Going the distance: memory and control processes in active dependency formation

Going the distance: memory and control processes in active dependency formation
Running&head:&MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&
CONSTRUCTION
&
&
&
&
&
Going&the&distance:&memory&and&control&processes&in&active&dependency&construction&
&
&
Matthew&W.&Wagers&
Department&of&Linguistics,&University&of&California,&Santa&Cruz&
&
Colin&Phillips&
Department&of&Linguistics,&Neuroscience&and&Cognitive&Science&Program&
University&of&Maryland,&College&Park&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
*Corresponding&Author:&
&
&
Matthew&Wagers&
Department&of&Linguistics&
University&of&California,&Santa&Cruz&
1156&High&Street&
Santa&Cruz,&CA&95064&
[email protected]&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
2
&
ACKNOWLEDGMENTS&
&
&
&
The&authors&gratefully&acknowledge&Shravan&Vasishth&for&helpful&analytical&advice&
&
and&discussion.&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
ABSTRACT&
&
&
PA&
&
FillerZgap& dependencies& make& strong& demands& on& working& memory& in& language&
&
comprehension& because& they& cannot& always& be& immediately& resolved.& In& a& series& of& three&
&
readingZtime& studies,& we& test& the& idea& that& these& demands& can& be& decomposed& into& active&
&
maintenance& processes& and& retrieval& events.& Results& indicate& that& the& fact& that& a& displaced&
&
phrase& exists& and& the& identity& of& its& basic& syntactic& category& both& immediately& impact&
&
comprehension& at& potential& gap& sites.& In& contrast,& specific& lexical& details& of& the& displaced&
&
phrase& show& an& immediate& effect& only& for& short& dependencies& and& a& much& later& effect& for&
&
longer& dependencies.& We& argue& that& coarseZgrained& information& about& the& filler& is& actively&
&
maintained& and& used& to& make& phrase& structure& parsing& decisions,& whereas& finerZgrained&
&
information& is& more& quickly& released& from& active& maintenance& and& consequently& has& to& be&
&
retrieved&at&the&gap&site.&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
1.&
&
Introduction&
4
&
&
&
An&adequate&description&of&the&language&comprehension&architecture&must&specify&how&the&
&
parser&encodes&and&retrieves&information&and&the&decision&principles&for&using&or&disposing&
&
of&those&encodings&–&what&Lewis&(2000)&decomposes&into&memory&and&control*processes.*The&
&
comprehension& of& fillerZgap& dependencies,& as& exemplified& by& the& relative& clause& in& (1),& has&
&
provided& a& fruitful& paradigm& for& understanding& the& interaction& between& the& memory& and&
&
control&processes&of&the&parser.&There&are&several&reasons&for&their&interest.&First,&fillerZgap&
&
dependencies& are& unbounded:& the& distance& between& filler& and& gap& can& be& arbitrarily& large,&
&
which& example& (2)& illustrates.& Consequently& a& representation& of& the& filler& must& be& durably&
&
encoded&and&reliably&retrievable&so&that&it&can&survive&spans&of&material&during&which&it&is&
not&being&processed.&
&
&
(1)&
One&clause&fillerZgap&dependency&
&
[NP&The&stones&[CP&that&the&pilgrim&[VP&placed&___&on&the&cairn&]]]&were&smooth.&
&
&
&
&
(2)&
(a)&
&
The&stones&[&that&the&monk&advised&the&pilgrim&[&to&place&___&on&the&cairn&]]&...&
(b)&
Three&clause&dependency&
&
The&stones&[&that&the&scriptures&advised&the&monk&[&to&tell&the&pilgrim&[&to&place&___&&
&
on&the&cairn&]]]&…&
&
&
Two&clause&dependency&
&
&
&
&
&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
&
PA&
Second,&fillerZgap&dependencies&are&‘islandZsensitive’;&there&are&certain&domains&that&
&
fillerZgap& dependencies& cannot& cross,& which& example& (3)& illustrates& for& phrases& in& subject&
position.&
&
&
(3)&
*The&stones&[&that&[&cairns&of&___&]&impressed&the&pilgrim&]&were&smooth.*
&
&
Ross& (1967)& identified& a& number& of& these& domains,& which& he& dubbed& ‘islands’,&
&
including&adjuncts,&coordinate&structures,&relative&clauses,&whZclauses,&and&factive&clauses.&A&
&
diverse&array&of&restrictions&thus&exist&on&where&a&gap&may&be&found.&Syntactic&theories&have&
&
attempted&to&explain&the&diversity&of&island&domains&as&the&surface&manifestation&of&fewer,&
&
more& abstract& rules& (Chomsky,& 1973,& 1986,& Manzini,& 1992,& Rizzi,& 1990).& From& the&
&
perspective&of&the&parser&it&is&sufficient&to&observe&that&not&every&argument&position&can&be&a&
&
gap& site.& This& suggests& that& the& control& processes& for& dependency& construction& must&
incorporate&a&number&of&contingencies.&
&
&
&
Finally,& fillerZgap& dependencies& are& widespread& in& natural& languages& and& cut& across&
&
many&grammatical&constructions.&Constituent&questions,&relative&clauses,&topicalizations&and&
&
comparatives,& among& others,& all& include& unbounded& dependencies& that& involve& a& relation&
&
between& a& constituent& at& the& edge& of& a& sentence& and& a& site& of& thematic& interpretation&
&
contained& within& that& sentence.& Whether& fillerZgap& parsing& is& viewed& as& consisting& in&
&
specialized& routines& or& simply& as& a& temporallyZextended& version& of& local& dependency&
&
formation,&&the&architectural&commitments&that&fillerZgap&parsing&reveals&are&likely&to&reflect&
core&properties&of&the&comprehension&system.&
&
&
&
RealZtime&fillerZgap&dependency&construction&has&been&studied&intensively&in&the&past&
&
20&years&and&a&notably&consistent&account&of&the&properties&of&dependency&construction&has&
&
emerged.&In&section&1.1,&we&briefly&review&the&evidence&that&dependency&construction&occurs&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
6
&
asZsoonZasZpossible& and& that& it& respects& the& constraints& of& the& grammar.& In& section& 1.2,& we&
&
turn& to& our& main& question,& involving& how& fillers& are& stored& and& retrieved& from& memory&
&
during&the&comprehension&process.&One&psycholinguistic&tradition,&beginning&with&Wanner&&&
&
Maratsos&(1978),&holds&that&fillers&are&maintained&in&a&distinguished&memory&state&until&they&
&
can& be& discharged& to& complete& the& dependency.& This& viewpoint& can& be& challenged&
&
empirically,&as&there&is&little&direct&evidence&that&such&a&distinguished&state&exists.&It&can&also&
&
be&challenged&conceptually,&as&some&otherwise&wellZsupported&memory&models&do&not&easily&
&
tolerate& maintenance& of& the& kind& envisioned& (McElree,& 2006).& The& alternative& is& to& assume&
&
that& the& filler& is& not& maintained,& but& that& it& is& retrieved& later& at& its& thematic& interpretation&
&
site.&&
&
&
We& have& previously& observed& that& the& locus& of& dependency& construction& appears& to&
&
shift&from&occurring&at&the&verb&to&a&nonZverbZadjacent&gap&site&when&the&length&of&the&fillerZ
&
gap&dependency&is&increased&(Wagers&&&Phillips,&2009).&This&shift&in&timing&with&respect&to&
&
heads& in& the& input& is& difficult& to& explain& on& either& a& pure& maintenance& or& pure& retrieval&
&
account.& In& a& series& of& three& experiments,& we& explore& under& what& conditions& the& timing& of&
&
dependency&construction&appears&to&shift&from&early,&preZgap&sites&to&late,&postZgap&sites.&We&
&
find& that& for& early& dependency& formation,& coarseZgrained,& categorical& details& are& reliably&
&
present&and&more&robust&than&finerZgrained&lexicallyZanchored&detail.&We&propose&a&hybrid&
&
maintenanceZretrieval& model& to& account& for& these& new& data,& in& a& way& which& combines& the&
&
insights&of&the&two&major&perspectives&on&unbounded&distance&dependency&formation.&
&
&
&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
1.1 Constructing*unbounded*dependencies*on4line&
&&
&
PA&
&
A& fairly& detailed& understanding& of& how& fillerZgap& processing& proceeds& has& been& compiled.&
&
Existing& findings& primarily& inform& us& about& control& processes,& that& is,& those& processes& that&
&
determine& when& and& under& what& circumstances& fillerZgap& dependency& formation& is&
&
attempted.& Intuitively,& processing& a& fillerZgap& dependency& requires& recognizing& the& filler,&
&
identifying& a& gap& position,& and& linking& the& filler& to& the& gap& position1.& An& important& finding&
&
regarding&fillerZgap&processing&is&that&comprehenders&attempt&to&construct&dependencies&in&
&
advance& of& unambiguous& information& about& the& gap& position.& That& is,& comprehenders&
&
anticipate& the& location& of& potential& gap& sites& and& predictively& construct& a& relation& with& the&
&
filler.& This& mechanism& is& referred& to& as& active& dependency& formation,& and& the& supporting&
&
evidence&comes&from&a&broad&variety&of&measures&and&languages&(for&review&see&Phillips&&&
&
Wagers,&2007).&One&type&of&evidence&comes&from&a&readingZtime&slowdown&in&sentences&like&
&
(4a).&In&this&sentence,&the&comprehender&encounters&an&overt&direct&object,&‘us’,&where&a&gap&
&
can& be& expected.& Reading& times& at& that& word& are& elevated& relative& to& reading& times& for& the&
&
same&word&in&a&closely&matched&sentence&that&lacks&a&fillerZgap&dependency,&where&no&gap&is&
&
expected&(4b).&This&effect&is&termed&the&filled*gap*effect*(Crain&&&Fodor,&1985;&Stowe,&1986;&
&
Lee,&2004).&
&
(4)&
a.&
My&brother&wanted&to&know&who&Ruth&will&bring&us&home&to&__&at&Christmas.&
&
b.&
My&brother&wanted&to&know&if&Ruth&will&bring&us&home&to&Mom&at&Christmas.&
&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
1
&
&
&
&
&
We use the terms gap and gap position in a theory neutral way: none of our discussion turns upon
whether the tail of an unbounded dependency is an empty category or not.
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
8
&
Another& type& of& evidence& comes& from& processing& disruptions& at& the& verb& when& the& filler&
&
would& be& a& semantically& implausible& argument& of& the& verb,& as& in& (5).& Traxler& && Pickering&
&
(1996)& found& elevated& reading& times& for& implausible& fillerZverb& combinations& at& the& verb,&
&
and& in& similar& configurations& Garnsey,& Tanenhaus& && Chapman& (1989)& observed&
&
enhancement&of&the&N400,&an&ERP&component&modulated&by&factors&associated&with&lexicoZ
&
semantic&integration&or&expectancy.&
&
(5)&
&
&
That’s& the& pistol/garage& with& which& the& heartless& killer& shot& the& hapless& man& ___&
&
yesterday&afternoon.&
&
&
&
It&is&because&processing&disruptions&precede&direct&evidence&for&the&actual&location&of&the&gap&
&
that&fillerZgap&processing&has&been&called&‘active’.&In&other&words,&comprehenders&posit&gaps&
&
eagerly,& preferring& to& construct& a& dependency& early& that& might& subsequently& have& to& be&
&
retracted,&rather&than&wait&for&unambiguous&evidence&of&the&gap&site.&Though&both&the&filledZ
&
gap& effect& and& the& semantic& anomaly& effects& observed& to& date& point& to& this& conclusion,& it& is&
&
important& to& recognize& that& that& these& effects& derive& from& different& underlying& sources& of&
&
information.&For&the&filledZgap&effect,&it&is&the&existence&of&a&dependency&(Stowe,&1986)&or&the&
&
syntactic& category& of& the& filler& (Lee,& 2004)& that& underlies& the& contrast.& For& the& semantic&
&
anomaly& effect,& the& integration& of& often& idiosyncratic& lexical& features& of& the& filler,& the& verb,&
&
and&their&relation&to&realZworld&knowledge&underlies&the&effect.&
&&
&
An& important& question& is& whether& active& dependency& formation& is& restricted& by&
&
grammatical&constraints&on&potential&gap&sites.&A&number&of&studies&have&therefore&examined&
&
whether&island&constraints&guide&the&comprehender.&Several&ERP&studies&have&shown&that&a&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
processing& disruption& occurs& if& the& parser& is& holding& an& unsatisfied& whZphrase& when& it&
&
encounters&the&edge&of&an&island&domain&(Kluender&&&Kutas,&1993;&McKinnon&&&Osterhout,&
&
1996)& or& is& forced& to& posit& a& gap& inside& an& island& (Neville,& Nicol,& Barss,& Forster,& && Garrett,&
&
1991).&Behavioural&studies&have&further&shown&that&active&dependency&formation&effects&are&
&
absent&inside&island&domains.&For&example,&Stowe&(1986)&found&no&filled&gap&effect&inside&a&
&
subject&noun&phrase.&Likewise,&Traxler&&&Pickering&(1996)&found&no&plausibility&effect&inside&
&
a&relative&clause&island,&a&finding&recently&replicated&in&secondZlanguage&learners&(Omaki&&&
&
Schulz,& 2011).& Evidence& from& additional& behavioural& studies& (Bourdages,& 1992,& Pickering,&
&
Barton&&&Shillcock,&1994,&McElree&&&Griffith,&1998)&is&consistent&with&these&results.&It&seems&
&
that&comprehenders&recognize&island&domains&and&subsequently&refrain&from&positing&gaps&
&
inside& them.& More& recently,& studies& have& shown& that& comprehenders& are& sensitive& in& realZ
&
time& to& important& subtleties& of& the& grammar& of& fillerZgap& dependencies,& including& where&
&
parasitic&gaps&may&be&licensed&(Phillips,&2006)&and&where&acrossZtheZboard&extraction&must&
&
occur& (Wagers& && Phillips,& 2009).& In& sum,& these& studies& provide& evidence& about& the& control&
&
processes&guiding&comprehension:&the&comprehender&integrates&grammatical&knowledge&in&
guiding&its&expectations&about&where&it&may&complete&the&fillerZgap&dependency.&
&
&
&
A&relatively&straightforward&description&thus&emerges&for&the&control&processes&that&
&
support& fillerZgap& dependency& construction.& First,& comprehenders& attempt& to& terminate& an&
&
open& fillerZgap& dependency& wherever& the& grammar& allows.& Second,& the& comprehension&
&
mechanism& does& not& ‘wait& and& see’& where& the& actual& gaps& in& a& sentence& are,& but& posits& the&
&
gaps& wherever& they& might& occur,& based& on& the& partial& input& and& the& upcoming& structure& it&
entails.&&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
1.2
&
Memory*resources*for*filler4gap*dependencies*
10
&
&
&
The& temporal& delay& between& the& encoding& of& the& filler& and& potential& sites& in& the& sentence&
&
where&it&can&be&integrated&may&be&substantial.&More&importantly,&this&delay&may&be&occupied&
&
by&potentially&irrelevant&processing&operations.&For&example,&in&(6),&after&the&comprehender&
&
processes& the& filler,& she& must& (at& least)& assemble& the& subject& phrase,& attach& its& predicate,&
&
build& that& verb’s& complement& clause,& and& link& the& main& clause& subject& to& its& thematic& role&
assigner&in&the&embedded&clause&–&all&of&this&before&the&gap&site&is&encountered.&
&
&
(6)&&
&
Which&shrines&is&the&famous&monk&most&likely&to&have&visited&___&?&
&
&
&&
&
Because& it& participates& in& nonZlocal& dependencies,& the& encoding& of& the& filler& has& to& be&
&
sufficiently& durable& to& survive& arbitrary& spans& of& material& during& which& it& is& not& being&
&
processed.& Note& that& this& problem& is& true& both& of& cases& where& there& is& considerable&
&
hierarchical& distance& between& the& filler& and& the& gap& site,& but& also& of& cases& where& the&
&
hierarchical&distance&is&minimal&and&the&temporal&delay&is&nonetheless&substantial,&as&in&(7),&
where&the&subject&is&modified&by&a&prepositional&phrase.&
&
&
(7)&&
&
&
Which&shrines&did&the&monk&from&the&ancient&monastery&visit&____?&
&
&
&
There&are&at&least&two&architectural&possibilities&for&how&the&task&of&integrating&a&filler&with&
&
its&gap&host&can&be&accomplished.&Under&a&maintenance*view,*the&filler&is&stored&in&a&shortZ
&
term&memory&location&until&the&point&at&which&it&can&be&integrated&directly&with&a&gap&host.&
&
Under& a& retrieval* view,& the& filler& is& not& purposely& kept& in& shortZterm& memory& and& must& be&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
retrieved& and& reactivated& when& the& parser& posits& a& gap.& According& to& the& maintenance&
&
account,&the&filler&is&continuously&accessible&to&the&parser&but&potentially&exacts&an&overhead&
&
cost& by& consuming& the& comprehender’s& shortZterm& memory& capacity.& According& to& the&
&
retrieval& account,& there& is& no& taxing& maintenance& but& there& is& a& possibility& for& error& later&
&
when& the& filler& has& to& be& reactivated.& Both& kinds& of& account& can& be& found& in& the& literature,&
&
though& there& has& been& relatively& little& attention& paid& to& distinguishing& them& (with& the&
&
notable& exception& of& McElree,& Foraker,& && Dyer,& 2003).& In& part& this& is& because& experimental&
&
