iment.Acalibration routine wasperformed atthe beginning of competinghypotheses.Moreover,afterthepost-verbregion,the
theexperiment,andtheexperimentermonitoredthecalibration lexicalitemswerenotheldconstantacrossconditionsandthere-
accuracy throughout the session, recalibrating when necessary. foreanyobserveddifferenceswouldbedifficulttointerpret.The
The experiment started with written instruction on the display islandconditionscontainedoneextraword,i.e.,therelativepro-
and sevenpractice trials. Atthe beginning of eachtrial,a black noun(e.g.,who),whichcouldhaveaffectedreadingtimesinthe
circlewasdisplayedontheleftsideofthemonitor,whichcorre- pre-verb region as well as regression measures for subsequent
spondedtothelocationofthebeginningofthesentence.Thetext regions.
wasdisplayedaftertheparticipantsuccessfullyfixatedonthecir- FollowingthedataanalysisproceduresusedinStaub(2007),
cle.Afterreadingeachsentence,theparticipantpressedabutton fourreadingtimemeasureswerecomputedforthethreeregions
to remove the sentence display. Each sentence was followed by ofinterests:firstfixationduration,firstpasstime,regressionpath
ayes-nocomprehensionquestion,andtheparticipantanswered time,andpercentregressions(Rayner,1998;RaynerandPollatsek,
the comprehension question by pressing a left or right button. 2006;StaubandRayner,2007).Firstfixationdurationisthedura-
Comprehensionquestionsneveraddressedthecritical filler-gap tion of the very first fixation in a region, regardless of whether
portion ofthe sentence.Theentire experimentlasted∼35min. thereisasinglewordormultiplewordsinthatregion.Thismea-
The experiment protocol for this study was approved by the sure is often used as an index of lexical difficulty (e.g., Reichle
Homewood Institutional Review Board at the Johns Hopkins et al., 2003) but is also informative about the earliest syntactic
University. processesthatimmediatelyfollowlexicalaccess(e.g.,Frazierand
Rayner,1982;Sturt,2003).
DataAnalysis Thefirst-passreadingtimeiscalculatedbysummingthefixa-
Comprehensionaccuracyforthetargettrialswas90.7%,andtri- tionsinaregionbetweenthetimewhentheeye-gazefirstenters
alsinwhichparticipants answeredthecomprehensionquestion the region from the left and the time when the eye-gaze exits
incorrectly were removed from the eye movement analyses, as theregioneithertotheleftortheright.First-passreadingtimes
data from these trials may reflect inattentive reading. For the alsoindexearlylexicalandsyntacticprocessesassociatedwitha
remaining data, an automatic procedure pooled short contigu- region, but given that they consist of multiple fixations on the
ous fixations. The procedure incorporated fixations of lessthan sameregion,theymayalsoreflectslightlylaterprocessesthanthe
80msintolargerfixationswhentheyoccurredwithinonecharac- firstfixationmeasure.
terofeachotheranddeletedanyremainingfixationsoflessthan Regression path times are the sum of fixations from the time
80 ms, because little information can be extracted during such whenthe eye-gazefirstentersaregionfrom the lefttothetime
shortfixations(RaynerandPollatsek,1989).Unusuallylongfixa- whentheeye-gazeexitstheregiontotheright.Regressionpath
tionsgreaterthan800mswerealsoremoved,becausetheyusually time is identical to first-pass reading time if the eye-gaze first
reflecttrackerlossesorotheranomalousevents.Thisprocedure exitsthe region tothe right, butifthe eye-gazeexitsthe region
resultedintheexclusionof4.86%ofallfixations. to the left, then regression path timesare longer than the first-
Forthepurposeofanalysisoftheeyemovementdata,thesen- passtimeastheyincludeallfixationsinpreviousregionsaswell
tencesweredividedintofiveanalysisregions,asshowninTable2. asre-fixationsontheregionbeforeexitingtheregiontotheright.
We report eye movement data in the following three regions: Thus,regressionpathtimesarelikelytoreflectslightlylaterpro-
(a)thepre-verbregion(theauthorinnon-islandconditions,the cesses,suchasintegrationofthecriticalregionwiththepreceding
FrontiersinPsychology|www.frontiersin.org 9 April2015|Volume6|Article384
Omakietal. Hyper-activegapfilling
context. The percent regressions indicate the probability that a In the pre-verb region, the first pass time and regression
readermadearegressiveeyemovementtoprecedingregionsafter path measures showed a main effect of structure (p < 0.001),
fixating a given region. This measure includes only regressions with longer reading times in the island conditions than in the
madeduringthereader’sfirstpassthroughtheregion,anddoes non-islandconditions.Thiseffectwasexpectedbecausethepre-
notincluderegressionmadeafterre-fixatingtheregion. verb region in the island conditions contained the extra word
Readingtimedata(i.e.,firstfixation,firstpass,andregression who, which made it more likely to attract multiple fixations in
pathdurations)wereanalyzedusinglinearmixedeffectsmodels that region. No other significant effects were observed in this
(Baayen et al., 2008), and percent regressions were analyzed by region.
mixed effects logistic regression, asthe dependentmeasurewas In the verb region, evidence for the hyper-active gap filling
categorical(seeJaeger,2008).Themixedeffectsmodelsincluded hypothesiswasfoundinfirstfixationdurationsaswellasinfirst
randominterceptsforparticipantsanditems.Weusedthesame passmeasures.Bothmeasuresshowedamaineffectofstructure
procedure asExperiment 1 tosimplify the random slope struc- withlongerreadingtimesfornon-islandconditions(ps<0.05).
tureuntilthemodelsconvergedinallregionsandeyemovement First fixation durations showed a marginal interaction of struc-
measures. This procedure led us to adopt verb transitivity as a tureandverbtransitivity(p=0.06),andfirstpasstimesshowed
randomslopeforparticipantsanditemsforallfixationmeasures a significant interaction (p < 0.05). Planned pairwise compar-
andregions,exceptforpercentregressionmeasuresinthepost- isons on first fixation durations and first pass times revealed
verbregion.Here,weremovedtheverbtransitivityrandomslope thatreadingtimesinthenon-island,intransitiveconditionwere
for participants, asthe transitivity biasvariance acrossdifferent significantly longer than in the island, intransitive condition
verbs(ifany)ismorelikelytoinfluencethedatathanvariancein (ps<0.01),butnosignificant difference wasobservedbetween
participants’experiencewiththeverbs. the transitive conditions. No significant effect wasobserved for
When the critical region demonstrated a significant inter- theregressionpathdurations.Therewasamaineffectofstruc-
action of verb and structure type, a planned comparison was tureinpercentregressions(p<0.05),withahigherpercentage
conducted with separate mixed effects models to test for sys- ofregressioninthe islandconditions, whichlikelyreflected the
tematicdifferencesbetweentheislandandnon-islandconditions greaterstructuralcomplexityintheislandconditions.
within each verb type. These models included participants and In the post-verb region, there was a marginally significant
itemsasrandomintercepts. interaction of verb and structure type (p = 0.066), but no sig-
nificanteffectwasobservedinothereye-movementmeasures.
Results
Discussion
Table3presentstheparticipantmeansoneachmeasureforeach
regionaswellasthestandarderrorsoftheparticipantmeans,and Experiment 2 used an eye-tracking during reading method to
Table4presentsasummaryofthestatisticalanalyses. investigate whether the parser uses verb transitivity informa-
tion in deciding whether to postulate a gap at the verb object
position. Firstfixationdurationsandfirstpasstimesforintran-
sitiveverbsweresignificantly longerinastructure thatallowsa
TABLE3|Experiment2participantmeanreadingtimesinmilliseconds
(standarderror)andpercentregressions. gap (non-island condition) than when the same verb appeared
in an island configuration. This effect was not observed when
Measure Pre-verbregion Verbregion Post-verbregion the critical verb was transitive. The fact that there was a read-
ingdisruption for intransitive verbsbutnot for transitive verbs
Firstfixation
is consistent with the prediction of the hyper-active gap filling
Transitive,non-island 212(8) 249(12) 242(9)
hypothesis. If the parser creates an object gap and integrates
Transitive,island 217(7) 240(7) 243(9)
the filler into the object position before having access to verb
Intransitive,non-island 207(8) 256(10) 246(10)
transitivity information, reading disruption in the non-island
Intransitive,island 208(5) 231(8) 237(7)
First-passtime intransitiveconditionshouldresultfromthemismatchbetween
Transitive,non-island 287(14) 277(13) 283(13) thepredictedtransitivityandactualtransitivityoftheverb.
Transitive,island 386(20) 275(10) 287(12) It is also important to note that in this experiment the crit-
Intransitive,non-island 299(15) 303(13) 296(14) ical mismatch effects were observed in the verb region, unlike
Intransitive,island 396(19) 266(11) 284(10) inExperiment1wherethemismatcheffectswereobservedonly
Regressionpathtime in the spill-over adverb region. This constitutes stronger evi-
Transitive,non-island 463(28) 373(24) 402(30) dence for hyper-active gap filling, because the mismatch effect
Transitive,island 636(43) 406(31) 447(35) musthaveresultedfrom properties ofthe verbitself.Theques-
Intransitive,non-island 472(38) 397(23) 492(35) tionofwhythe critical effectswereobservedinthe verbregion
Intransitive,island 619(41) 425(38) 469(26) in Experiment 2 (unlike in Experiment 1, where the effect was
Percentregressions found in the spill-over region) likely reflects task-based differ-
Transitive,non-island 33.1(5.0) 17.1(3.5) 17.9(3.4) ences whose effects are seen well beyond the current studies.
Transitive,island 33.2(4.0) 23.0(4.4) 24.4(3.4) Inhibition of the button pressing action in self-paced reading
Intransitive,non-island 27.1(4.8) 16.2(2.8) 27.5(3.7) tasks is likely more difficult than inhibition of saccades in an
Intransitive,island 32.7(4.7) 24.4(3.7) 22.4(3.1) eye-trackingtask.
FrontiersinPsychology|www.frontiersin.org 10 April2015|Volume6|Article384