design,& methodologies,& and& linguistic& manipulations& have& varied& between& studies,& making&
&
direct&comparisons&difficult.&Secondly,&despite&a&diversity&of&measures&and&the&fact&that&a&key&
&
property& of& fillerZgap& dependencies& is& their& unboundedness,& a& relatively& narrow& range& of&
&
dependency& lengths& has& been& examined.& As& we& will& argue& in& a& moment,& at& very& short&
dependency&lengths&the&two&accounts&are&difficult&to&distinguish&empirically.&
&
&
&
&In& the& remainder& of& this& section& we& will& review& some& of& the& major& accounts& of& the&
&
memory& resources& deployed& in& fillerZgap& comprehension.& Then& we& will& lay& out& the& logic& of&
&
our& study,& whose& goal& is& to& disentangle& maintenance& versus& retrieval& accounts& of& fillerZgap&
&
processing& by& examining& the& effects& dependency& length& has& on& differing& linguistic&
manipulations.&&
&
&
&
One& influential& maintenance& proposal& holds& that& the& filler& phrase& is& encoded& in& a&
&
distinguished& memory& register& until& it& can& be& assigned& a& thematic& role& in& the& sentence.&
&
Wanner& && Maratsos& (1978)& proposed& an& Augmented& Transition& Network& (ATN)& model& of&
&
parsing&which&included&a&separate&buffer,&called&the&HOLD&cell,&for&displaced&NPs.&When&an&NP&
&
is&identified&as&the&head&of&a&relative&clause,&the&contents&of&the&NP&are&placed&in&the&HOLD&cell.&
The&
HOLD&
&
cell& allows& the& assignment& of& grammatical& function& to& be& put& off& until& an&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
12
&
appropriate&context&is&identified.&&For&example,&in&an&objectZextracted&relative&clause,&after&a&
&
transitive&verb&has&been&processed&the&ATN&attempts&to&analyze&subsequent&input&as&a&direct&
&
object& NP.& & If& it& fails& to& find& an& overt& NP& in& the& input,& as& would& be& the& case& if& the& clause&
&
contains&a&gap,&then&the&system&checks&to&see&whether&or&not&the& HOLD&cell&is∅&if&it&is&
&
not&empty,&the&system&retrieves&its&contents,&treating&them&as&input.&
&
&
This& parsing& sequence& does& not& qualify& as& an& active& strategy,& because& a& gap& is& not&
&
postulated& unless& the& parser& fails& to& encounter& a& verb’s& lexical& argument.& But& that& is& a& fact&
&
about& the& parser’s& control& structure,& which& can& be& modified& independently& of& the& memory&
&
architecture.&Indeed&Frazier&(1987)&suggested&that&the&nonZemptiness&of&the& HOLD&cell&itself&
&
could& serve& as& a& signal& to& postulate& a& gap& before& lexical& arguments& are& encountered.& This&
&
mechanism&is&formulated&as&the&Active&Filler&Strategy,&which&states&that&once&a&filler&has&been&
&
encountered&the&parser&is&“immediately&predispose[d]”&to&rank&gaps&more&highly&than&lexical&
&
arguments& (Frazier& && Flores& d’Arcais,& 1989).& Frazier& and& Flores& d'Arcais& suggest& that& this&
&
implies&a&filler&that&is&not&‘inert’.&At&predicates&following&the&filler&the&parser&considers&gap&
&
analyses& before& analyses& with& an& overt& lexical& argument& because& the& unintegrated& filler& is&
&
effectively& always& an& input.& Considered& in& this& way,& dependency& completion& may& be& active&
&
because&the&filler&is&available&in&the&processing&workspace&before&subsequent&inputs,&and&it&
&
therefore&effectively&outZcompetes&incoming&categories&for&attachment.&
&
&
There& have& been& many& demonstrations& that& fillerZgap& dependencies& exact& a& cost& on&
&
the& comprehender& (Wanner& && Maratsos,& 1978,& King& && Just,& 1991,& Gibson,& 1998,& 2000,&
&
Sprouse,&Wagers,&&&Phillips,&2012),&a&fact&sometimes&taken&as&evidence&for&the&maintenance&
&
view.&Several&electrophysiological&studies&have&also&been&brought&to&bear&on&the&question&of&
&
memory& load& during& an& open& dependency.& Both& King& && Kutas& (1995)& and& Fiebach,&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
Schlesewsky&&&Friederici&(2002)&found&that&in&objectZextracted&fillerZgap&dependencies&the&
&
averaged&EEG&record&shows&a&sustained&anterior&negativity&(SAN).&This&effect&is&modulated&
&
by&performance&and&participants’&memory&span&and&has&been&implicated&in&explicit&memory&
&
load&tasks&(Ruchkin,&Johnson,&Canoune&&&Ritter,&1990).&For&these&reasons,&the&presence&of&a&
&
sustained& anterior& negativity& in& open& fillerZgap& dependencies& has& been& interpreted& as& a&
&
direct&reflection&of&the&memory&load&consumed&by&actively&maintaining&the&filler.&However,&
&
this& interpretation& may& be& undercut& by& the& fact& that& the& SAN& does& not& reflect& a& cumulative&
&
effect& that& accrues& at& each& word,& but& instead& derives& mainly& from& the& first& words& of& the&
&
dependency&(King&&&Kutas,&1995;&Phillips,&Kazanina,&&&Abada,&2005).&Importantly,&it&remains&
&
an&open&question&what&the&actual&content&of&the&filler&is&when&the&dependency&is&unsatisfied.&
&
In& discussing& the& SAN,& Fiebach& et& al.& (2002)& point& out& that& the& electrophysiological& effect&
&
does& not& choose& between& alternative& accounts& of& what& precisely& is& being& maintained& in& a&
&
privileged& state.& It& could& be& a& full& semantic& or& syntactic& representation& of& the& filler,& or&
&
perhaps& just& a& few& features.& Alternatively,& it& may& not& be& the& content& of& the& filler& that& is&
&
represented& at& all,& but& rather& the& prediction& for& a& category& that& allows& completion& of& the&
dependency,&as&in&Dependency&Locality&Theory&(Gibson&2000).&
&
&
&
In&contrast&to&a&pure&maintenance&account,&and&supporting&a&retrieval&account,&there&
&
is& evidence& that& semantic& features& of& the& filler& are& not& maintained& and& must& be& reactivated&
&
later&into&a&state&suitable&for&integration.&Swinney&and&colleagues&have&shown&that&a&filler’s&
&
semantic&associates&are&primed&in&a&crossZmodal&lexical&decision&immediately&following&the&
&
introduction& of& the& filler& in& the& sentence& (Nicol& && Swinney& 1989;& Nicol,& Fodor& && Swinney&
&
1994).&However&this&priming&does&not&persist,&declining&across&the&sentence.&Priming&is&once&
&
again&observed&following&the&verb.&McElree&(2001)&reports&a&similar&pattern&using&a&probe&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
14
&
recognition& task.& The& fact& that& priming& is& not& observed& in& dependencyZmedial& positions&
&
suggests& that& either& the& filler& contents& are& not& actively& maintained& or& that,& in& their&
&
maintained& form,& they& cannot& influence& lexical& decision2.& These& findings& demonstrate& that&
&
some& properties& of& the& filler& are& not& equally& available& throughout& the& span& of& the&
&
dependency,&in&support&of&a&retrieval&view.&
&
The& possibility& that& no& filler& contents& are& actively& maintained& is& supported& by& the&
&
notion& that& there& is& a& fairly& stringent& limit& on& the& amount& of& information& that& can&
&
concurrently& occupy& shortZterm& memory.& There& is& good& empirical& support& for& such& a&
&
constraint& (Broadbent,& 1958,& McElree& && Dosher,& 1989,& Cowan,& 2001),& which& is& argued& by&
&
many&authors&to&correspond&to&an&architectural&component,&the&‘focus&of&attention’&(McElree,&
&
2006,& Oberauer& && Kliegl,& 2006).& The& focus& of& attention& is& sufficiently& limited& so& as& to&
&
effectively&exclude&information&unassociated&with&local&processing&events,&like&maintaining&a&
&
filler& phrase& while& other& constituents& of& the& sentence& are& being& interpreted.& If& the& filler’s&
&
contents& are& not& available& in& the& focus& of& attention,& then& they& must& be& retrieved& to& be&
&
integrated&with&the&verb&(McElree&et&al.,&2003,&Lewis&&&Vasishth,&2005).&Retrieval&accounts&
&
must& thus& answer& the& question& of& how& the& correct& constituent& is& retrieved& from& memory&
&
once& a& suitable& verb& is& reached.& Several& sources& for& these& cues& are& supported& in& the&
&
literature.& Cues& for& retrieval& could& in& principle& derive& from& specific& lexical& or& contextual&
&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
&
2& McKoon& && Ratcliff& (1994)& object& to& the& reactivation& interpretation& of& crossZmodal& lexical&
&
priming& in& these& sentential& contexts& on& other& grounds.& They& argue,& instead,& that& the&
goodnessZofZfit&of&the&lexical&decision&target&is&an&important&determinant&of&the&RT&patterns&
&
in&these&experiments.&Thus&lower&RTs&are&observed&in&postZverbal&positions&not&because&of&
priming&from&a&reactivated&filler,&but&because&the&lexical&decision&target&is&itself&a&noun&that&
&
could& occur& in& the& postZverbal& position.& The& probe& recognition& study& of& McElree& (2001)& is&
less&vulnerable&to&this&objection,&since&the&probe&words&were&adjectives&and&the&task&was&to&
&
determine&whether&a&synonymous&word&was&encountered&in&the&prior&context.&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
information&supplied&by&the&verb&and&other&heads&near&the&gap&site.&For&example,&a&verb&like&
&
amuse&selects&for&an&animate&internal&argument&and&therefore&animacy&is&a&plausible&cue&for&
&
retrieval&of&the&filler.&The&evidence&that&we&review&below&suggests&that&some&but&not&all&kinds&
&
of&verbZparticular&information&constrain&the&retrieval&of&the&filler.&The&cues&used&to&retrieve&
&
the&filler&could&also&derive&from&grammatical&rules&or&parsing&routines,&leading&to&cues&that&
&
are& independent& of& particular& sentences.& For& example,& filler& phrases& are& typically& (but& not&
&
always)&of&category&NP&and&they&(always)&occupy&a&specifier&position&high&in&the&clause.&Thus&
those&cues&could&be&used&alongside&the&lexicallyZspecific&cues.&
&
&
An&attractive&possibility&is&that&a&verb’s&selectional&restrictions&provide&the&necessary&
&
cues& to& retrieve& the& filler.& The& semantic& anomaly& studies& discussed& above& (Garnsey& et& al.,&
&
1989,&Traxler&&&Pickering,&1996)&provide&a&useful&constraint&on&this&idea.&In&those&studies,&an&
&
anomalous& combination& of& verb& and& filler& was& detected& either& at& the& verb& or& shortly&
&
thereafter.& In& Traxler& && Pickering’s& study& the& plausibility& contrast& was& evident& in& first&
&
fixations.& This& suggests& that& even& if& a& filler& is& not& a& good& object& for& the& verb,& it& may&
&
nonetheless&be&retrieved&at&the&verb&position.&Caution&is&required&here,&as&previous&studies&
&
have& not& entirely& distinguished& between& purely& selectional& restrictions& and& violations& of&
&
realZworld&expectations.&For&example,&the&verb&“punish”&requires&an&animate&theme&and&thus&
&
“punish& the& rock”& illZformed.& In& contrast,& a& phrase& like& “shoot& the& garage”& (Traxler& &&
&
Pickering,& 1996)& is& anomalous& because& it& is& unusual,& but& not& impossible.& Based& on& the&
&
absence& of& plausibility& mismatch& effects,& Boland,& Tanenhaus,& && Garnsey& (1995)& found& that&
&
active&dependency&completion&was&blocked&for&implausible&verbZobject&combinations&if&the&
&
verb&was&a&verb&like&‘remind’&and&the&displaced&object&was&an&inanimate&noun&like&‘movie.’&
&
Because&verbs&like&‘remind’&can&combine&with&both&a&NP&object&and&a&clause,&the&parser&may&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
16
&
thus&be&influenced&by&the&possibility&that&the&filler&could&be&combined&with&a&plausible&host&in&
&
the&upcoming&clause.&In&this&way&the&semantic&anomaly&effect&could&be&blocked.&Pickering&&&
&
Traxler& (2001)& provided& data& convergent& with& Boland& et& al.& (1995)& and& argued& that& these&
&
results&favor&a&mechanism&by&which&selectional&features&could&be&used&to&filter&dependency&
&
completion.& In& the& example& given,& the& filler& ‘movie‘& lacks& the& animacy& feature& that& ‘remind’&
&
selects& for& in& its& NP& argument.& Consequently& the& parser& blocks& combination& with& the& filler&
&
early.&&
&
Subcategorization& information& may& be& also& be& used& as& a& cue& and& indeed& Van& Dyke& &&
&
McElree&(2006)&have&suggested&more&generally&that&syntactic&cues&may&gate&semantic&cues&in&
&
the& retrieval& of& constituents& from& memory.& A& crucial& case& to& test& this& claim& is& intransitive&
&
verbs,&like&‘arrive’,&which&do&not&take&a&direct&object.&Presently&evidence&remains&somewhat&
&
mixed& on& whether& fillers& give& rise& to& active& effects& at& verbs& that& are& purely& intransitive&
&
(Staub,&2007,&Omaki,&Lau&&&Phillips,&2011).&However,&Frazier&&&Clifton&(1989)&and&Pickering&
&
&& Traxler& (2003)& have& both& found& evidence& that& fillers& give& rise& to& active& effects& for& verbs&
&
that&are&preferred&intransitives&but&nonetheless&optionally&transitive&(cf.&Stowe&et&al.,&1991).&
&
This&suggests&that&active&dependency&completion&can&lead&the&parser&to&postulate&gaps&that&
&
the&verb’s&argument&structure&statistics&might&otherwise&disprefer.&
&
The& idea& that& lexicallyZspecified& information& alone& cues& retrieval& meets& a& further&
&
challenge&in&some&recent&data&from&Wagers&&&Phillips&(2009).&They&compared&the&semantic&
&
anomaly& effect& for& short& versus& long& dependencies.& Dependencies& were& serially& lengthened&
&
by& adding& a& fiveZword& prepositional& phrase& inside& the& subject.& In& conditions& with& short&
&
dependencies& there& was& an& immediate& anomaly& effect& at& the& verb,& whereas& in& conditions&
&
with&long&dependencies&the&anomaly&effect&was&absent&at&the&verb&and&was&not&reliable&until&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
after& the& gap& site& (an& indirect& object& gap& that& occurred& a& few& words& later& in& the& sentence).&
&
This&finding&is&surprising&because&one&might&initially&assume&that&effects&that&are&triggered&
&
by&verb&retrieval&cues&ought&to&consistently&occur&at&the&same&position,&i.e.,&near&the&verb.&&&If&
&
the& retrieval& occurs& much& later& than& the& verb,& the& question& arises& how& the& verb& specific&
&
information&itself&survives&in&the&focus&of&attention.&The&shift&in&the&locus&of&the&effect&raises&
&
the& possibility& that& active& dependency& processing& occurs& only& at& very& short& distances& and,&
&
possibly,& that& previous& studies& may& have& systematically& undersampled& longerZdistance&
&
dependencies.&Past&studies&of&fillerZgap&dependency&construction&have&indeed&been&heavily&
&
skewed&to&very&short&distances.&A&sample&of&21&experiments3&performed&between&1986&and&
&
2004& revealed& that,& on& average,& 2.9& words& or& 1.8& constituents& linearly& intervened& between&
&
filler& and& gap.& These& experiments& spanned& different& paradigms:& 10& selfZpaced& reading&
&
studies,& 5& eyeZtracking& studies,& 3& sensicality& monitoring& studies,& 3& crossZmodal& priming&
&
studies.&The&strong&bias&to&examine&short&dependencies&calls&into&question&the&generality&of&
&
the& idea& that& fillerZgap& dependency& formation& reliably& occurs& at& verbs& that& intervene&
&
between& fillers& and& gaps.& & If& there& is& variation& in& the& timing& of& fillerZgap& dependency&
&
completion,& then& the& richness& of& the& set& of& retrieval& cues& available& to& the& comprehender&
might&vary&depending&on&when&the&retrieval&actually&occurred&with&respect&to&the&verb.&
&
&
&
&
&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
&
3&Tanenhaus,&Stowe&&&Carlson&(1985);&Stowe&(1986);&Swinney&et&al.&(1988);&Frazier&&&Clifton&
&
(1989);& Boland& et& al.& (1995);& Pickering,& Barton& && Shillcock& (1994);& Pickering& && Traxler&
(1996);& Traxler& && Pickering& (1996);& Clahsen& && Featherston& (1999);& McElree& (2000);&
&
Sussmann&&&Sedivy&(2003);&Aoshima,&Phillips&&&Weinberg&(2004);&Conklin,&Koenig&&&Mauner&
(2004);&Lee&(2004)&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
1.3
The*current*study*
18
&
&
The&research&reviewed&above&presents&two&different&views&of&how&unbounded&dependencies&
&
are&completed.&On&one&view,&the&filler&phrase&is&actively&maintained&throughout&the&course&of&
&
the&sentence.&By&virtue&of&this&(potentially&costly)&maintenance,&a&pressure&arises&to&rapidly&
&
complete&the&dependency&and&attachments&are&attempted&at&grammaticallyZlicit&gap&sites.&On&
&
the& other& view,& maintaining& the& filler& phrase& exceeds& the& system’s& capacity& and& it& must& be&
&
removed&from&the&focus&of&attention&until&it&is&reactivated&at&a&potential&gap&host.&Although&
&
discussions& of& unbounded& dependency& processing& have& tended& to& adopt& the& language& of&
&
either& the& maintenance& or& retrieval/reactivation& view,& there& is& a& natural& compromise&
&
(articulated&by&Fiebach&et&al.,&2002,&for&example).&Namely,&maintenance&could&apply&to&some&
&
but& not& all& features& of& the& filler& encoding,& with& reactivation& applying& to& its& unmaintained&
&
components.& The& present& study& seeks& to& explore& a& hybrid& maintenance/retrieval& view& by&
&
testing&what&aspects&of&a&filler&phrase&contribute&to&active&dependency&formation&at&longer&
&
dependency&lengths&and&what&aspects&become&attenuated.&
&
A&decomposition&of&fillerZgap&dependency&formation&into&more&fineZgrained&operations&
&
would& seem& to& be& necessary& to& answer& the& questions& about& memory& contents& raised& in&
&
Section& 1.2.& In& the& present& set& of& studies,& we& test& active& dependency& formation& at& different&
&
serial& lengths,& as& in& Wagers& && Phillips& (2009),& but& we& also& use& three& different& indices& of&
&
dependency& formation,& each& of& which& presumably& draws& upon& different& sources& of&
&
information:& the& filledZgap& effect,& the& semantic& anomaly& effect,& and& an& index& based& on&
&
subcategorization.&&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
2
2.1
&
Experiment&1:&The&[email protected]&Effect&
Rationale*
PA&
&
&
In& the& first& experiment,& we& used& the& filledZgap& effect& design& to& test& whether& active&
dependency&formation&persists&over&long&dependency&lengths.&&&
&
&
&
&
We& used& a& modified& version& of& the& filledZgap& design& devised& by& Lee& (2004),& which&
&
manipulates&the&need&for&active&gap&search&while&holding&constant&the&presence&of&a&filler&in&
&
memory.&Lee&contrasted&displaced&noun&phrases&(NPs)&with&displaced&prepositional&phrases&
&
(PPs).& In& unmarked& word& order& only& displaced& NPs& can& occupy& subject& or& object& position.&
&
Relative&to&reading&a&displaced&PP,&reading&a&displaced&NP&should&cause&difficulty&if&subject&or&
&
object&position&is&filled&by&another&phrase.&Example&(8)&illustrates&a&simplified&version&of&our&
experimental&contrast.&
&
(8)&
(a)&
The&stones!which&the&pilgrim&toppled&the&cairn&for&___&…&
&
&
(b)&
The&stones&for&which&the&pilgrim&toppled&the&cairn&___&…&
&
&
In&both&(8a)&and&(8b)&it&is&possible&to&form&a&dependency&at&the&critical&verb&(‘topple’).&But&
&
only& when& the& displaced& category& is& a& PP& (8b)& can& readers& unambiguously& predict& the& gap&
&
site.&If&the&comprehender&hypothesizes&that&the&gap&occupies&the&direct&object&position&at&the&
&
verb& in& (8a),& then& encountering& the& overt& NP& ‘the& cairn’& should& disconfirm& that& hypothesis.&
&
Meanwhile,& the& PP& category& of& the& filler& (8b)& should& prevent& the& comprehender& from&
&
hypothesizing& that& there& is& a& gap& in& direct& object& position.& Consequently& processing& the&
&
direct& object& is& predicted& to& be& easier& in& (8b)& than& in& (8a).& Lee’s& design& represents& an&
&
improvement& over& previous& filledZgap& effect& designs,& because& it& contrasts& sentences& that&
both&have&fillerZgap&dependencies&and&identical&interpretations.&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
20
&
We&created&contrasts&like&those&in&(8)&for&three&different&dependency&lengths:&a&short&
&
singleZclause& dependency,& a& longer& singleZclause& dependency,& and& a& longer& biZclausal&
&
dependency.& In& this& design& the& filledZgap& effect& becomes& a& test& of& whether& category&
&
information& is& maintained& over& the& course& of& a& longZdependency.& A& sample& materials& set& is&
&
given&in&Table&1.&&
<<&INSERT&TABLE&1&HERE&>>&
&
&
&
2.2 Participants,&Materials&and&Methods&
Participants&were&36&native&speakers&of&English&from&the&University&of&Maryland&community.&
&
Each&received&$10&for&taking&part&in&the&experiment,&which&lasted&45&minutes.&
&
&
This& experiment& crossed& the& factors& Filler* Category& and& Length& in& a& 2×3& factorial&
&
design.&Filler&Category&was&either&‘NP’&or&‘PP’.&Length&was&either&‘Short,&‘+PP’,&or&‘+CP’&(where&
&
CP&stands&for&Complementizer&Phrase,&the&syntactic&category&of&an&embedded&sentence).&In&
&
the&Short&condition&a&twoZword&subject&and&an&adverb&separated&the&filler&from&the&verb.&In&
&
+PP&conditions&the&subject&was&modified&by&a&fiveZword&prepositional&phrase&(PP).&Finally,&in&
&
+CP&conditions,&the&sentence&inside&the&Short&condition&relative&clause&was&embedded&under&
&
a&verb&like&‘say’&or&‘think’,&verbs&that&do&not&readily&host&a&NP&object.&We&chose&such&verbs&in&
&
order& to& minimize& the& possibility& of& premature& dependency& completion& at& the& main& clause&
&
verbs.& Like& the& +PP& condition,& the& embedding& clause& was& comprised& of& five& words.& Table& 1&
&
illustrates& one& full& materials& set.& The& full& materials& list,& for& all& experiments,& is& given& in& the&
&
supplementary&materials.&
&
The&object&NP&comprised&the&critical&region&in&this&design&because&it&was&where&a&filledZ
&
gap& effect& could& first& occur.& Object& NPs& were& always& threeZword& sequences& of& the& form&
&
‘determinerZadjectiveZnoun’& or& ‘determinerZnounZnoun’& (i.e.,& a& nounZnoun& compound).& The&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
relative& clause& always& contained& an& adverbial& phrase& in& final& position& (e.g.,& in& the& example&
&
above,&“while&wearing&a&mask”).&This&phrase&provided&a&strong&signal&that&a&constituent&was&
missing,&by&means&of&the&sequence&preposition4preposition&(e.g.,&“for&while”).&
&
&
Finally,& the& Short& conditions& were& always& preceded& by& a& fiveZword& embedding&
&
preamble&so&that&the&position&of&the&critical&region&was&ordinally&matched&across&conditions.&
&
For&the&sample&set&in&Table&1,&the&Short&preamble&was&“The&biologist&was&distressed&that&…“.&
&
Sentences& were& presented& on& a& desktop& PC& using& the& Linger& software& package& (Doug&
&
Rohde,& MIT)& in& a& selfZpaced,& wordZbyZword,& moving& window& paradigm& (Just,& Carpenter,& &&
&
Woolley,& 1982).& Each& trial& began& with& a& screen& presenting& a& sentence& in& which& the& words&
&
were& masked& by& dashes& while& spaces& and& punctuation& remained& intact.& Each& time& the&
&
participant& pressed& the& space& bar& a& word& was& revealed& and& the& previous& word& was& reZ
&
masked.& A& yes/no& comprehension& question& appeared& all& at& once& on& the& screen& after& each&
&
sentence.&The&‘f’&key&was&used&for&‘yes’&and&the&‘j’&key&was&used&for&‘no’.&OnZscreen&feedback&
&
was&provided&for&incorrect&answers.&Participants&were&instructed&to&read&at&a&natural&pace&
&
and& answer& the& questions& as& accurately& as& possible.& In& all& of& the& selfZpaced& reading&
&
experiments&presented&here,&participants&were&never&informed&that&sentences&would&contain&
&
grammatical& errors.& Order& of& presentation& was& randomized& for& each& participant.& 7& practice&
items&were&presented&before&the&beginning&of&the&experiment.&
&
&
2.3 Analysis&
&
&
The&following&analysis&procedure&was&applied&to&all&data&sets&reported&in&this&paper&
&
so&we&describe&it&in&detail&here.&&
&
&
Both&RT&and&accuracy&data&were&analyzed&with&mixedZeffects&modeling.&We&estimated&
&
models&using&the&lme4*package&(Bates,&2011)&in&the&R&software&environment&(R&Development&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
22
&
Core& Team,& 2011).& Accuracy& data& were& treated& with& a& logistic& model;& RT& data& were&
&
logarithmically&transformed&before&being&entered&into&a&linear&model.&Both&experimentallyZ
&
manipulated& factors& were& coded& with& orthogonal& contrasts.& The& two& levels& of& the& Filler*
&
Category&condition&were&coded&with&sum&contrasts&(±1;&where&+1&represents&NP&Fillers).&The&
&
three&levels&of&Length&were&coded&using&Helmert&contrasts:&Short&conditions&were&compared&
&
to& the& mean& of& +PP/+CP& conditions& (reported& as& the& Length.Long& coefficient);& and& +PP&
&
conditions& were& directly& compared& to& +CP& conditions& (reported& as& the& Length.Clause&
&
coefficient).& This& coding& makes& sense& in& view& of& the& questions& we& posed:& are& short&
&
dependencies&different&from&long&dependencies?&Are&different&kinds&of&length&relevant?&&
&
Reading& time& data& were& treated& for& outliers& in& the& following& manner,& adopting&
&
suggestions&of&Baayen&&&Milin&(2010).&Initially&no&observations&were&excluded&from&the&RTs.&
&
Linear&models&were&fit&to&the&log&reading&times&and&the&residuals&were&inspected&for&fit&to&the&
&
normal&distribution.&Observations&with&absolute&standardized&residuals&greater&than&2&were&
&
then&excluded.&This&method&thus&identifies&outliers&on&the&criterion&of&how&much&they&stress&
&
the& assumptions& of& the& statistical& model.& On& average,& our& analyses& excluded& 5%& of&
&
observations.& Results& of& untrimmed& and& residualZtrimmed& analyses& are& presented& as&
&
coefficient&tables4.&&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
&
&
4& In& any& experiment& of& modest& complexity,& a& number& of& implicit& choices& are& made& by& the&
&
analyst,&more&than&the&explicit&steps&reported&in&a&research&report.&Simmons&et&al.&(2011)’s&
&
recent& study& illustrates& how& such& flexibility& in& the& experimental& pipeline& can& lead& to&
&
increased&effective&falseZpositive&rates.&We&have&attempted&to&disclose&all&major&steps&in&the&
&
analysis&of&the&data&we&present.&In&the&supplementary&materials,&we&supply&coefficient&tables&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
We& first& fit& models& with& crossed& random& factors;& these& sometimes& failed& to& converge&
&
and&never&presented&a&different&profile&of&coefficients&or&t/zZvalues.&We&report&therefore&the&
&
simpler& models& with& subject/item& intercepts& only.& PZvalues& were& determined& by& Markov&
&
Chain& Monte& Carlo& sampling& of& the& posterior& distribution& from& the& model& with& no& nested&
experimental&factors,&using&the&pvals.fnc*function&in&R&(Baayen,&2008).&
&
&
&
&
2.4 Results&
&
2.4.1& Comprehension&accuracy&
&
&
Mean&accuracy&was&84%.&Comprehension&accuracy&by&condition&is&reported&in&Table&
&
2.& There& was& no& effect& of& Filler& type.& The& only& reliable& contrasts& were& in& Length:& both&
&
between&Short&conditions&and&the&average&of&the&long&conditions&(Length.Long:&0.58&±&0.23,&z*
&
=& 2.5,& p<.05)& and& between& the& two& long& conditions.& (Length.Clause:& 0.89& ±& 0.24,& z& =& 3.8,&
&
p<.001).&A&paired&comparison&between&just&the&Short&and&+PP&conditions&was&not&significant.&
&
In&other&words:&(1)&+CP&lengthened&sentences&were&harder&to&understand;&(2)&the&filler&type&
&
did&not&affect&performance&on&the&comprehension&questions.&
&
&
&
2.4.2& Reading&times&
&
&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
&
for&the&models&over&untrimmed&data.&On&the&first&author’s&web&site,&we&supply&the&raw&data&
&
accompanied&by&R&scripts&used&to&process&it,&so&that&the&reader&can&replicate&or&modify&the&
analyses&reported&here.&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
24
&
Reading& time& results& are& reported& in& Figure& 1& as& logZtransformed& RTs& from& the&
&
untrimmed&dataset.&Words&14Z16&are&annotated&as&“active&filling&regions”,&since&these&occur&
&
before& direct& evidence& of& the& gap& position.& Words& 18Z20,& occurring& after& unambiguous&
&
evidence&in&the&input&for&the&gap&position,&are&annotated&as&“gapZdriven”&regions.&&Below&we&
&
report&4&regions&of&analysis:&the&verb&(Word&13)&and&preZverb&region&(Word&12),&the&active&
&
filling& region,& and& the& gapZdriven& region.& Coefficient& tables& for& the& linear& mixedZeffects&
&
models& on& this& regions& are& reported& in& Table& 3.& In& brief,& evidence& of& active& dependency&
&
construction&was&found&for&all&dependency&lengths,&with&some&variation&in&the&timing&and&the&
&
size&of&the&effect.&&
&
&
&
No&effects&were&observed&on&the&verb,&though&there&was&a&tendency&for&PP&conditions&
&
to&be&read&more&quickly.&This&suggests&that&we&had&a&good&baseline&for&evaluating&the&filledZ
&
gap& effect& in& the& critical& active& filling& region.& An& interesting& pattern& of& reading& times& in& the&
&
adverb& deserves& comment,& though& it& does& not& seem& to& impinge& upon& our& interpretation& of&
&
the&filledZgap&effect:&Short&conditions&were&read&more&quickly&than&either&long&condition,&and&
&
there&was&an&interaction&with&Filler&Category.&Inspection&of&the&means&suggests&this&reflects&a&
&
slowZdown&for&PP&fillers&in&the&PP&length&condition&(t&=&Z1.9,&p&<&.05).&Our&ability&to&interpret&
&
this&interaction&confidently&is&limited,&since&items&were&not&matched&prior&to&the&adverb&(a&
&
consequence& of& our& design).& The& difference& may,& however,& reflect& the& fact& that& two& PP&
&
constituents& were& recently& processed& (the& filler,& and& the& subject& postZmodifier).& & If& we&
&
incorporate& reading& times& from& the& previous& word& as& a& covariate,& the& pairwise& difference&
&
between&the&two&Filler&conditions&in&+PPZlength&conditions&is&no&longer&significant&(t*=&Z&1.3,&
&
p&=&.19),&suggesting&that&spillZover&is&a&relevant&concern.&&&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
&
PA&
The& critical& region& of& interest& is& the& Object& NP,& which& we& dubbed& the& ‘active& filling’&
&
region.& Here& we& observed& a& significant,& positive& difference& between& NP& conditions& and& PP&
&
conditions.& We& interpret& this& as& a& filled& gap& effect,& following& Lee& (2004).& Overall,& CP&
&
conditions& were& also& read& more& slowly& than& PP& conditions.& Crucially,& there& was& no&
&
interaction&of&the&Filler&effect&with&either&of&the&Length&contrasts.&Inspection&of&the&wordZbyZ
&
word&means&in&Figure&1&makes&it&clear&that,&for&each&of&the&Length&conditions,&there&was&a&
&
more&punctate&effect&that&varied&in&onset:&Region&15&for&Short&conditions,&Region&16&for&+PPZ
&
length&conditions,&and&Region&14&for&+CPZlength&conditions.&The&effect&sizes&for&each&of&the&
&
maximal& pairwise& comparisons,& expressed& as& difference& in& mean& log& RT& normalized& by&
&
standard&deviation&log&RT&(Cohen’s&d),&was&0.16&for&Short&conditions,&0.08&for&PP&conditions,&
and&0.14&for&CP&conditions.&&*
&
&
&
&Finally,&the&gapZdriven&region&was&defined&as&the&three&words&following&unambiguous&
&
evidence&for&the&gap.&Overall,&Short&conditions&were&read&more&quickly.&There&was&no&overall&
&
effect&of&Filler&Category,&though&there&was&a&nonZsignificant&tendency&for&NP&filler&conditions&
&
to& be& read& more& quickly.& However,& an& interaction& between& Filler& Category& and& Length&
&
amplified& the& NP& filler& speedZup& in& the& Short& condition& and& reversed& it& in& the& +PPZlength&
conditions:&this&is&apparent&in&the&RT&‘spike’&in&Figure&1&for&the&NP&conditions.&
<<&INSERT&TABLE&3&HERE&>>&
<<&INSERT&FIGURE&1&HERE&>>&
&
&
&
&
&
&
2.5
Discussion*
&
&
In&this&experiment&we&tested&whether&participants&actively&constructed&the&fillerZgap&
&
dependency&at&all&dependency&lengths&by&contrasting&NP&fillers&with&PP&fillers.&This&contrast&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
26
&
tests&the&availability&of&syntactic&category&information&across&the&three&dependency&lengths.&&
&
NPZfiller& sentences& were& expected& to& be& read& more& slowly& than& PPZfiller& sentences& in& the&
&
direct&object&region&if&the&dependency&was&constructed&actively,&i.e.,&if&the&dependency&was&
&
constructed& prior& to& direct& evidence& for& the& position& of& the& gap.& This& filledZgap& effect& was&
&
found& for& all& dependency& lengths.& Participants& thus& actively& constructed& the& fillerZgap&
&
dependency,&even&for&long&biZclausal&or&PPZextended&monoZclausal&dependencies.&The&fillerZ
&
gap& distances& tested& in& this& experiment& greatly& exceeded& those& used& in& most& studies&
&
completed& to& date& using& this& paradigm:& 9& words& and& 5& or& 6& major& constituents& linearly&
&
intervened&between&the&filler&and&the&critical&verb&for&both&+PP&and&+CP&Length&conditions,&
&
compared& with& an& average& of& 2.9& words& in& previous& studies.& We& can& thus& conclude& more&
&
confidently& that& fillerZgap& dependency& construction& itself& remains& active& across& clause&
&
boundaries.&It&appears&that&sensitivity&to&the&filler’s&grammatical&category&is&retained&when&
&
comprehenders&actively&posit&gaps,&since&participants&reliably&distinguished&the&NPZfiller&and&
&
PPZfiller&conditions.&&
&
&
The& filledZgap& effect& was& observed& as& a& punctate& effect,& whose& onset& varied& across&
&
conditions.&That&is,&the&entire&direct&object&region&was&not&rendered&more&difficult,&but&one&
&
particular& word& was.& Interestingly,& this& effect& appeared& earliest& for& the& +CP& Length&
&
dependency,& on& the& determiner& of& the& direct& object.& It& appeared& in& the& middle& word& of& the&
&
direct& object& for& short& dependencies& (an& adjective& or& a& noun),& and& on& the& last& word& of& the&
&
direct&object&for&+PP&Length&dependencies&(a&noun).&Due&to&the&limited&temporal&resolution&
&
of&the&selfZpaced&reading&technique,&we&cannot&draw&clear&conclusions&about&the&finer&timeZ
&
scale& dynamics& of& the& process.& However& in& every& condition& pair& the& effect& occurred& during&
&
the&processing&of&the&direct&object,&so&we&can&still&conclude&that&the&gap&was&posited&before&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
direct& evidence& of& the& missing& constituent’s& location.& One& possibility& for& explaining& the&
&
variation&in&timing&is&that&different&sets&of&processing&events&immediately&precede&the&verb&
&
in&the&different&conditions.&If&the&status&or&difficulty&of&parsing&decisions&made&prior&to&the&VP&
&
onset& affect& the& point& when& the& processor& begins& to& evaluate& a& hypothesized& fillerZgap&
&
dependency,&then&variation&in&the&observed&timing&of&the&filledZgap&effect&would&be&expected.&
&
However& it& is& worth& emphasizing& that& there& was& only& an& effect& of& Filler& Category,& and& no&
interaction&with&Length.&
&
&
&
These& results& provide& clear& evidence& of& active& dependency& formation& for& all&
&
dependency&lengths.&However&there&was&also&a&reading&time&slowdown&in&NPZfiller&sentences&
&
in&the&postZgap&region&of&+PP&Length&conditions.&This&observed&contrast&was&unanticipated.&It&
&
could&be&taken&as&evidence&that&linking&the&gap&with&the&NP&filler&is&more&difficult&when&the&
&
subject&of&the&clause&also&contains&a&PP.&A&possible&explanation&for&this&difficulty&stems&from&
&
the&fact&that,&once&the&direct&object&analysis&has&been&proven&incorrect,&a&gapZcontaining&PP&
&
may& be& predicted.& It& may& be& that& establishing& an& expectation& for& this& PP& could& be& more&
&
difficult&in&sentences&that&already&contain&a&similar&phrase&(i.e.,&in&the&+PP&Length&sentences).&
&
This& suggestion& is& plausible& if& there& is& retrieval& interference& for& attachment& (Lewis& &&
Vasishth,&2005),&but&future&studies&would&be&required&to&test&this&conjecture.&
&
&
&
The& present& results& suggest& that& category& information& is& well& preserved& across&
&
different& dependency& lengths.& The& finding& is& compatible& with& the& idea& that& basic& category&
&
information&could&be&either&maintained&or&retrieved&in&an&interferenceZrobust&manner,&but&it&
&
does& not& address& whether& other& contents& of& a& filler& phrase& are& maintained.& Therefore,& in&
&
Experiment& 2& we& make& the& measure& of& dependency& formation& contingent& upon& semantic&
features&of&the&filler,&in&order&to&test&whether&those&are&preserved&across&different&lengths.&&
&
&
&
PA&
Running&head:&MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&
CONSTRUCTION
3
&
3.1
&
Experiment&2:&Semantic&anomaly&detection&
Rationale*
&
In&the&second&experiment&we&used&a&different&measure&of&active&dependency&formation:&the&
semantic& anomaly& effect& (Garnsey,& Tanenhaus& && Chapman,& 1989,& Traxler& && Pickering,&
1996).&Example&(9)&illustrates&a&semantic&anomaly&contrast.&If&the&filler&is&initially&treated&as&
the& direct& object& of& the& verb& ‘erect’,& then& the& initial& interpretation& describes& a& plausible&
situation&in&(9a),&but&not&in&(9b).&&
&
(9)&
(a)&
The&monument!which&the&pilgrim&erected&___&to&appease&the&gods&…&
&
(b)&
The&gods&which&the&pilgrim&erected&the&monument&to&appease&___&…&
&
For&this&reason&a&reading&time&slowdown&is&expected&at&the&verb&or&shortly&thereafter.&&
&
We&created&contrasts&like&(9)&for&three&different&dependency&lengths:&a&short&singleZ
clause&dependency,&a&longer&singleZclause&dependency,&and&a&longer&biZclausal&dependency.&
If&the&semantic&details&of&the&filler&are&maintained&in&the&focus&of&attention&to&participate&in&
active& dependency& formation& across& the& different& dependency& lengths,& then& we& expect& to&
observe&a&semantic&anomaly&effect&either&at&the&critical&verb&or&shortly&thereafter.&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
3.2
&
&
Materials*and*Methods**
PA&
&
Participants& were& 36& native& speakers& of& English& from& the& University& of& Maryland&
&
community&who&received&$10&for&taking&part&in&the&experiment.&
&
&
&
This&experiment&crossed&the&factors&Filler&Plausibility&and&Length&in&a&2&⋅&3&factorial&
&
design.& Filler& Category& was& either& ‘plausible’& or& ‘implausible’.& Length& was& either& ‘Short’,&
&
‘+PP’,&‘+CP’,&identical&to&Experiment&1.&Table&4&illustrates&a&full&materials&set.&
&
<<INSERT&TABLE&4>>&
&
In&this&design&the&verb&is&the&critical&region&because&it&is&where&the&semantic&fit&of&the&
&
filler& can& first& be& detected.& However& because& such& effects& are& prone& to& spillZover& in& selfZ
&
paced& reading& tasks,& the& direct& object& region& was& made& 4& words& in& length.& Consequently,&
&
bottomZup&evidence&for&a&missing&constituent&would&also&be&unavailable&for&some&time&after&
&
the&verb.&A&reading&time&slowdown&in&the&implausible&conditions&either&at&the&verb&or&in&the&
&
4Zword& direct& object& region& could& be& considered& an& effect& of& active& dependency&
&
construction.&&
&
The&verbs&and&fillers&chosen&in&this&experiment&were&the&same&as&Wagers&&&Phillips&
&
(2009).&The&verbs&were&sprayZloadZtype&alternating&locative&verbs&(Fraser,&1971,&Rappaport&
&
&&Levin,&1986),&which&take&two&internal&arguments:&the&Figure&and&the&Ground&arguments,&
&
arguments&which&correspond&to&a&moving&object&and&a&location,&respectively.&For&example,&
&
in& the& sentence& “The& squirrels& crammed& their& cheeks& with& acorns,”& the& argument& their*
&
cheeks* is& the& Ground& and& acorns& is& the& Figure.& All& verb& phrases& in& the& experiment& allow&
&
alternation&between&GroundZFigure&and&FigureZGround&argument&orderings&(e.g.,&“crammed&
&
acorns& into& their& cheeks”),& but& the& GroundZFigure& ordering& of& arguments& was& used& in& all&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
30
&
experimental& sentences.& Crucially& the& verbZfiller& combinations& were& normed& so& that& the&
&
implausible& fillers& were& equally& implausible& as& either& Figure& or& Ground& (see& Wagers& &&
&
Phillips,& 2009,& for& norming& details).& Because& of& this,& comprehenders& could& not& use& the&
&
plausibility&of&the&filler&as&a&particular&argument&of&verb&to&guess&its&grammatical&role.&Thus&
&
by&factoring&out&cues&to&the&filler’s&grammatical&role&from&the&verb,&we&could&be&confident&
&
about& attributing& plausibility& effects& in& the& reading& times& before& the& gap& to& active&
&
dependency&completion&processes.&
&
Moreover,& this& norming& responds& to& the& observations& of& Boland& et& al.& (1995)& and&
&
Pickering& && Traxler& (2001)& that& available& alternative& argument& positions& could& mute& a&
&
semantic&anomaly.5&To&satisfy&the&double&constraint&of&anomaly&as&either&figure&or&ground&
&
argument,&it&proved&to&be&practically&necessary&to&make&the&anomaly&one&which&depended&
&
on&real&world&knowledge,&rather&than&selectional&restrictions.&Partially&as&a&consequence&of&
&
this& constraint,& we& were& able& to& avoid& a& confound& with& the& animacy& of& the& filler.& The&
&
processing&complexity&of&object&relative&clauses&is&sensitive&to&the&animacy&of&both&the&filler&
&
and&the&relative&clause&subject.&In&particular,&when&the&filler&and&RC&subject&positively&match&
&
in&animacy,&the&RC&is&more&difficult&to&comprehend&(Traxler&et&al.,&2002).&In&18&of&our&item&
&
sets,& both& the& plausible& and& implausible& fillers& were& inanimate;& in& 5& sets,& the& implausible&
&
filler&was&animate&and&the&plausible&filler&was&animate;&in&1&set,&the&implausible&filler&was&
&
inanimate,& and& the& plausible& filler& was& animate.& In& order& to& test& for& possible& effects& of&
&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
&
Although&the&design&of&the&current&study&required&that&the&verbs&be&distinct&from&the&verbs&
in& Experiment& 1,& there& are& several& similarities.& The& verbs& in& both& Experiment& 1& and& in&
&
Experiment& 2& allow& two& internal& arguments& (Theme/Benefactive& in& Experiment& 1;&
Figure/Ground&in&Experiment&2).&Verbs&in&both&experiments&allow&alternative&orderings&of&
&
the&arguments.&Finally&verbs&in&both&experiments&only&require&one&of&the&arguments&to&be&
realized& syntactically.& Thus& the& relative& complexity& of& the& two& classes& is& plausibly&
&
equivalent.&
5&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
animacy,& we& performed& analyses& identical& to& the& reported& analyses& except& that& they&
&
excluded& the& sets& involving& extraction& of& an& animate& argument;& these& did& not& reveal&
&
qualitatively& different& results& and& patterns& of& significance& were& preserved.& Therefore& we&
&
only&report&analyses&with&all&items&included.&An&analysis&without&animate&extractions&can&be&
replicated&from&the&materials&provided&on&the&authors’&web&site.&
&
&
Analysis& preceded& identically& to& the& analysis& in& Experiment& 1.& Filler& conditions& were&
&
sum& coded& for& Filler& Category& (Implausible& =& +1).& Helmert& contrasts& for& Length& were&
&
identical& to& the& Length& contrasts& in& Experiment& 1.& One& participant& was& excluded& from&
&
analysis&due&to&exceptionally&low&accuracy&(63%;&greater&than&2&standard&deviations&from&
the&mean,&expressed&in&logits).&
&
&
&
&
3.3
&
Results!
3.3.1* Comprehension*accuracy*
&
&
&
&
Mean& comprehension& accuracy& was& 91%.& Comprehension& accuracy& by& condition& is&
&
reported&in&Table&5.&The&only&significant&difference&was&between&PPZlength&conditions&and&
&
CPZlength&conditions,&the&former&being&answered&more&accurately&(0.79&±&0.33,&t&=&2.4,&p&<&
.05).&
&
&
&
<<INSERT&TABLE&5>>&
&
&
&
3.3.2* Reading*times*
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
32
&
Reading& time& results& are& reported& in& Figure& 2.& Coefficient& tables& from& the& linear&
&
mixedZeffects&models&of&the&four&regions&of&analysis&are&given&in&Table&6.&Words&12Z16,&for&
&
which& a& slowZdown& due& to& implausibility& would& count& as& evidence& for& active& dependency&
&
completion,&are&annotated&as&“active&filling”&regions&and&analyzed&separately&as&the&verb&and&
&
the&direct&object&regions.&Words&18Z20,&for&which&a&slowZdown&due&to&implausibility&would&
&
show&that&participants&recognized&the&direct&evidence&for&the&gap&location,&are&annotated&as&
&
“gapZdriven”& regions.& Word& 11& was& analyzed& as& the& preZcritical& adverb& region,& and& there&
&
were&no&effects&in&that&region.&
&
&
Active&filling&effects&were&most&robustly&found&in&the&Short&Length&conditions.&In&the&
&
active& filling& regions& (verb& and& direct& object),& there& was& a& slowZdown& for& the& implausible&
&
filler&conditions,&indicating&sensitivity&to&the&filler’s&fit&as&an&argument&of&the&critical&verb.&In&
&
the& verb& region,& this& slowZdown& was& only& present& in& Short& conditions.& In& the& direct& object&
&
region,&it&was&present&as&a&simple&effect,&but&was&smaller&for&the&+PP/+CP&length&conditions&
&
overall.& A& pairwise& test& of& the& Filler& Plausibility& contrast,& restricted& to& just& +CP& Length&
&
conditions,&showed&that&it&was&absent&for&that&length&(t&=&Z.41,&p&=&.70).&&
&
However,& in& the& gapZdriven& regions& (words& 18Z20),& both& the& +PP& and& +CP& Length&
&
conditions& showed& clear& evidence& of& sensitivity& to& the& plausibility& manipulation.& Here& the&
&
opposite&pattern&was&observed&compared&to&the&active&region:&the&effect&of&plausibility&was&
&
greatest&in&this&region&for&+CP&length&conditions.&&
&
Overall& this& pattern& of& results& contrasts& with& the& filledZgap& effect& in& Experiment& 1,&
&
which& was& detected& for& all& length& conditions.& It& replicates& the& difference& observed& by&
&
Wagers&&&Phillips&(2009)&for&Short&v.&+PPZlength&dependencies,&and&it&extends&the&finding&to&
&
+CP&length&dependencies.&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
<<INSERT&FIGURE&2>>&
<<INSERT&TABLE&6>>&
&
PA&
&
&
3.4
Discussion*
&
The&goal&of&this&study&was&to&test&whether&the&semantic&anomaly&index&of&active&dependency&
&
formation& is& observed& at& long& dependencies.& At& all& dependency& lengths& we& found& that&
&
comprehenders&were&ultimately&sensitive&to&the&semantic&fit&of&the&filler&with&the&verb&that&
&
hosted&its&gap.&However&the&onset&of&that&sensitivity,&as&reflected&in&slower&reading&times&for&
&
implausible& verbZfiller& combinations,& varied& with& dependency& length.& For& long& biZclausal&
&
dependencies&comprehenders&only&showed&sensitivity&to&plausibility&once&there&was&direct&
&
evidence&in&the&input&for&the&location&of&the&missing&constituent.&This&contrasted&strikingly&
&
with& short& monoZclausal& dependencies,& where& sensitivity& arose& immediately& at& the& verb.&&
&
MonoZclausal& dependencies& that& were& lengthened& by& modifying& the& subject& with& a& PP&
&
showed&weaker&sensitivity&to&the&semantic&fit&of&the&filler&prior&to&the&gap&region.&However,&
&
the& RT& difference& in& the& active& region& was& reliably& smaller& for& +PP& Length& dependencies&
&
than&for&the&short&monoZclausal&dependencies.&This&contrast&between&anomaly&detection&in&
&
short& and& +PP& Length& dependencies& replicates& the& previous& finding& by& Wagers& && Phillips&
&
(2009;&Experiment&3).&
&
This& pattern& of& results& can& be& interpreted& in& one& of& two& ways:& either& (a)& active&
&
dependency&formation&ceased&by&the&time&the&most&deeply&embedded&verb&was&reached&in&a&
&
biZclausal& dependency,& or& (b)& the& lexicoZsemantic& features& of& the& filler& required& to& elicit&&
&
implausibility&effects&did&not&participate&in&the&active&part&of&dependency&construction&for&
&
longer&dependencies.&The&results&of&Experiment&1&make&interpretation&(a)&unlikely.&In&that&
&
experiment& the& filledZgap& effect& was& present& in& monoZclausal& and& biZclausal& dependencies&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
34
&
alike.&Therefore,&some&version&of&interpretation&(b)&is&likely&to&be&correct.&It&is&important&to&
&
observe,&however,&that&it&is&not&the&case&that&longer&dependencies&are&simply&more&taxing&
&
for& comprehenders,& and& that& the& effect& of& semantic& fit& declines& as& dependency& length&
&
increases.&In&both&of&the&long&dependency&conditions&the&effect&of&semantic&fit&was&reliable&
&
in& the& postZgap& region,& and& not& different& from& short& dependency& conditions.& Indeed,& the&
&
most& robust& postZgap& effect& was& found& for& +CP& Length& dependencies,& which& showed& no&
&
evidence& of& a& plausibility& effect& during& the& active& filling& region.& Therefore& the& lexicoZ
&
semantic& features& of& the& filler& are& not& forgotten.& Rather& they& specifically& become&
&
inaccessible&or&ineffective&for&active&dependency&formation.&&
&
In& the& context& of& the& results& of& Experiment& 1,& we& propose& that& longZdistance&
&
dependency& formation& reflects& the& integration& of& information& in& two& states.& The& heart& of&
&
active& dependency& formation& is& the& act& of& predictive& structure& building:& dependencies& are&
&
projected&forward&in&time&by&the&comprehender,&in&advance&of&direct&evidence&for&the&gap&
&
location.& However,& we& propose& that& the& extent& to& which& the& dependency& is& initially&
&
elaborated&or&evaluated&is&limited&by&the&information&about&the&filler&that&is&carried&forward&
&
in& time.& In& other& words,& the& comprehender& maintains& what& she& can& about& the& filler& in& an&
&
active&representational&state,&and&this&is&what&guides&initial&dependency&construction.&This&
&
‘active’& state& can& be& identified& with& the& focus& of& attention& or& the& capacity& of& concurrent&
&
processing&(Broadbent,&1958,&McElree,&2006,&Jonides,&Lewis,&Nee,&Lustig,&Berman&&&Moore,&
&
2008).&
&
&Since& this& capacity& is& limited,& a& complete& representation& of& the& filler& cannot& be&
&
maintained.& When& other& processing& events& intervene,& some& details& of& the& filler& may& be&
&
displaced.& Therefore& the& initial& dependency& might& be& thought& of& as& a& provisional&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
representation,&the&details&of&which&are&filled&out&by&retrieving&the&complete&representation&
&
of&the&filler&from&memory,&the&process&normally&identified&as&reactivation&(Nicol&&&Swinney,&
1989).&&
&&
&
&
The&results&of&Experiments&1&and&2&together&suggest&that&the&category&identity&of&the&
&
filler&is&maintained&across&all&dependency&lengths,&whereas&its&semantic&details&are&not.&This&
&
has& both& architectural& and& functional& advantages.& From& an& architectural& standpoint,& the&
&
information& that& encodes& category& identity& is& more& compact& than& the& information& about&
&
semantic& features.& This& may& be& because& the& category& identity& is& a& coarser,& more& general&
&
representation&for&which&there&are&only&a&few&possibilities.&In&contrast,&individual&semantic&
&
features& (e.g.,& +ANIMATE,& +CONCRETE,& +MASS,& +LIQUID,& & etc.)& are& finerZgrained,& and& there& are&
&
many& more& possibilities.& From& a& functional& standpoint,& category& information& is&
&
grammatically&more&useful&in&evaluating&the&wellZformedness&of&a&dependency.&A&sentence&
may&express&an&unusual&or&implausible&state&of&affairs,&and&yet&be&grammatical.&&
&
&
We&develop&this&account&in&greater&detail&in&the&general&discussion,&but&immediately&it&
&
raises& the& question& of& what& the& appropriate& level& of& ‘coarseness’& is& for& maintenance.& In&
&
Experiment&3,&we&examine&another&measure&of&active&dependency&formation,&in&which&the&
&
information&that&must&be&maintained&about&the&filler&is&not&as&general&as&category&identity,&
but&is&still&more&constrained&than&the&semantic&features.&
&
&
&
&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
4
4.1
&
Experiment&3:&Subcategorization&Match&
Rationale*
36
&
&
&
In&the&third&experiment&we&devised&another&measure&of&active&dependency&formation&
&
by&manipulating&subcategorization&match.&It&is&based&upon&the&fact&that&arguments&bearing&
&
particular& thematic& roles& must& occur& as& the& complement& of& specific& prepositions.& For&
&
example,&the&verb&‘inherit’&requires&a&source&argument.&In&English,&source&arguments&are&the&
&
complement&of&the&preposition&‘from’&(10a).&In&contrast,&verbs&like&‘entrust’&require&a&goal&
&
argument.&Goal&arguments&typically&appear&as&the&complement&of&the&preposition&‘to’&(10b).&
&
(10)& (a)&
The&orphan&inherited&the&prayer&book&from/*to&his&uncle.&
&
The&monk&entrusted&the&prayer&book&*from/to&his&novice.&&
(b)&
&
&
If&we&displace&the&Source/Goal&argument,&the&restriction&on&preposition&identity&still&holds.&
&
(11)& (a)&
The&uncle&from/*to&whom&the&orphan&inherited&the&prayer&book&
&
The&novice&*from/to&whom&the&monk&entrusted&the&prayer&book.&
(b)&
&
&
If&comprehenders&are&sensitive&to&not&just&the&category&identity&of&a&displaced&PP,&but&also&to&
&
the&particular&preposition&that&heads&the&PP,&then&there&should&be&a&reading&time&slowdown&
&
when& the& displaced& preposition& is& not& appropriate& for& the& kind& of& semantic& argument& the&
&
verb&requires.&
&
&
The& subcategorization& mismatch& manipulation& was& tested& at& two& different&
&
dependency& lengths:& short& monoclausal& dependencies& and& PPZextended& monoclausal&
&
dependencies.&We&chose&not&to&include&a&CPZextended&dependency&in&this&experiment&since&
&
most& of& the& embedding& verbs& used& in& Experiments& 1Z2& allow& PP& arguments& headed& by& to&
&
and&from&(‘conclude&from&experience,’&‘say&to&a&friend’,&etc).&This&would&allow&a&fronted&PPZ
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
filler&to&be&prematurely&interpreted&in&the&main&clause,&potentially&obscuring&the&effects&of&
&
interest&at&the&critical&verb.&
&
&&
&
4.2
&
Materials*and*methods*
&
&
&
Participants& were& 54& native& speakers& of& English& who& were& recruited& via& Amazon&
&
Mechanical& Turk& (we& attempted& to& recruit& as& many& participants& as& possible& over& a& 2Zday&
&
period).& They& received& $4& for& taking& part& in& the& experiment.& Participants& completed& the&
study&
using&
&
the&
online&
experiment&
platform&
Ibex&
(A.&
&
Drummond,&
&
http://spellout.net/ibexfarm),&which&allows&a&selfZpaced&reading&experiment&to&be&deployed&
&
in& the& browser.& An& earlier& study& of& the& same& design,& conducted& in& the& lab& with& 18& native&
&
speakers,&yielded&the&same&results;&this&allays&some&potential&concerns&about&the&robustness&
of&realZtime&psycholinguistic&data&derived&from&a&webZbased&participant&sample.&
&
&
&
This& experiment& crossed& the& factors& Filler& Match& and& Length& in& a& 2×2& factorial&
&
design.&The&Filler&Match&factor&contrasted&‘Match’&and&‘Mismatch’&conditions,&corresponding&
&
to&the&subcategorization&match&between&the&verb&and&the&PP.&The&Length&factor&contrasted&
&
‘Short’&and&‘Long’&conditions.&As&in&Experiment&1,&the&Short&Length&conditions&separated&the&
filler& and& the& verb& by& a& twoZword& subject& and& an& adverb.& The& Long& conditions& attached&&a&
&
fiveZword& PP& modifier& to& the& subject.& Table& 7& illustrates& a& full& materials& set.& Items& were&
&
balanced&so&that&the&matching&preposition&was&to&in&12&item&sets&and&from&in&12&item&sets.&
&
Materials&were&distributed&according&to&a&Latin&Square&across&four&lists&so&that&participants&
each&read&six&sentences&per&condition.&&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
38
&
The& analysis& details& were& as& described& in& Experiment& 1.& Regions& of& interest& were&
&
structurally& aligned& so& that& the& VP& words& common& to& all& conditions& in& an& item& set&
&
corresponded&to&the&same&regions&across&conditions.&Because&there&were&only&two&levels&of&
&
the& Length& factor,& it& was& sum& coded& in& this& study,& with& Long& (+PP)& represented& by& the&
&
positive&coefficient.&&
&
4.3*
&
&
&
Results*
&
&
4.3.1* Accuracy!
&
&
Comprehension& accuracy& is& reported& in& Table& 8.& Overall& accuracy& was& 84%.& There&
&
was&a&3%&decrement&in&accuracy&associated&with&long&dependencies,&and&a&2%&decrement&
&
for& verbs& with& mismatching& PP& arguments.& However& none& of& these& effects,& nor& their&
&
interaction,&achieved&significance.&&
&
&
&
<<&INSERT&TABLE&8&>>&
&
&
4.3.2& Reading&times&
&
&
Reading& time& results& are& reported& in& Figure& 3.& The& main& finding& from& the& reading&
&
time&data&is&that&sensitivity&to&the&verbZPP&match&was&observed&principally&only&in&the&short&
&
dependency&conditions.&As&in&Experiment&2,&we&report&four&regions&of&analysis:&the&adverb,&
&
two& ‘active& filling’& regions:& the& verb& and& the& direct& object& NP,& and& the& 3& following& words&
&
(‘gapZdriven’).&Model&coefficients&are&given&in&Table&9.&In&the&adverb&and&verb&regions&there&
&
were&no&effects.&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
In& the& object& NP& region,& there& was& an& anomaly& effect:& Filler& Mismatch& conditions&
&
were&read&more&slowly&than&Filler&Match&conditions.&However&there&was&also&an&interaction&
&
with&Length,&such&that&the&contrast&was&almost&entirely&abolished&for&Long&conditions.&This&
&
is&apparent&in&the&untrimmed&data&as&well,&as&Figure&3&makes&clear.&In&the&gap&region,&the&
&
anomaly&effect&of&Filler&Mismatch&approached&marginal&significance,&as&did&the&interaction&
with&Length;&but&neither&effect&was&reliable&at&α&=&.05.&
&
&
<<&INSERT&FIGURE&3&>>&
<<&INSERT&TABLE&9&>>&
&
&
&
&
&
&
&
&
Discussion*
This&experiment&provided&evidence&that&comprehenders&were&sensitive&online&to&the&
&
subcategorization& restriction& that& verbs& place& on& a& displaced& PPZargument,& but& only& when&
&
the&distance&between&the&displaced&PP&and&the&verb&was&short.&When&more&than&a&twoZword&
&
subject& intervened& between& the& PP& filler& and& the& verb,& there& was& no& indication& in& the&
&
reading& times& that& participants& detected& a& mismatch.& This& finding& suggests& that& the&
&
information&about&the&identity&of&the&preposition&is&not&well&preserved&over&a&relatively&long&
&
dependency.& Only& in& the& short& conditions& was& there& evidence& that& a& mismatching& PP& was&
&
rapidly& detected,& based& on& reading& times& at& the& verb& and& in& the& following& region.& The&
&
presence& of& a& slowdown& in& this& region& indicates& that& comprehenders& evaluated& the&
&
displaced&PP&as&a&potential&argument&of&the&verb.&&
&
&
An& important& difference& between& this& experiment& and& Experiment& 2& should& be&
&
highlighted.&In&Experiment&2,&the&semantic&anomaly&led&to&a&slowdown&in&the&active&filling&
&
regions& in& the& short& dependency& conditions,& but& not& in& the& longer& dependency& conditions.&
&
4.3
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
40
&
Nevertheless,&the&long&dependency&conditions&did&eventually&show&an&effect&of&the&semantic&
&
anomaly,& in& the& postZgap& regions.& In& contrast,& in& the& present& experiment& there& was& no&
&
statisticallyZsignificant&slowdown&in&the&long&dependency&conditions.&It&could&be&that&there&
&
was&a&high&level&of&variability&in&the&timing&and&strength&of&the&mismatch&effect&at&the&level&
&
of&individual&trials&or&participants,&with&the&consequence&that&we&were&not&able&to&observe&a&
&
punctate& mismatch& effect& in& the& average& reading& times.& Furthermore,& the& absence& of& a&
&
reliable& effect& even& in& the& gap& region& raises& the& concern& that& our& study& lacked& the&
&
appropriate&power&to&detect&it.&However,&in&a&pilot&study&(n&=&18)&we&did&replicate&the&exact&
&
pattern,& with& a& different& sample:& college& students& in& the& lab.& We& supply& those& data& in& our&
&
supplementary&materials.&6&&
&
&
Alternatively,& it& could& be& that& the& mismatch& was& never& detected,& unless& it& was&
&
detected&actively.&In&a&followZup&grammaticality&study&with&RSVP&presentation&and&speeded&
&
judgment,&we&presented&32&participants&with&the&target&sentences&from&Experiment&3&(the&
&
composition& of& the& fillers& was& different).& In& the& Short& dependency& conditions& participants&
&
accepted& grammatical& sentences& in& 83%& of& trials,& but& they& also& only& rejected& the&
&
ungrammatical&sentences&in&29%&of&trials.&In&the&Long&dependency&conditions&participants&
&
accepted&grammatical&sentences&in&67%&of&trials,&and&rejected&ungrammatical&sentences&in&
&
43%& of& trials.& In& a& mixedZeffects& logistic& regression& over& percent& correct,& there& were& only&
&
main& effects& of& Match& (Z.71& ±& 0.49,& p& <& .001)& and& Length& (Z.95& ±& 0.49,& p& <& .001),& and& no&
&
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
&
6&If&we&took,&as&an&expectation&of&the&effect&size,&the&average&difference&between&Filler&Match&
&
conditions&in&Short&dependencies,&excluding&the&first&word&in&the&direct&object&region&(see&
Figure& 3),& we& would& have& a& small& positive& difference& corresponding& to& a& withinZsubjects&
&
corrected& Cohen’s& d& of& 0.33.& For& a& oneZsided& tZtest& and& a& repeated& measures& design,& a&
sample&size&of&58&would&be&required&to&achieve&a&power&of&.80&(1Zβ)&(G*Power&3,&Faul&et&al.&,&
&
2007).&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
interaction.&What&is&most&notable&is&that&overall&sensitivity&to&this&violation&was&low&for&both&
&
lengths.&We&computed&dZprime,&a&measure&of&sensitivity&that&corrects&for&response&bias&by&
&
taking&the&difference&between&the&hit&rate&and&the&false&alarm&rate&(expressed&as&a&zZscore;&
&
MacMillan& && Creelman,& 2004).& In& our& case,& the& hit& rate& corresponded& to& the& endorsement&
&
rate& for& grammatical& sentences;& and& the& false& alarm& rate& the& endorsement& rate& for&
&
ungrammatical& sentences.& In& short& dependencies,& the& resulting& dZprime& was& 0.40& and& for&
&
long& dependencies& it& was& 0.26.& Participants& therefore& seemed& to& mostly& endorse& short&
&
sentences& as& grammatical,& regardless& of& match,& and& to& reduce& endorsement& for& long&
&
sentences& overall.& Sensitivity& to& the& mismatch& remained& very& small& in& both& cases.& This&
&
pattern&corroborates&the&idea&that&individuals&do&not&retain&sensitivity&to&the&lexical&identity&
&
of& the& preposition& in& these& sentences& for& a& long& time,& even& in& short& sentences.& A& possible&
&
reason&is&that&the&verb&is&so&semantically&constraining&about&the&possible&role&of&the&PP,&that&
&
only&information&about&the&NP&expressed&inside&the&PP&is&useful&to&retain.&We&discuss&this&
possibility&in&greater&detail&in&the&General&Discussion.&
&
&
5
5.1
&
General&discussion&
Summary*of*results*
&
&
In& three& experiments& we& tested& the& sensitivity& of& different& measures& of& active&
&
dependency& formation& to& extended& dependency& lengths.& Characterizing& the& difference&
&
between&early,&active&dependency&completion&and&later&nonZactive/gapZdriven&reactivation&
&
of& the& filler& is& potentially& a& window& onto& how& the& comprehender& manages& memory&
&
resources& in& the& course& of& comprehending& an& unbounded& dependency.& In& order& to&
&
distinguish& active& and& nonZactive& phases& of& dependency& completion,& longer& dependencies&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
42
&
become& a& crucial& testing& ground.& However,& there& have& been& surprisingly& few& studies&
&
examining&longer&dependencies.&In&Experiments&1&and&2&we&tested&the&filledZgap&effect&and&
&
the& plausibility& effect& respectively,& both& of& which& are& commonly& used& in& studies& on&
&
processing&fillerZgap&dependencies&(e.g.,&Garnsey,&Tanenhaus&&&Chapman,&1989,&Lee,&2004,&
&
Stowe,& 1986,& Traxler& && Pickering,& 1996).& In& Experiment& 3& we& tested& the& effect& of&
&
subcategorization&mismatch.&These&measures&differ&in&terms&of&the&kind&of&the&information&
&
needed& about& the& filler& phrase& to& distinguish& the& normal& and& (sometimes& temporarily)&
&
anomalous&sentences&in&each&study:&the&filledZgap&effect&is&sensitive&to&the&maximal&category&
&
of&the&filler&(e.g.,&‘which&chemicals’&v.&‘for&which&chemicals’);&the&subcategorization&effect&is&
&
sensitive&to&the&identity&of&a&PP&filler’s&closedZclass&head&(e.g.,&‘to&which&courier’&v.&‘for&which&
&
courier’);& and& the& plausibility& effect& is& sensitive& to& particular& semantic& features& of& an& NP&
&
filler’s&openZclass&head&(e.g.,&‘acorns’&v.&‘cats’).&
&
&
The&filledZgap&effect&tested&in&Experiment&1&was&apparent&at&all&dependency&lengths.&
&
The&logic&of&the&filledZgap&manipulation&was&as&follows:&if&the&parser&posited&a&direct&object&
&
gap& for& NP& fillers,& but& not& for& PP& fillers,& then& there& should& be& longer& reading& times& in& the&
&
direct& object& region& for& NP& conditions& compared& to& PP& conditions,& likely& reflecting&
&
reanalysis&that&was&required&in&the&NPZfiller&sentences.&The&fact&that&a&filledZgap&effect&was&
&
found& at& all& dependency& lengths& suggests& that& the& parser& actively& completed& the& fillerZgap&
&
dependency&in&all&of&those&contexts,&and&that&it&specifically&projected&direct&object&gaps&for&
&
NP&fillers.&&
&
&
In& contrast& to& the& filledZgap& effect,& the& plausibility& effect& in& Experiment& 2& showed&
&
more&variability&across&conditions.&The&logic&of&the&plausibility&effect&was&as&follows:&if&the&
&
parser&immediately&analyzed&the&filler&as&an&argument&of&the&critical&verb,&then&fillers&that&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
were&semantically&anomalous&arguments&should&lead&to&increased&reading&times&at&the&verb&
&
or& in& the& direct& object& region& (the& ‘active& filling’& regions).& In& the& active& filling& regions& the&
&
plausibility&effect&was&strong&in&short&dependency&conditions,&but&completely&absent&in&long&
&
biclausal& dependency& conditions.& In& the& long,& PPZextended& dependency& conditions& the&
&
plausibility&effect&was&present,&but&weak.&Of&particular&importance&is&the&fact&that&the&long&
&
conditions& that& showed& weak& or& no& plausibility& effects& in& the& active& filling& regions&
&
nevertheless& showed& a& clear& plausibility& effect& in& the& postZgap& regions.& Comprehenders&
&
therefore& did& not& fail& to& notice& the& semantically& anomalous& arguments,& but& they& did& not&
&
detect& the& semantic& anomaly& before& reaching& the& gap.& & We& observed& the& same& pattern& of&
&
results& in& an& earlier& experiment& (Wagers& && Phillips,& 2009,& Experiment& 3),& which& only&
included&the&+PP&Lengthened&condition.&&
&
&
&
Finally,& the& subcategorization& mismatch& effect& in& Experiment& 3& showed& a& profile&
&
similar&to&the&plausibility&effect&in&Experiment&2.&The&logic&of&this&index&was&as&follows:&if&the&
&
parser& immediately& analyzed& the& PP& filler& as& an& argument& of& the& critical& verb,& then& fillers&
&
whose& prepositional& head& did& not& match& the& requirements& of& the& verb& should& lead& to&
&
increased& reading& times& at& the& verb& or& in& the& direct& object& region& that& preceded& the& gap.&
&
However&the&mismatch&effect&was&only&apparent&in&the&short&dependency&conditions.&As&in&
&
Experiments& 1& and& 2& the& active& dependency& formation& effect& in& short& dependencies& was&
&
large&and&occurred&early&in&the&regions&of&interest.&Paralleling&the&findings&of&Experiments&1&
&
and&2,&the&mismatch&effect&was&attenuated&in&a&long,&PPZextended&dependency.&In&fact,&in&this&
&
study& the& effect& in& the& PPZextended& conditions& was& completely& absent& in& active& regions& or&
&
after&the&gap.&It&is&therefore&unclear&how&robustly&comprehenders&recovered&sensitivity&to&
&
the& identity& of& the& filler’s& prepositional& head& in& long& dependencies.& A& followZup& speeded&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
44
&
grammaticality& test& indicated& that& endZofZsentence& judgments& were& not& very& accurate;&
&
however,&there&was&no&interaction&with&length.&
&
5.2
&
&
Memory*and*control*processes*in*active*dependency*construction*
&
&
&
We& conclude& that& fillerZgap& dependencies& are& actively& constructed& regardless& of&
&
dependency&length.&This&can&explain&the&pattern&of&results&observed&in&Experiment&1,&and&by&
&
Frazier& && Clifton& (1989):& even& long,& biclausal& fillerZgap& dependencies& are& formed& actively.&
&
What& must& be& accounted& for,& then,& is& why& longer& dependencies& failed& to& show& any& active&
&
effects& involving& properties& of& the& filler& phrase& that& were& more& specific& than& syntactic&
&
category.& The& likely& explanation,& we& argue,& is& that& certain& information& is& immediately&
&
accessible& to& the& processor& to& make& parsing& decisions,& while& other& information& must& be&
&
retrieved&from&passive&memory&once&the&dependency&is&formed.&Findings&from&a&variety&of&
&
cognitive&tasks&indicate&that&memory&is&partitioned&into&a&highly&capacityZlimited&focal&state&
&
(Cowan,& 2001,& Garavan,& 1998,& Jonides& et& al.,& 2008,& McElree,& et& al.,& 1989,& 2000,& 2006,&
&
Verhaegen,& et& al.,& 2004)& and& a& virtually& unlimited& nonZfocal& state.& Information& in& the& focal&
&
state&can&be&incorporated&into&onZgoing&processing&directly.&However,&information&that&has&
&
been& displaced& from& this& state& must& be& reincorporated& by& retrieval& operations,& a& finding&
&
supported& by& speedZaccuracy& tradeZoff& timeZcourse& studies& of& list& memory& (McElree& &&
&
Dosher,& 1989;& McElree,& 1998;& McElree,& 2006)& and,& more& recently,& studies& of& subjectZverb&
&
dependencies&in&language&comprehension&(McElree&et&al.,&2003).&&
&
&
Interpreted& in& view& of& this& architecture,& our& data& suggest& that& the& comprehension&
&
system& completes& fillerZgap& dependencies& based& on& two& complementary& processes:& &a&
“leading”&process,&which&projects&structure&ahead&of&the&input&based&on&a&limited&amount&of&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
information&in&the&focus&of&attention;&and&a&“lagging”&process&which&involves&the&retrieval&or&
&
reactivation& of& the& full& filler& phrase.& The& filledZgap& effect’s& relative& insensitivity& to& length&
&
derives&from&the&fact&that&active&dependency&formation&can&be&driven&by&a&minimal&amount&
&
of& information,& i.e.,& the& information& that& a& filler& of& category& XP& exists,& and& that& this& small&
&
amount& of& information& can& survive& during& the& course& of& the& sentence,& even& in& the& face& of&
&
unrelated& processing& events.& The& filledZgap& effect& depends& upon& the& minimal& amount& of&
&
information& necessary& to& drive& active& filling,& whereas& the& other& measures& of& active& gap&
&
creation&depend&on&maintenance&of&fuller&syntactic&and&semantic&details.&In&Experiments&2&
&
and&3,&when&the&fillerZgap&dependency&was&initially&formed&in&long&dependencies,&lexicallyZ
&
specific& information& about& the& filler& phrase& was& not& available& that& would& lead& the&
&
comprehender& to& notice& any& anomalies.& Once& the& comprehender& retrieved& the& filler,&
however,&the&acceptability&of&the&dependency&could&be&evaluated.&&&
&
&
&
The&core&idea&that&our&data&support&is&that&certain&features&of&the&representation&are&
&
privileged& in& anticipation& of& how& future& information& should& be& integrated.& This& can& be&
&
achieved&in&a&variety&of&models.&For&example,&the&proposed&interplay&of&predictive,&leading&
&
processes& and& retrospective,& lagging& processes& can& be& combined& in& a& straightforward&
&
hybridization& of& the& two& leading& mechanisms& of& fillerZgap& processing:& a& maintenance&
&
mechanism,& as& in& Wanner& && Maratsos’& (1978)& HOLD& CELL& hypothesis& (and& Frazier& (1987)’s&
&
related&idea&that&the&filler&is&somehow&not&‘inert’);&and&a&reactivation&mechanism,&proposed&
&
on&the&basis&of&crossZmodal&lexical&priming&and&probe&recognition&tasks&(Bever&&&McElree,&
&
1989;& Nicol& && Swinney,& 1989;& Nicol,& Fodor& && Swinney,& 1994).& This& decomposition& of&
&
mechanisms& actually& seems& to& be& reflected& rather& directly& by& the& dissociation& of& the&
&
plausibility&effect&in&Experiment&2:&the&effect&appeared&on&the&verb&for&short&dependencies,&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
46
&
but&downstream&from&the&gap&for&longer&dependencies.&However&the&question&arises&of&why&
&
retrieval& does& not& occur& immediately,& rather& than& after& the& position& where& the& gap& site& is&
&
confirmed.&One&possibility&is&simply&that&other&interpretive&relations&are&being&constructed&
&
simultaneously.& After& projecting& the& gap& position,& the& processor& shifts& its& resources& to&
&
interpret&the&direct&object.&Data&from&crossZmodal&priming&studies&are&relevant&here,&but&are&
&
somewhat&mixed&(see&McKoon&&&Ratcliff,&1994,&Nicol,&Fodor&&&Swinney,&1994).&Among&the&
&
studies& that& have& found& evidence& for& reactivation,& Nicol& && Swinney& (1989)& report& that&
&
facilitation&(in&RTs)&for&a&visuallyZpresented,&semantic&associate&of&the&filler&is&found&in&the&
&
position& immediately& following& the& verb.& On& the& other& hand,& in& a& synonym& judgment& task,&
&
McElree&(2001)&obtained&the&greatest&facilitation&(in&accuracy)&several&words&downstream,&
&
after&which&the&gap&had&been&unambiguously&located.&It&is&important&to&note&that&in&both&of&
&
these& cases,& the& gap& was& in& direct& object& position.& In& our& case,& the& gap& was& in& an& oblique&
&
position.&Further&research&is&therefore&needed&to&test&the&generality&of&the&results.&
&
An& alternative& way& of& differentially& privileging& different& kinds& of& information& in&
&
processing& is& through& different& data& types.& Syntactic& constituents& (nodes,& items)& might& be&
&
distinguished&from&syntactic&dependencies&(arcs,&associations)&in&terms&of&their&mnemonic&
&
properties.& This& opens& up& several& degrees& of& freedom.& Firstly,& we& could& grant& that& certain&
&
features&inhere&or&are&bound&to&arcs&while&others&are&bound&to&constituents.&Secondly,&we&
&
could& distinguish& between& predictions& for& particular& constituents& and& predictions& for&
&
dependencies& between& constituents.& Finally,& different& data& types& may& have& their& own&
&
advantages& and& vulnerabilities& in& working& memory:& the& constituents& themselves& could& be&
&
susceptible& to& interference,& that& the& dependencies& linking& them& could& be& susceptible,& or&
&
both.&How&could&this&help&explain&our&data?&If&comprehenders&posit&the&link&between&filler&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
and&gap&predictively,&this&component&of&the&representation&and&not&access&to&the&features&of&
&
the& filler& item& itself& could& drive& the& filledZgap& effect.& The& specific& filler& constituent& would&
&
need&to&be&reactivated&to&interpret&it&and&thus&to&generate&the&anomaly&effect.&Moreover&if&a&
&
single& labeled& dependency& link,& bearing& information& about& syntactic& category,& were& a&
&
relatively& strong& or& unique& component& of& the& representation,& its& distinctness& at& retrieval&
&
could& render& it& more& robust& to& interference& (Nairne,& 2006).& & Thus& an& account& is& possible&
&
without& direct& appeal& to& maintaining& a& syntactic& constituent;& but& robust& predictive&
activation&of&some&information&does&seem&to&be&required.&
&&
&
&
Our& present& data,& and& understanding,& probably& cannot& choose& between& all& these&
&
possibilities,& but& they& do& point& the& way& for& future& investigation& on& how& distinct& features&
&
types&interact&with&one&another&in&working&memory&during&language&processing.&In&the&next&
section,&we&expand&upon&the&relationship&between&maintenance&and&distinctness.&
&
&
&
&
5.3
The*capacity*of*concurrent*processing*and*underspecification**
&
&
The& amount& of& information& that& can& be& concurrently& processed& is& clearly& limited&
&
(Broadbent,& 1958),& but& the& exact& nature& of& these& limitations& for& processing& linguistic&
&
structure& remains& poorly& understood.& For& list& memory& it& is& clear& that& focal& attention& does&
&
not& apply& merely& to& the& last& percept,& and& that& chunking& and& task& expectations& play& an&
&
important& role& in& determining& what& is& maintained& in& focal& attention& (McElree,& 2006).& An&
&
idea&with&deep&roots&in&psycholinguistics&is&that&a&capacity&bottleneck&causes&the&chunking&
&
of&an&expression&into&its&major&constituents,&a&notion&which&motivated&the&click&dislocation&
&
experiments&of&the&1960s&(Bever,&Lackner&&&Kirk,&1969).&More&recently,&in&the&ACTZR&model&
&
of&sentence&processing&(Lewis&&&Vasishth,&2005),&it&is&one&maximal&projection&(i.e.,&an&XP)&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
48
&
that&can&be&maintained&in&the&‘active’&buffer.&What&these&ideas&have&in&common&is&that&the&
&
capacity&bottleneck&functionally&carves&up&expressions&into&contiguous&extents&of&structure.&
&
The& span& of& concurrent& processing& thus& specifically& limits& the& breadth& of& information& that&
&
can&be&represented&at&any&given&moment.&
&
&
An& alternative& view,& that& is& consistent& with& our& proposal& for& processing& fillerZgap&
&
dependencies,&is&that&limits&on&concurrent&processing&exert&their&influence&primarily&on&the&
&
depth&of&information&that&can&be&represented&at&any&given&moment.&Instead&of&maintaining&
&
all&the&features&of&a&given&constituent&with&full&precision,&the&comprehender&may&choose&to&
&
discard& some& features& from& focal& attention,& in& order& to& accommodate& information& from&
&
other&constituents.&For&fillerZgap&dependencies,&maintaining&a&fully&articulated&encoding&of&
&
the&filler&phrase&across&the&entire&extent&of&the&dependency&while&processing&intermediate&
&
portions& of& the& sentence& may& simply& be& impossible.& But& it& may& nonetheless& be& feasible& to&
&
maintain&a&paredZdown&representation&of&the&fillerZphrase,&with&just&enough&information&to&
&
make& the& most& crucial& parsing& decision,& i.e.,& where& to& posit& a& gap.& In& the& shortest& length&
&
conditions,&when&active&effects&were&observed&for&all&contrast&types,&we&propose&that&most&
&
of&the&features&of&the&filler&remained&in&focal&attention&and&were&yet&to&be&displaced.&&&
&
&
There&are&models&which&motivate&the&paring&down&of&concurrentlyZactive&features&in&
&
a& partiallyZconstructed& representation& without& strictly& separating& maintenance& and&
&
retrieval& processes.& For& example,& Oberauer& && Kliegl& (2006)& present& a& formal& model& that&
&
traces&the&construct&of&maintenance&capacity&to&the&underlying&cause&of&feature&overwriting.&
&
In& this& view,& items& are& lost& from& active& processing& because& other& items,& with& similar&
&
component&features,&have&overwritten&them.&Capacity&is&thus&a&function&of&distinctness,&just&
&
like& retrievability& (Nairne,& 2006).& This& model& thus& achieves& a& broader& unification& of&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
interference& theory& and& forgetting.& For& language& processing,& it& could& be& that& certain&
&
syntactic& features& are& simply& more& interferenceZrobust& than& specific& lexical& features.&–& i.e.,&
&
the&syntactic&category&of&the&filler&in&Experiment&1&is&less&likely&to&be&overwritten&than&the&
&
lexical&features&in&Experiment&2.&Those&‘robust’&features&would&constitute&part&of&the&paredZ
&
down& representation.& This& might& seem& initially& implausible,& since& syntactic& categories& are&
&
fewer&in&number&than&the&set&of&potential&lexical&features;&thus&the&syntactic&category&of&the&
&
filler&and&other&constituents&are&more&likely&to&overlap&than&are&their&lexical&features.&But&it&
&
is&important&to&keep&in&mind&that&we&do&not&know&the&exact&lexical&features&responsible&for&
&
anomaly& detection& in& Experiment& 2& or& the& syntactic& features& responsible& for& the& filled& gap&
effect&in&Experiment&1.&&&
&
&
&
Most& generally,& this& line& of& inquiry& raises& the& question& of& whether& we& can& have& a&
&
theory& of& what& kinds& of& information& are& interferenceZrobust& in& language& processing& and&
&
what& kinds& are& not.& Distinctness& is& clearly& a& crucial& part& of& the& picture.& But& we& conjecture&
&
that& a& complementary& limitation& derives& from& the& binding& of& specific& lexical& material& to&
&
positions&in&the&structure,&an&idea&supported&by&recent&computational&modelling&of&parsing&
&
(Van&der&Velde&&&de&Kamps,&2006;&for&nonZlinguistic&information:&Treisman&&&Gelade,&1980,&
&
Oberauer&&&Vockenberg,&2009).&Thus&we&distinguish&the&storage&of&atomic&features&from&the&
&
storage&of&feature/feature&bindings.&In&our&case,&we&might&suppose&that&syntactic&category&
&
features&are&bound&together&with&the&+WH&filler&feature&(i.e.,&introduced&by&‘which’&in&‘which&
&
acorns’),& but& that& lexical& details& are& linked& to& the& NP& restrictor& (e.g.,& ‘acorns’& in& ‘which&
&
acorns’).&Multiple&NPs&were&encountered&along&the&dependency&path&which&could&interrupt&
&
the&binding&of&the&lexical&features&to&the&NP&restrictor&via&overwriting.&However,&no&further&
&
+WH& categories& were& encountered,& effectively& protecting& the& information& bound& to& that&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
50
&
feature.& This& view& has& two& advantages:& firstly,& it& is& consistent& with& the& idea& that& syntactic&
&
categories& are& complex& and& structured& (Jackendoff,& 1977,& Gazdar,& Klein,& Pullum,& && Sag&
&
1985).& Secondly,& it& makes& a& strong& prediction:& if& it& is& correct,& experimentally& introducing&
&
further& (nested)& filler& phrases& (e.g.,& a& relative& clause& attached& to& a& subject)& should& disrupt&
&
the&filledZgap&effect.&&
&
&
Limits& on& the& depth& of& representation& can& be& thought& of& as& a& kind& of&
&
underspecification,& a& concept& which& has& received& renewed& attention& in& psycholinguistics,&
&
(Frisson,&2009;&Sturt&et&al.,&2004;&Weinberg,&1993).&By&underspecifying&most&constituents&in&
&
the&syntactic&context,&the&comprehender&frees&up&capacity&to&encode&and&process&incoming&
&
information,&but&nonetheless&has&at&the&ready&some&amount&of&global&context&upon&which&to&
&
base& parsing& decisions.& Though& locality& biases& are& evident& in& many& comprehension&
&
processes,& nonZlocal& dependencies& can& often& be& constructed& with& great& accuracy& (Phillips,&
&
Wagers& && Lau,& 2011).& Thus& the& ability& to& have& ready& access& to& information& about& global&
&
structure& would& be& an& advantage.& Understanding& the& generality& of& this& conclusion& will&
&
depend& on& examining& other& species& of& linguistic& dependency& in& a& way& that& varies& the&
&
predictability&and&distinctness&of&their&dependent&elements.
&
&
&
&
&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
REFERENCES&
Baayen,&H.&(2008).&Analyzing*Linguistic*Data.*Cambridge,&UK:&Cambridge&UP.&
&
&
PA&
&
&
&
Baayen,& R.H.& && Milin,& P.& (2010).& Analyzing& reaction& times.& International* Journal* of*
Psychological*Research,*3,*12Z28.&
&
&
&
&
Bates,& D.,& Maechler,& M.& && Bolker& B.& (2011).& lme4:& Linear& mixedZeffects& models& using& S4&
classes.&R&package&version&0.999375Z42.&http://CRAN.RZproject.org/package=lme4&
&
&
&
&
Bever,&T.&G.,&Lackner,&J.&R.&&&Kirk,&R.&(1969).&The&underlying&structures&of&sentences&are&the&
primary&units&of&immediate&speech&processing.*Perception*&*Psychophysics,*5,&225–234&
&
&
&
&
Bourdages,&J.S.&(1992).&Parsing&complex&NPs&in&French.&In&Goodluck,&H.,&&&Rochemond,&M.S.&
&
(eds.),&Island*constraints:*Theory,*acquisition*and*processing.*Dordrecht:&Kluwer&Broadbent&
1958&
&
Broadbent,&D.&E.&(1958).&Perception&and&communication.&New&York:&Oxford&UP.&&
&
&
&
&
&
&
Chomsky,&N.&(1977).&On&WhZMovement.&In&Culicover,&P.W.,&Wasow,&T.,&&&A.&Akmajian,&eds.,&
Formal*syntax.&New&York:&Academic&Press.&71Z132.&
&
Chomsky,&N.&(1987).&Barriers.&Cambridge,&MA:&MIT&Press.&
&
&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
52
&
Cowan,& N.& (2001).& The& magical& number& 4& in& shortZterm& memory:& a& reconsideration& of&
&
mental&storage&capacity.&Behavioral*and*Brain*Sciences&24,&87Z185.&
&
&
&
Crain,&S.,&&&Fodor,&J.D.&(1985).&How&can&grammars&help&parsers?&In&Dowty,&D.,&Kartunnen,&L.,&
&
&&Zwicky,&A.&(eds.),&Natural*language*parsing.&Cambridge:&Cambridge&University&Press.&
&
&
&
Ericsson,& K.& A.,& && Kintsch,& W.& (1995).&& LongZterm& working& memory.&& Psychological* Review,*
&
102(2),&211Z245.&&
&
&
&
Faul,& F.,& Erdfelder,& E.,& Lang,& A.ZG.,& && Buchner,& A.& (2007).& G*Power& 3:& A& flexible& statistical&
&
power& analysis& for& the& social,& behavioral,& and& biomedical& sciences.&Behavior* Research*
&
Methods,*39,&175Z191..&
&
&
&
Fiebach,&C.J.,&Schlesewsky,&M.,&&&Friederici,&A.D.&(2002).&Separating&syntactic&memory&costs&
&
and& syntactic& integration& costs& during& parsing:& The& processing& of& German& WHZquestions.&
&
Journal*of*Memory*and*Language,&47,&250Z272.&
&
&
&
Ford,& M.& (1983).& A& method& of& obtaining& measures& of& local& parsing& complexity& throughout&
&
sentences.&Journal*Verbal*Learning*&*Verbal*Behavior,*22,&203–18.&
&
&
&
Frazier,& L.& (1987).& Syntactic& processing:& Evidence& from& Dutch.& Natural* Language* and*
&
Linguistic*Theory,*5,&519–560.&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
Frazier,&L.&&&Clifton,&C.,&Jr.&(1989)&Successive&cyclicity&in&the&grammar&and&the&parser.&
Language*and*Cognitive*Processes,&4,&93Z126.&
&
&
PA&
&
&
&
Frazier,& L.,& && Flores& D’Arcais,& G.B.& (1989).& Filler–driven& parsing:& a& study& of& gap& filling& in&
Dutch.&Journal&of&Memory&of&Language,&28,&331–44.&
&
Frisson,&S.&(2009).&Semantic&Underspecification&in&Language&Processing.&Language*and*
Linguistics*Compass,&3,&111Z127.&&
&
&
&
&
&
&
&
Garavan,&H.&(1998).&Serial&attention&within&working&memory.&Memory&&&Cognition,&26,&263Z
&
276.&
&
&
&
Garnsey,&S.M.,&Tanenhaus,&M.K.&&&Chapman,&R.M.&(1989).&Evoked&potentials&and&the&study&of&
&
sentence&comprehension.&Journal*of*Psycholinguistic*Research,*18,&51–60.&
&
&
&
Gazdar,& G.,& Klein,& E.,& Pullum,& G.,& and& Sag,& I.& (1985)& Generalized* Phrase* Structure* Grammar.*
&
Harvard&University&Press,&Cambridge,&Mass.&
&
&
&
Gibson,&E.&(1998).&Syntactic&complexity:&Locality&of&syntactic&dependencies.&Cognition,&68,&1–
&
76.&
&
&
&
Jackendoff,&R.&(1977).&X4Bar*Syntax:*a*Study*of*Phrase*Structure.*Cambridge,&MA:&MIT&Press.&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
54
&
Jonides,&J.,&Lewis,&R.L.,&Nee,&D.E.,&Lustig,&C.A.,&Berman,&M.G.,&&&Moore,&K.S.&(2008).&The&mind&
&
and&brain&of&shortZterm&memory.&Annual*Review*of*Psychology,*59,*15.1Z15.32&
&
&
&
Just,& M.A.,& Carpenter,& P.A.,& and& Wooley,& J.D.& (1982).& Paradigms& and& Processes& in& Reading&
&
Comprehension.&Journal*of*Experimental*Psychology:*General&,&111&(2),&228Z238.&&
&
&
&
King,& J.,& && Just,& M.A.& (1991).& Individual& differences& in& syntactic& processing:& the& role& of&
&
working&memory.&Journal*of*Memory*and*Language,&30,&580–60.&
&
&
&
Kluender,& R.,& && Kutas,& M.& (1993).& Subjacency& as& a& processing& phenomenon.& Language* and*
&
Cognitive*Processes,&8,&573Z633.&
&
&
&
Lee,&M.ZW.&(2004).&Another&look&at&the&role&of&empty&categories&in&sentence&processing&(and&
&
grammar).&Journal*of*Psycholinguistic*Research,*33,&51–73.&
&
&
&
Lewis,&R.&L.&(2000).&Specifying&architectures&for&language&processing:&Process,&control,&and&
&
memory& in& parsing& and& interpretation.& In& M.& W.& Crocker,& M.& Pickering,& && C.& Clifton& (Eds.),&
&
Architectures& and& mechanisms& for& language& processing& (pp.& 56–89).& Cambridge,& UK:&
&
Cambridge&University&Press.&
&
Lewis,&R.&L.,&&&Vasishth,&S.&(2005).&An&ActivationZBased&Model&of&Sentence&Processing&as&
Skilled&Memory&Retrieval.&Cognitive*Science,&29,&375Z419.&
&
&
&
&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
McElree,&B.&(2000).&Sentence&comprehension&is&mediated&by&contentZaddressable&memory.&
Journal*of*Psycholinguistic*Research,*29(2),&111Z123.&
&
&
&
&
McElree,&B.&(2001).&Working&memory&and&focal&attention.&J*Exp*Psychol:&Learn.*Mem.*Cogn.*
&
27,!817435.*
&
&
&
McElree,&B.&(2006).&Accessing&recent&events.&In&B.&H.&Ross&(Ed.),&The*psychology*of*learning*
&
and*motivation,&vol.&46.&San&Diego:&Academic&Press.&
&
&
&
McElree,&B.,&&&Dosher,&B.&A.&(1989).&Serial&position&and&set&size&in&shortZterm&memory:&&Time&
&
course&of&recognition.&Journal*of*Experimental*Psychology:*General,*118,&346Z373.&&
&
&
&
McElree,& B.,& Foraker,& S.,& && Dyer,& L.& (2003).& Memory& structures& that& subserve& sentence&
&
comprehension.&Journal*of*Memory*&*Language,*48(1),&67Z91.&&
&
&
&
McElree,& B.,& && Griffith,& T.,& (1998).& Structural& and& lexical& constraints& on& filling& gaps& during&
&
sentence& comprehension:& A& timeZcourse& analysis.& Journal* of* Experimental* Psychology:*
&
Learning,*Memory,*and*Cognition*24:&432Z60.&
&
&
&
McKinnon,& R.,& && Osterhout,& L.& (1996).& Constraints& on& movement& phenomena& in& sentence&
&
processing:&Evidence&from&eventZrelated&potentials.&Language*and*Cognition,*11,&495Z523.&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
56
&
McKoon&G&&&Ratcliff&R&(1994)&Sentential&context&and&onZline&lexical&decision&tasks.&Journal*of*
&
Experimental*Psychology:*Language,*Memory*and*Cognition,&20,&1239Z43.&
&
&
&
Nairne,&J.&S.&(2006).&Modeling&Distinctiveness:&Implications&for&General&Memory&Theory.&In&
&
R.&R.&Hunt&&&J.&B.&Worthen&(Eds.),&Distinctiveness*and*Memory*(pp.&27–46).&Oxford&University&
&
Press,&USA.&
&
&
&
Neville,&H.J.,&Nicol,&J.,&Barss,&A.,&Forster,&K.,&Garrett,&M.&(1991).&SyntacticallyZbased&sentence&
&
processing& classes:& evidence& from& eventZrelated& potentials.& Journal* of* Cognitive*
&
Neuroscience,*3,&151Z165.&
&
&
&
Nicol,& J.L.,& Fodor,& J.D.,& && Swinney,& D.& (1994).& & Using& cross–modal& lexical& decision& tasks& to&
&
investigate& sentence& processing.& & Journal* of* Experimental* Psychology:* Learning,* Memory* &*
&
Cognition,*20,&1229–38.&
&
&
&
Nicol,& J.,& && Swinney,& D.& (1989).& The& role& of& structure& in& coreference& assignment& during&
&
sentence&comprehension.&Journal*of*Psycholinguistic*Research,*18,&5Z19.&
&
&
&
Oberauer,& K.,& && Kliegl,& R.& (2006).& A& formal& model& of& capacity& limits& in& working& memory.&
&
Journal*of*Memory*and*Language,*55,*601Z626.&
&
&
&
Oberauer,& K.,& && Vockenberg,& K.& (2009).& Updating& of& working& memory:& lingering& bindings.&
&
The*Quarterly*Journal*of*Experimental*Psychology,*62,*967Z987.&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
&
PA&
&
Omaki,& A.,& && Schulz,& B.& (2011).& FillerZgap& dependencies& and& island& constraints& in& secondZ
language&sentence&processing.&Studies*in*Second*Language*Acquisition,*33,*563Z588.&
&
&
&
&
Omaki,&A.,&Lau,&E.,&Davidson&White,&I.,Phillips,&C.&(2010).&HyperZactive&gap&filling:&PreZverbal&
object&gap&creation&in&English&fillerZgap&dependency&processing.&Submitted.&
&
Phillips,&C.&(2006).&The&real–time&status&of&island&phenomena.&Language,&82,&795–823.&
&
&
&
&
&
&
Phillips,& C.,& Kazanina,& N.,& && Abada,& S.H.& (2005).& ERP& effects& of& the& processing& of& syntactic&
long–distance&dependencies.&Cognitive*Brain*Research,*22,&407–428.&&
&
&
&
&
Phillips,& C.& && Wagers,& M.W.& (2007).& Relating& Structure& and& Time& in& Linguistics& and&
&
Psycholinguistics.&In&Gaskell,&G.&(ed.),&Oxford*Handbook*of*Psycholinguistics,&Oxford:&Oxford&
University&Press.&
&
&
&
&
Phillips,& C.,& Wagers,& M.,& && Lau,& E.& (2011).& Grammatical& illusions& and& selective& fallibility& in&
&
realZtime&language&comprehension.&In&Runner,&J.&(ed.),&Experiments*at*the*Interfaces,*Syntax*
and*Semantics,*Vol*37,*Bingley,&UK:&Emerald&Publications.&
&
&
Pickering,& M.J.,& Barton,& S.,& && Shillcock,& R.& (1994).& Unbounded& dependencies,& island&
&
constraints& and& processing& complexity.& In& Clifton,& C.,& Frazier,& L.,& && Rayner,& K.& (eds.),&
&
Perspectives*on*Sentence*Processing.*London:&Lawrence&Erlbaum.&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
58
&
R&Development&Core&Team&(2011).&R:&A&language&and&environment&for&statistical&computing.&
&
R&Foundation&for&&Statistical&Computing,&Vienna,&Austria.&http://www.RZproject.org/.&
&
&
&
Ratcliff,& R.& (1993).& Methods& for& dealing& with& reaction& time& outliers.& Psychological* Bulletin,&
&
114,&510Z532.*
&
&
&
Ross,& J.& R.& (1967).& Constraints& on& variables& in& syntax.& Ph.D.& Dissertation,& Massachusetts&
&
Institute&of&Technology.&&
&
&
&
Ruchkin,&D.S.,&Johnson,&R.,&Jr.,&Canoune,&H.,&&&Ritter,&W.&(1990).&ShortZterm&memory&storage&
&
and&retention:&An&eventZrelated&brain&potential&study.&Electroencephalography*and*Clinical*
&
Neurophysiology,&76,&419Z39.&
&
&
&
Simmons,&J.P.,&Nelson,&L.D.,&&&Simonsohn,&U.&(2011).&FalseZpositive&psychology:&undisclosed&
&
flexibility& in& data& collection& and& analysis& allows& presenting& anything& as& significant.&
&
Psychological*Science,*22,*1359Z1366.&
&
&
&
Sturt,& P.,& Sanford,& A.J.,& Stewart,& A.,& Dawydiak,& E.& (2004).& Linguistic& focus& and& goodZenough&
&
representations:&an&application&of&the&change&detection&paradigm.&Psychonomic*Bulletin*and*
&
Review,*11,*882Z888.&
&
&
&
Staub,&A.&&(2007).&&The&parser&doesn't&ignore&intransitivity,&after&all.&&Journal*of*Experimental*
&
Psychology:*Learning,*Memory,*and*Cognition,*33,&550Z569.&&
&
&
&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
&
PA&
&
&
Stowe,& L.& (1986).& Parsing& wh–constructions:& evidence& for& on–line& gap& location.& Language*
and*Cognitive*Processes,&1,&227–46.&
&
&
&
&
Traxler,& M.J.,& && Pickering,& M.J.& (1996).& Plausibility& and& the& processing& of& unbounded&
dependencies.&Journal*of*Memory*and*Language,&35,&454–475.&
&
&
&
&
Traxler,& M.J.,& Morris,& R.K.,& && Seely,& R.E.& (2002).& Processing& subject& and& object& relative&
clauses:&evidence&from&eye&movements.&Journal*of*Memory*and*Language,*47,*69Z90.&
&
&
&
&
Treisman,& A.M.,& && Gelade,& G.& (1980).& A& featureZintegration& theory& of& attention.& Cognitive*
&
Psychology,*12,*97Z136.&
&
&
&
van&der&Velde,&F.&&&de&Kamps,&M.&(2006).&Neural&blackboard&architectures&of&combinatorial&
structures&in&cognition.&Behavioral*and*Brain*Sciences,*29,&37Z70.&
&
&
&
&
Verhaegen,&P.,&Cerella,&J.,&Basak,&C.&(2004).&A&working&memory&workout:&how&to&expand&the&
&
focus& of& serial& attention& from& one& to& four& items& in& 10& hours& or& less.& J.* Exp.* Psychol.:* Learn.*
Mem.*Cogn.*30,&1322Z37.&
&
&
&
&
&
Wagers,&M.,&&&Phillips,&C.&(2009).&Multiple&dependencies&and&the&role&of&the&grammar&in&realZ
time&comprehension.&Journal*of*Linguistics,*45,&395Z433.&
&
&
&
&
&
PA&
MEMORY&AND&CONTROL&PROCESSES&IN&ACTIVE&DEPENDENCY&CONSTRUCTION&
60
&
Wanner,& E.,& && Maratsos,& M.& (1978).& An& ATN& Approach& to& Comprehension.& In& Halle,& M.,&
&
Bresnan,&J.,&&&Miller,&G.A.&(eds.),&Linguistic*Theory*and*Psychological*Reality,&Cambridge,&MA:&
&
MIT&Press.&
&
&
&
Weinberg,& A.& (1993).& Parameters& in& the& theory& of& sentence& processing:& Minimal&
&
Commitment&theory&goes&east.&Journal*of*Psycholinguistic*Research,*22,*339Z364.
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
&
TABLE&1&
Sample&materials&set&for&Experiment&1&
&
!Each!of!the!six!complete!sentences!in!this!set!is!derived!by!filling!in!the!underscored!
blank!in!the!Item!Frame!with!the!phrases!in!the!LENGTH!×!FILLER!cells.!Filler!phrases!
are!indicated!in!bold!font,!the!critical!filledEgap!region!is!indicated!by!double!
underlining,!and!the!gap!is!indicated!an!underscore.!In!the!Short!Length!conditions!a!
fiveEword!preamble!also!preceding!the!Item!Frame!to!control!for!the!ordinal!
position!of!the!critical!region.!For!this!set,!the!preamble!was!“The!biologist!was!
distressed!that!...”!
!
Sample!Materials!Set!for!Experiment!1!
LENGTH!
FILLER!
Short!
NP!
CATEGORY!
PP!
+PP!
NP!
PP!
+CP!
NP!
PP!
!
&
ITEM!FRAME:!
“...![T]he!chemicals!...!_________!...!might!still!become!contaminated!
…”!
…!which!the!technician!carefully!prepared!the!clean!tubes!for!___!
while!wearing!a!mask!…!
…!for!which!the!technician!carefully!prepared!the!clean!tubes!!___!
while!wearing!a!mask!…!
…!which!the!technician!at!the!medical!research!facility!carefully!
prepared!the!clean!tubes!for!___!while!wearing!a!mask!…!
…!for!which!the!technician!at!the!medical!research!facility!
carefully!prepared!the!clean!tubes!___!while!wearing!a!mask!…!
…!which!the!young!biologist!said!that!the!technician!carefully!
prepared!the!clean!tubes!for!___!while!wearing!a!mask!…!
…!for!which!the!young!biologist!said!that!the!technician!carefully!
prepared!the!clean!tubes!___!while!wearing!a!mask!…!
!
!
TABLE&2&
Experiment&1&Comprehension&accuracy&
&
Average!percentage!correct!over!participants,!with!row,!column!and!grand!means.!
Standard!error!of!the!cell!means!across!subjects!is!reported!in!parentheses.!N!=!36!
!
LENGTH!
ACCURACY&
NP&
FILLER!CATEGORY!
PP&
!
Short&
88%!
(2%)!
88%!
(2%)!
88%!
+PP!
87%!
(3%)!
84%!
(3%)!
86%!
!
+CP!
78%!
(3%)!
77%!
(3%)!
77%!
!
84%!
83%!
84%!
TABLE&3&
!
&
Experiment&1&MixedBeffect&model&coefficient&tables!
Left:&models!on!residualEtrimmed!data.!Right:&on!untrimmed!data.!In!FILLER!contrasts,!NP!
conditions!are!positive!(+1);!in!LENGTH.LONG!contrasts,!Short!conditions!are!+2/3;!in!
LENGTH.CLAUSE!contrasts,!+PP!conditions!are!(+1/2).!!P0values!are!reported!rounded!to!2!
decimal!places,!except!where!they!are!less!than!.005.!Dark!shading!highlights!coefficients,!
excluding!the!intercept,!with!a!pEvalue!less!than!.05;!light!shading!for!.05!≤!p!<!.10.!
RESIDUAL&TRIMMED&DATASET&
ADVERB!
ESTIMATE!
INTERCEPT!
!
FILLER!
LEN.LONG:!!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
VERB!
INTERCEPT!
FILLER!
LEN.LONG:!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
ACTIVE!FILLING!
OBJECT!NP!
INTERCEPT!
!
FILLER!
LEN.LONG:!!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
GAP2DRIVEN!
INTERCEPT!
FILLER!
LEN.LONG:!!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
STD.!!
t!
pMCMC&
EST.!
.056!
.011!
100! <!.001!
E.29! .77!
.023!
E1.1!
ERROR!
t!
pMCMC&
.051!
.008!
110!
E.45!
<!.001!
.65!
5.90!
E.003!
E.049!
.017!
E2.9!
<!.005!
E.026!
E.028!
.039!
E.001!
.020!
.017!
.020!
STD.!!
E1.4!
2.3!
E.07!
.16!
.02!
.95!
E.029!
E.016!
E.007!
t!
pMCMC&
EST.!
5.91!
E.003!
.051!
.009!
120!
E.40!
<!.001!
.69!
5.96!
.002!
.027! E1.1! .27!
E.023! E.27! .50!
.027! .70!
.81!
STD.!!
t!
pMCMC&
ERROR!
.055!
110! <!.001!
.012!
.19!
.85!
E.001!
.019!
E.07!
.95!
E.011!
.026!
E.43!
.66!
E.031!
.020!
E.013!
.022!
.019!
.022!
STD.!!
E1.4!
1.1!
E.58!
.15!
.29!
.56!
E.050!
.009!
E.014!
E1.7!
.34!
E.48!
.10!
.74!
.63!
t!
pMCMC&
EST.!
.030!
.026!
.030!
STD.!!
t!
pMCMC&
5.87!
.01!
.038!
.004!
140!
2.4!
<!.001!
.02!
5.91!
.009!
.046!
.006!
130! <!.001!
1.6!
.11!
.003!
.008!
.35!
.73!
.008!
.013!
.67!
.50!
E.029!
E.001!
.001!
.010!
.009!
.010!
STD.!!
E2.7!
E.09!
.10!
.008!
.93!
.92!
E.03!
.01!
.01!
E1.9!
.93!
.67!
.06!
.35!
.50!
t!
pMCMC&
EST.!
.015!
.013!
.015!
STD.!!
t!
pMCMC&
5.86!
E.006!
.04!
0.004!
156!
E1.5!
<!.001!
.15!
5.90!
.001!
.041!
.006!
140! <!.001!
.08!
.93!
E.02!
.009!
E2.2!
.03!
E.041!
.013!
E3.2!
<!.005!
E.009!
E.020!
.020!
.011!
.009!
.011!
E.83!
E2.1!
1.8!
.41!
.04!
.07!
.008!
E.024!
0.032!
.015!
.013!
.015!
.57!
E1.8!
2.2!
.57!
.07!
.03!
ESTIMATE!
ESTIMATE!
!
ERROR!
UNTRIMMED&DATASET&
5.87!
E.004!
ESTIMATE!
!
STD.!!
ERROR!
ERROR!
ERROR!
ERROR!
ERROR!
.26!
&
TABLE&&4&
Sample&materials&set&from&Experiment&2&
&
The!plausibility!manipulation!is!indicated!with!bold!face,!the!critical!region!with!
double!underlining,!and!the!gap!position!with!an!underscore.!
!
!
Length&
Short&
+PP&
+CP&
!
!
!
Filler&
Sentence&
plausibility&
Plausible!
The!acorns!which!the!squirrels!quickly!crammed!their!small!
puffy!cheeks!with!___!before!scurrying!out!of!the!park!had!fallen!
from!a!sickly!oak.!
Implausible! The!cats!which!the!squirrels!quickly!crammed!their!small!puffy!
cheeks!with!___!before!scurrying!out!of!the!park!were!basking!in!
the!sun.!
Plausible!
The!acorns!which!the!squirrels!with!the!bushy!black!tails!quickly!
crammed!their!small!puffy!cheeks!with!___!before!scurrying!out!of!
the!park!had!fallen!from!a!sickly!oak.!
Implausible! The!cats!which!the!squirrels!with!the!bushy!black!tails!quickly!
crammed!their!small!puffy!cheeks!with!___!before!scurrying!out!of!
the!park!were!basking!in!the!sun.!
Plausible!
The!acorns!which!the!experienced!naturalist!concluded!that!the!
squirrels!quickly!crammed!their!small!puffy!cheeks!with!___!
before!scurrying!out!of!the!park!had!fallen!from!a!sickly!oak.!
Implausible! The!cats!which!the!experienced!naturalist!concluded!that!the!
squirrels!quickly!crammed!their!small!puffy!cheeks!with!___!
before!scurrying!out!of!the!park!were!basking!in!the!sun.!
!
TABLE&5&
Experiment&2&Comprehension&accuracy&
Average!percentage!correct!over!participants,!with!row,!column!and!grand!means.!
Standard!error!of!the!cell!means!across!subjects!is!reported!in!parentheses.!N!=!35.!
!
ACCURACY&
Plausible&
Filler&Category&
Implausible&
!
Short&
91%!
(2%)!
90%!
(3%)!
91%!
Length&
+PP!
95%!
(2%)!
94%!
(2%)!
94%!
+CP!
89%!
(3%)!
88%!
(3%)!
89%!
!
!
92%!
90%!
91%!
TABLE&6&
Experiment&2&MixedBeffect&model&coefficient&tables!
Left:&models!on!residualEtrimmed!data.!Right:&on!untrimmed!data.!In!FILLER!contrasts,!
Implausible!fillers!are!positive!(+1);!in!LENGTH.LONG!contrasts,!Short!conditions!are!+2/3;!in!
LENGTH.CLAUSE!contrasts,!+PP!conditions!are!(+1/2).!!P0values!are!reported!rounded!to!2!
decimal!places,!except!where!they!are!less!than!.005.!Dark!shading!highlights!coefficients,!
excluding!the!intercept,!with!a!pEvalue!less!than!.05;!light!shading!for!.05!≤!p!<!.10.!
!
&
RESIDUAL&TRIMMED&DATASET&
ADVERB!
INTERCEPT!
ESTIMATE!
!
FILLER!
LEN.LONG:!!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
ACTIVE!FILLING!
VERB!
INTERCEPT!
!
FILLER!
LEN.LONG:!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
ACTIVE!FILLING!
OBJECT!NP!
INTERCEPT!
!
FILLER!
LEN.LONG:!!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
GAP2DRIVEN!
INTERCEPT!
FILLER.!
LEN.LONG:!!
SHT!–MEAN(CP,PP)!
LEN.CLS:!PP!E!CP!
FILLER:LEN.LONG!
FILLER:LEN.CLS!
ERROR!
UNTRIMMED&DATASET&
STD.!!
t!
pMCMC&
EST.!
.055!
.011!
110! <!.001!
E.29! .34!
ERROR!
t!
pMCMC&
5.80!
.006!
.04!
.008!
140!
.77!
<!.001!
.44!
5.90!
E.003!
.001!
.016!
.06!
.95!
E.029!
.027!
E1.1!
.40!
.018!
.011!
E.005!
.019!
.02!
.01!
STD.!!
.95!
.71!
E.26!
.34!
.48!
.79!
E.026!
E.007!
0.016!
E1.1!
E.03!
.69!
.19!
.20!
.75!
t!
pMCMC&
EST.!
.023!
.027!
.023!
STD.!!
t!
pMCMC&
5.84!
E.004!
.043!
.008!
135!
E.51!
<!.001!
.62!
5.88!
.005!
.046!
.011!
128! <!.001!
.44! 0.67!
E.018!
.018!
E1.0!
.33!
E.012!
.024!
E.51!
.61!
E.0002!
.043!
.003!
.020!
.018!
.020!
STD.!!
.01!
2.4!
.16!
.98!
.02!
.87!
E.001!
.022!
.013!
E.05!
.93!
.47!
.96!
0.35!
0.67!
t!
pMCMC&
EST.!
.027!
.023!
.027!
STD.!!
t!
pMCMC&
5.81!
.01!
0.035!
.004!
160!
3.4!
<!.001!
<!.001!
5.85!
.02!
0.04!
.005!
160! <!.001!
3.9! <!.001!
.01!
.001!
1.3!
.18!
.02!
.01!
1.6!
.12!
.01!
.02!
.01!
.009!
.008!
.009!
STD.!!
1.4!
2.5!
1.3!
.16!
.01!
.20!
.01!
.03!
.03!
.90!
2.9!
2.3!
.37!
<!.005!
t!
pMCMC&
EST.!
.01!
.01!
.01!
STD.!!
t!
pMCMC&
5.82!
.02!
.034!
.004!
170!
4.6!
<!.001!
<!.001!
5.85!
.03!
.035!
.006!
170! <!.001!
5.5! <!.001!
E.005!
.009!
E.52!
.60!
E.012!
.012!
E1.0!
.31!
E.019!
.01!
E.02!
.010!
.009!
.011!
E1.7!
1.1!
E1.9!
.09!
.29!
.06!
E.017!
.008!
E.002!
.014!
.012!
.014!
E1.2!
0.64!
E.12!
.23!
.53!
.91!
ESTIMATE!
ESTIMATE!
ESTIMATE!
!
STD.!!
ERROR!
ERROR!
ERROR!
ERROR!
ERROR!
ERROR!
.02!
TABLE&7&
Sample&material&set&for&Experiment&3&
The!Match!manipulation!is!indicated!with!boldface,!the!start!of!the!critical!region!
with!double!underlining,!and!the!gap!position!with!an!underscore.!
!
!
Length& Match&
Short&
Long&
!
!
Item!Frame:&
“The!courier!…!unfortunately!wrecked!his!bike!in!traffic.”!
Match&
…!to&whom&the!secretary!warily!entrusted!the!confidential!
business!correspondence!___!after!some!hesitation!…!
Mismatch& …!from&whom!the!secretary!warily!entrusted!the!confidential!
business!correspondence!___!after!some!hesitation!…!
Match&
…!to&whom!the!secretary!for!the!highEpowered!defense!attorney!
warily!entrusted!the!confidential!business!correspondence!___!
after!some!hesitation!…!
Mismatch& …!from&whom!the!secretary!for!the!highEpowered!defense!
attorney!warily!entrusted!the!confidential!business!
correspondence!___!after!some!hesitation!…!
!
TABLE&8&
Experiment&3&Comprehension&Accuracy&
Average!percentage!correct!over!participants,!with!row,!column!and!grand!means.!
Standard!error!of!the!cell!means!across!subjects!is!reported!in!parentheses.!!N!=!54.!
&
Length&
Short&
Long&
85%!!
85%!
(2%)!
(2%)!
86%!
80%!
(2%)!
(2%)!
85%!
82%!
ACCURACY&
Match&
Filler&
Subcat&
Match&
Mismatch&
!
&
!
&
!
!
85%!
83%!
84%!
TABLE&9&
Experiment&3&MixedBeffect&model&coefficient&tables!
Left:&models!on!residualEtrimmed!data.!Right:&on!untrimmed!data.!In!FILLER!contrasts,!
Mismatch!conditions!are!positive!(+1);!in!LENGTH!contrasts,!Long!conditions!are!+1.!P0
values!are!reported!rounded!to!2!decimal!places,!except!where!they!are!less!than!.005.!Dark!
shading!highlights!coefficients,!excluding!the!intercept,!with!a!pEvalue!less!than!.05;!light!
shading!for!.05!≤!p!<!.10.!
!
&
RESIDUAL&TRIMMED&DATASET&
ADVERB!
INTERCEPT!
ESTIMATE!
!
FILLER!
LENGTH!
FILLER:LENGTH!
ACTIVE!FILLING!
VERB!
INTERCEPT!
!
FILLER!
LENGTH!
FILLER:LENGTH!
ACTIVE!FILLING!
OBJECT!NP!
INTERCEPT!
!
FILLER!
LENGTH!
FILLER:LENGTH!
GAP2DRIVEN!
INTERCEPT!
FILLER!
LENGTH!
FILLER:LENGTH!
!
!
!
!
!
!
!
!
!
!
!
&
6.16!
E.002!
.009!
E.002!
ESTIMATE!
6.19!
.002!
.001!
.001!
ESTIMATE!
6.11!
.008!
.003!
E.012!
ESTIMATE!
!
6.09!
.006!
.0001!
E.005!
STD.!!
UNTRIMMED&DATASET&
t!
pMCMC&
EST.!
.053!
.007!
.007!
.007!
STD.!!
120!
E.23!
1.3!
E.35!
<.001!
.81!
.19!
.64!
6.21!
E.009!
E.001!
E.005!
t!
pMCMC&
EST.!
.046!
.008!
.008!
.008!
STD.!!
130!
.24!
.18!
.12!
<.001!
.80!
.84!
.90!
6.23!
E.007!
E.011!
.009!
t!
pMCMC&
EST.!
160!
2.6!
.95!
E3.8!
<.001!
<.001!
6.15!
.005!
E.004!
E.014!
t!
pMCMC&
EST.!
160!
1.6!
.03!
E1.6!
<.001!
6.13!
.007!
E.004!
E.005!
ERROR!
ERROR!
ERROR!
.039!
.003!
.003!
.003!
STD.!!
ERROR!
.036!
.003!
.003!
.003!
.01!
.36!
.10!
.98!
.13!
STD.!!
ERROR!
.060!
.010!
.010!
.010!
STD.!!
ERROR!
t!
pMCMC&
100!
E.83!
E.10!
E.47!
<.001!
t!
pMCMC&
.42!
.93!
.60!
.050! 120! <.001!
.012! E.65!
.51!
.012! E.95!
.36!
.012! .75!
.48!
STD.!!
t!
pMCMC&
ERROR!
.05! 130! <.001!
.005! 1.1!
.29!
.005! E.73!
.47!
.005! E2.8! .005!
STD.!!
t!
pMCMC&
ERROR!
.040! 150! <.001!
.005! 1.4!
.17!
.005! E.85!
.40!
.005! E.91!
.36!
6.1
5.8
5.9
6.0
NP filler
PP filler
gap−driven
active filling
5.7
reading time (ln ms)
Short
11
12
13
14
15
16
17
18
19
20
21
region
xrange
6.1
6.0
5.9
5.8
11
12
13
14
15
16
17
18
19
20
21
region
xrange
5.8
5.9
6.0
6.1
+CP
gap−driven
active filling
5.7
reading time (ln ms)
gap−driven
active filling
5.7
reading time (ln ms)
+PP
11
12
13
14
15
16
17
18
19
20
21
region
Figure'1'
Experiment'1'Reading'Times'
Each!panel!corresponds!to!a!different!Length!condition.!Closed'
symbols!represent!NP!fillers,!open!symbols!PP!fillers.!Reading!times!
are!reported!in!natural!log!ms!(ln!ms).!Error!bars!indicate!standard!
error!of!the!cell!means.'
!
Sample'sentence:'
[…]!the!chemicals!(for)!which![…]CP!the!technician![…]PP!
carefully12!prepared13!the14!clean15!tubes16!(for17)!while18!wearing19!
a20!mask21!might!still!become!contaminated.!
!
!
!
5.8
6.0
plausible filler
implausible filler
gap−driven
active filling
5.6
reading time (ln ms)
Short
10
11
12
13
14
15
16
17
18
19
20
21
region
xrange
6.0
5.8
10
11
12
13
14
15
16
17
18
19
20
21
region
xrange
5.8
6.0
+CP
gap−driven
active filling
5.6
reading time (ln ms)
gap−driven
active filling
5.6
reading time (ln ms)
+PP
10
11
12
13
14
15
16
17
18
19
20
21
region
Figure'2'
!
!
Experiment'2'Reading'times!
Each!panel!corresponds!to!a!different!Length!condition.'Closed'
symbols!represent!Plausible!fillers,!open'symbols!Implausible!fillers.!
Reading!times!are!reported!in!natural!log!ms!(ln!ms).!Error!bars!
indicate!standard!error!of!the!cell!means.!
!
Sample'sentence:!
!!!!!!!!!!!!!…!x!…!]9P10!quickly11!crammed12!their13!small14!puffy15!cheeks16!with17!!
!
___!before18!scurrying19!out20![of!the!park!…!]>20!
!
6.0
6.1
6.2
filler match
filler mismatch
gap−driven
active filling
5.9
reading time (ln ms)
6.3
Short
12
13
14
15
16
17
18
19
20
21
region
xrange
6.2
6.1
6.0
gap−driven
active filling
5.9
reading time (ln ms)
6.3
Long
12
13
14
15
16
17
18
19
20
21
region
Figure'3'
!
Experiment'3'Reading'times'
Each!panel!corresponds!to!a!different!Length!condition.!Closed'
symbols'represent!Subcategorization!Matching!fillers,!open'symbols!
Mismatching!fillers.!Reading!times!are!reported!in!natural!log!ms!(ln!
ms).!Error!bars!indicate!standard!error!of!the!cell!means.!
Sample'sentence:'
The!courier!to/from!whom!the!secretary![for!the!highPpowered!
defense!attorney]!…!warily12!entrusted13!the14!confidential15!
business16!correspondence17!after18!some19!hesitation20!…!
!
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement