parsing with focus particles in context: eye ... · 14 and interrogative contexts jointly...

2

3

4

567

10

111213141516171819

20

21

22232425262728

Journal of

YJMLA 3083 No. of Pages 23, DTD=5.0.1

22 August 2005 Disk Used J. Jayalakshmi (CE) / Anand (TE)ARTICLE IN PRESS

Journal of Memory and Language xxx (2005) xxx–xxx

www.elsevier.com/locate/jml

Memory andLanguage

ROOF

Parsing with focus particles in context: Eye movementsduring the processing of relative clause ambiguities q

Ruth Filik a,*, Kevin B. Paterson b, Simon P. Liversedge c

a Department of Psychology, University of Glasgow, UKb School of Psychology, University of Leicester, UK

c Department of Psychology, University of Durham, UK

Received 16 March 2004; revision received 21 July 2005

ECTEDP

Abstract

Sedivy (2002) proposed that using only and prior referential context to specify contrastive focus can guide the pars-ing of relative clause ambiguities. We report two studies investigating this hypothesis, using sentences that either tem-porarily allowed or disallowed a transitive main clause interpretation. Sentence-completions demonstrated that onlyand interrogative contexts jointly influenced the frequency of relative clause completions to ambiguous fragments.Eye tracking demonstrated that conjoint effects of only and context influenced initial parsing decisions only whenthe active transitive analysis was unavailable. The results are consistent with previous observations that the influenceof contrastive focus on sentence processing depends on which syntactic analyses are available to the parser (Liversedge,Paterson, & Clayes, 2002; Paterson, Liversedge, & Underwood, 1999).� 2005 Published by Elsevier Inc.

Keywords: Focus particles; Contrastive focus; Referential context; Syntactic ambiguity resolution

29303132333435

ORR

There is a long-standing debate concerning whetherprocesses involved in establishing the reference of a lin-guistic expression can influence decisions about whichanalysis to assign to a syntactic ambiguity (e.g., Alt-mann & Steedman, 1988; Altmann, Garnham, & Den-nis, 1992; Altmann, Garnham, & Henstra, 1994;Altmann, van Nice, Garnham, & Henstra, 1998; Binder,

UNC

363

383940414243

0749-596X/$ - see front matter � 2005 Published by Elsevier Inc.

doi:10.1016/j.jml.2005.07.004

q RF conducted Experiment 1 as part-fulfilment of a PhD inPsychology at the University of Derby and is grateful to theExperimental Psychology Society for a study grant to visit theUniversity of Durham. We are grateful to Roger Van Gompelfor helpful discussions, and to Gerry Altmann and twoanonymous reviewers for useful comments.* Corresponding author. Fax: +44 0 141 330 4606.E-mail address: [email protected] (R. Filik).

Duffy, & Rayner, 2001; Britt, 1994; Britt, Perfetti, Gar-rod, & Rayner, 1992; Ferriera & Clifton, 1986; Frazier &Clifton, 1996; Murray & Liversedge, 1994; Sedivy, 2002;Spivey-Knowlton & Tanenhaus, 1994; Spivey-Knowl-ton, Trueswell, & Tanenhaus, 1993). Much researchinforming this debate has involved the study of sentenc-es that contain temporary syntactic ambiguities (e.g., 1).

1. The businessmen loaned money at low interestwere told to record their expenses.

In (1), the phrase loaned money at low interest is tem-porarily ambiguous between a main clause analysis spec-ifying what the businessmen did and a reduced relativeclause analysis specifying which businessmen are underconsideration. It is syntactically disambiguated at were

mailto:[email protected]

T

4445

447

4849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697

9899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146

147148149

150152

153

2 R. Filik et al. / Journal of Memory and Language xxx (2005) xxx–xxx



UNCORREC

told as a reduced relative clause sentence. Sentence (2) ismade unambiguous by including who were.

6 2. The businessmen who were loaned money at lowinterest were told to record their expenses.

Garden path theory (e.g., Frazier & Rayner, 1982) re-quires that the parser computes syntactic analyses serial-ly, with initial parsing decisions made solely on the basisof knowledge about permissible grammatical structures.Readers preferentially adopt a main clause analysis ofthe ambiguity in sentences like (1) because it is structur-ally less complex than a relative clause analysis. Conse-quently, they should experience difficulty when sentenceslike (1) are disambiguated as reduced relatives. After theparser has made an initial syntactic choice, a thematicprocessor uses a broad range of information to assessthe quality of the resulting interpretation (e.g., Rayner,Carlson, & Frazier, 1983), and potentially to guide itsreanalysis.

An alternative approach (Crain & Steedman, 1985;also Altmann & Steedman, 1988) requires that readersadopt whichever analysis produces the least referentiallycomplex semantic representation. Crain and Steedmanargued that part of the process of interpreting text in-volves the reader constructing a mental representationof its meaning (i.e., a discourse model) and incremental-ly integrating each phrase with the existing discoursemodel. On encountering an ambiguity, each alternativesyntactic analysis is made available to a semantic proces-sor, which preferentially selects an analysis that refers toexisting discourse entities or accommodates the fewestreferential presuppositions about entities that are notmade explicit in the text.

Sentence (1) begins with a definite expression (i.e., thebusinessmen) that presupposes a unique referent in theprior discourse context (e.g., Heim, 1982). Duringsemantic integration the processor attempts to establishan anaphoric link between this expression and its refer-ent. When a single potential antecedent exists in priordiscourse context then readers can readily establish a un-ique anaphoric link. However, if the context containstwo or more potential antecedents then readers will re-quire modifying information, such as that supplied bya relative clause, to uniquely identify the referent. Crainand Steedman argued that readers are predisposed toadopt the relative clause analysis of an ambiguity whenit disambiguates the referent of an expression, therebyeliminating reading difficulty for ambiguous sentencessuch as (1). In the absence of preceding referential con-text, or when context does not include a previouslyestablished referent, readers may accommodate referen-tial presuppositions by instantiating a unique referent inthe discourse model. As modifying information is not re-quired to disambiguate the referent, readers will adopt

EDPROOF

the main clause analysis of a relative clause ambiguityand experience difficulty if it subsequently is disambig-uated as a relative clause.

Some evidence suggests that prior referential contextcan affect the processing of syntactic ambiguities (e.g.,Altmann & Steedman, 1988; Altmann et al., 1992; Alt-mann et al., 1994; Altmann et al., 1998; Sedivy, 2002;Spivey-Knowlton & Tanenhaus, 1994; Spivey-Knowltonet al., 1993). However, other evidence suggests that ref-erential factors have a limited role in parsing (Britt,1994; Britt et al., 1992; Clifton & Ferreira, 1989; Ferriera& Clifton, 1986; Murray & Liversedge, 1994).

Although Crain and Steedman (1985) originally pre-sented their account as the Referential theory of sen-tence processing, several subsequent theories includereferential factors as one of several possible extra-syn-tactic influences on parsing, most notably Constraint-satisfaction theories (e.g., McRae, Spivey-Knowlton, &Tanenhaus, 1998; MacDonald, 1994; MacDonald,Pearlmutter, & Seidenberg, 1994; Tanenhaus & True-swell, 1995). Constraint-satisfaction requires that thealternative analyses of a syntactic ambiguity are con-structed in parallel and compete for adoption. Multiplefactors, including verb sub-categorization preferences,semantic plausibility, and discourse context, imposeconstraints that favor the alternative analyses, and pro-cessing difficulty occurs when a constituent is encoun-tered that conflicts with the currently most favoredanalysis. Constraint-satisfaction predicts an effect of ref-erential context on parsing, with the magnitude of thiseffect being dependent on the relative strength of thisconstraint in relation to other constraints favoring thealternative analyses. Its advocates (e.g., Spivey & Tanen-haus, 1998; Spivey-Knowlton & Sedivy, 1995) often useoff-line data, such as analyses of corpora or sentencecompletions, to quantify the relative strength of con-straints favoring alternative syntactic analyses at a par-ticular point in the sentence.

With the present studies we were concerned with anextension of the Referential theory by Ni, Crain, andShankweiler (1996), which requires that the referentialpresuppositions of focus particles such as only can deter-mine how a syntactic ambiguity initially is analyzed. Fo-cus particles are used to indicate contrastive focus, andaccording to Rooth (1992), only presupposes a contrastbetween a focus set of entities that is made explicit in thesentence and a set of alternatives. Ni et al. argued thaton encountering a phrase that includes only, e.g., onlybusinessmen in (3), the reader immediately attempts toincorporate its presuppositions by determining which fo-cus and contrast sets are to be included in the discoursemodel.

3. Only businessmen loaned money at low interestwere told to record their expenses.

T

154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202

203204205206207208209210211212213214215

216217218219220221222223224225226227228229230231232233234235236

237

238

239240241242243244245246247248249250251

R. Filik et al. / Journal of Memory and Language xxx (2005) xxx–xxx 3



NCORREC

In principle, readers may contrast a focus set denotedby the head noun with another set of entities (e.g., busi-nessmen with lawyers) or, alternatively, they might con-trast subsets of the head noun and anticipate modifyinginformation, such as that supplied by a relative clause,that disambiguates the focus set. Ni et al. argued thatreaders preferentially contrast subsets of the head nounas this avoids referential presuppositions about entitiesthat are not made explicit in the text. Thus, on encoun-tering only businessmen in (3), readers will contrast twosets of businessmen and anticipate modifying informa-tion that disambiguates the focus set, thereby predispos-ing readers to adopt a relative clause analysis of theambiguity.

In an eye-tracking experiment Ni et al. found thatonly eliminated reading difficulty for the disambiguatingphrase (e.g., were told in 3) of ambiguous sentences.However, including a pronominal adjective (e.g., onlywealthy businessmen) reintroduced reading difficulty, asin this case the need for modifying information was sat-isfied by the adjective, and the readers no longer antici-pated modifying information in the form of a relativeclause. A self-paced reading study by Sedivy (2002,Experiment 1), comparing definite versions of sentenceswith and without only (e.g., only the businessmen com-pared with the businessmen), also showed that contras-tive focus can eliminate reading difficulty for syntacticambiguities. However, Clifton, Bock, and Rado (2000)found no effects of only on the parsing of relative clauseambiguities, and Paterson et al. (1999) argued that onlymay affect sentence processing without impinging on ini-tial parsing decisions.

Paterson et al. (1999) noted that Ni et al. (1996) hadused a large disambiguating region for many items, of-ten comprising two words (e.g., were told in 3), and ar-gued that reading times at this region might includefixations made during syntactic reanalysis as well as dur-ing initial parsing.1 Consequently, the effects they ob-tained might have been due to only aiding recoveryfrom an initial misanalysis rather than guiding parsingdecisions. To address this possibility, Paterson et al. con-ducted an eye-tracking study using materials that weresyntactically disambiguated at a single word. They usedsentences such as Only teenagers allowed a party invited a

juggler straightaway, that temporarily permitted an ac-tive transitive analysis with the second noun-phrasebeing analyzed as a direct object (e.g., Only teenagers al-

lowed a party in the evening.). Previous research had sug-gested that readers are predisposed to adopt the active
U 252
253254255256257258

1 Sedivy�s (2002) Experiment 1 employed Ni et al.�s (1996)materials, using the same disambiguating region. Therefore,Paterson et al.�s (1999) arguments concerning the use of largedisambiguating regions would apply to Sedivy�s replication ofthe study too.

EDPROOF

transitive analysis of a syntactic ambiguity whenever itis available (MacDonald, 1994; Townsend & Bever,2001), and that the preference cannot be over-riddenby prior referential context (Britt et al., 1992; Ferriera& Clifton, 1986; Murray & Liversedge, 1994).

Paterson et al. (1999) compared ambiguous sentenceswith unambiguous counterparts that were created byadding who were after the head noun. There was clearevidence that readers experienced difficulty due to ini-tially misanalyzing the ambiguity, with longer first passreading times at the critical verb (e.g., invited) of ambig-uous items, and no modulating effect of only. However,only did affect other aspects of sentence processing.More time was spent reprocessing text following the dis-ambiguation of sentences without only, which Patersonet al. interpreted as evidence that only had facilitatedthe reanalysis of ambiguous items when they were dis-ambiguated as reduced relative clause constructions.Furthermore, first pass reading times were longer atthe critical verb of both ambiguous and unambiguoussentences without only. Paterson et al. took this to dem-onstrate that using only to create an expectation formodifying information had affected the semantic inte-gration of relative clause information. As the criticalverb marked the completion of the relative clause ofunambiguous sentences and disambiguated the ambigu-ous ones, it may have provided the first opportunity forreaders to integrate the relative clause with the discoursemodel. It appeared that relative clause information wasmore easily integrated when it satisfied the need formodifying information that only created.

A follow-up eye-tracking study by Liversedge et al.(2002) used sentences such as Only motorists stopped in

the car park received a warning about their outdated per-

mits. These included a prepositional phrase (e.g., in the

car park) that ruled out the possibility of an active tran-sitive analysis, and were temporarily ambiguous betweenan intransitive analysis and a reduced relative clauseanalysis. With the most strongly preferred analysis ruledout, readers must select between two relatively dispre-ferred analyses. Under these circumstances syntacticprocessing might be susceptible to extra-syntactic fac-tors, such as the processing of contrastive focus. Livers-edge et al. found no effects in first pass reading times butreaders made fewer regressive saccades and spent lesstime re-inspecting text following syntactic disambigua-tion for sentences with than without only, which theytook as evidence for an effect of only on initial parsingdecisions. Readers also took longer to complete readingthe remainder of the ambiguous sentences irrespective ofwhether they included only. Thus, although there was aneffect of only at disambiguation, it did not fully eliminatereading difficulty for ambiguous sentences.

One possibility, suggested by Sedivy (2002), is thatthe influence of only is modulated by the availability ofcontrast sets in the prior referential context. Sedivy

T

259260261262263264265266267268269270271272273274275276277278279280281282

28284285286287

288289290291292293294295296297298299300301302303304305306307308309310311312313314

315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361

362

363364365366367




UNCORREC

proposed that when an explicit contrast is made avail-able in the prior context then only causes the reader toestablish a contrast between this set and one denotedby the noun it modifies. As the need to establish a con-trast has been satisfied, the reader will not anticipate fur-ther modifying information. However, when priorcontext does not include an explicit contrast then only

will cause the reader to contrast subsets of the headnoun and anticipate modifying information that speci-fies how the subsets differ. Under these circumstances,only and prior referential context might over-ride pars-ing preferences, causing readers to adopting the relativeclause analysis of an ambiguity.

Sedivy conducted four phrase-by-phrase self-pacedreading experiments. With her first experiment, Sedivyreplicated Ni et al.�s (1996) findings, using their materi-als. In Experiment 2, Sedivy investigated the interactiveinfluence of only and prior referential context on pars-ing. She used materials with one- or two-referent con-texts (which she referred to respectively as ‘‘nocontrast’’ and ‘‘contrast’’ contexts) and temporarilyambiguous sentences that were syntactically disambig-uated as reduced relative or main clause constructions(e.g., 4).

3 4. All the secretaries (and accountants) were made totake a tough computing course. Only secretariesprepared for the exam (passed) and earned payraises for their work.

The ambiguous sentences always included only andrelative clause versions differed from main clause coun-terparts by including a disambiguating verb (e.g.,passed). One-referent contexts introduced a single setof entities (e.g., secretaries) that was repeated as thehead noun of the ambiguous sentence, and two-referentcontexts introduced two sets (e.g., secretaries andaccountants) with one repeated as the head noun ofthe ambiguous sentence. Sedivy obtained an interactionof context and ambiguity. She attributed this effect tothe disambiguating phrase of ambiguous main clauseitems (e.g., and earned) being read more quickly intwo- than one-referent contexts and, conversely, the dis-ambiguating phrase of ambiguous reduced relatives(e.g., passed) being read more quickly in one- thantwo-referent contexts; although, importantly, neithercontrast was statistically reliable. No such effects werefound for sentences without only in Experiment 3. ThusSedivy took the results of Experiments 2 and 3 to dem-onstrate that only and prior referential context jointlycreated expectations for modifying information that dic-tated how the ambiguity was to be analyzed.

Sedivy�s results are important because they demon-strate the context dependency of only, and show howusing a combination of only and prior referential contextcan affect the processing of structurally ambiguous sen-

EDPROOF

tences. However, although the results clearly demon-strated that only and prior referential context canaffect sentence processing, they did not unambiguouslydemonstrate that these factors determined how theambiguity initially was analyzed. As noted above, Pater-son et al. (1999) also demonstrated that the need to sat-isfy referential presuppositions can facilitate theprocessing of relative clause information in sentenceswith only. However, by comparing the processing ofambiguous and unambiguous items, Paterson et al.showed that facilitation occurred independently of ini-tial parsing decisions. In her second experiment, Sedivyexamined the influence of only and prior context on theprocessing of sentences containing relative clause ambi-guities but without examining their effect on unambigu-ous counterparts. Thus, it is not possible to determinewhether Sedivy�s effects occurred during the construc-tion of an initial syntactic analysis or if they occurredlater in processing, and were therefore similar in natureto the semantic integration effects that Paterson et al.obtained.

The present studies used eye tracking to examine theinfluence of only and prior context on the parsing of rel-ative clause ambiguities. We did not compare sentenceswith only to sentences without only, as this differencewas investigated in previous work (e.g., Liversedgeet al., 2002; Paterson et al., 1999). Instead, the currentstudies were designed primarily to investigate how only

interacts with context. In conducting these studies, weused interrogatives (e.g., which, who, whether) to set updifferent expectations for information provided in a rel-ative clause sentence (e.g., Frazier & Clifton, 2002;Pesetsky, 1987). Several previous studies have madeeffective use of wh-words and indirect questions to gen-erate contextual expectations (e.g., Altmann et al.,1998; Hanna, Spivey-Knowlton, & Tanenhaus, 1996;Liversedge, Pickering, Branigan, & Van Gompel, 1998;Liversedge, Pickering, Clayes, & Branigan, 2003). InExperiment 1, we investigated whether only and an inter-rogative context could eliminate reading difficulty forrelative clause ambiguities that temporarily permit anactive transitive analysis. In Experiment 2, we examinedwhether these factors affected the processing of relativeclause ambiguities that cannot be assigned an activetransitive analysis, but which are temporarily ambiguousbetween a relative clause analysis and an intransitiveanalysis.

Experiment 1

Experiment 1 investigated the influence of only andinterrogative context on the parsing of relative clauseambiguities that temporarily permitted an active transi-tive analysis. Paterson et al. (1999) had previouslyshown that the initial parsing of the syntactic ambiguity

T

368369370371372

37374375

376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423

424425426427428429430431432433434435436437438439440441442443

444445446447

448449450451452453454455456457458

459

460

461462463

464

465466467468469470471472

473474475476




UNCORREC

in this form of construction was impervious to the influ-ence of only. Participants read sentence pairs comprisinga context sentence followed by a structurally ambiguousor unambiguous relative clause sentence (e.g., 5).

3 5. Tom wondered [who/which builders] fitted a kitch-en. Only builders (who were) paid a deposit fitted akitchen within the week.

The ambiguous sentences began with only and had aninitial NP V NP structure (e.g., Only builders paid a

deposit) that temporarily permitted an active transitiveanalysis with the second noun-phrase analyzed as a di-rect object. The unambiguous sentences were identicalto the ambiguous ones, except for the inclusion of a rel-ative pronoun and auxiliary verb (e.g., who were), whichdisambiguated the sentence structure. The interrogativeswhich and who were used to set up different expectationsfor information provided in the relative clause sentence.Which-phrases indicated that a contrast was to be madebetween two subsets of a noun (e.g., builders). The sub-divided noun was repeated as the head of the followingsentence, and the relative clause analysis of the ambigu-ity provided modifying information that specified howthe sets differed. Who-phrase contexts queried whichset of entities performed a particular action, with the an-swer provided by the head noun of the relative clausesentences. Thus, whereas which-contexts supported only

in specifying a contrast between subsets of the headnoun, who-contexts did not.

The contextual manipulation used in the presentexperiment (and in Experiment 2) was not of the typethat often is used when testing claims made by the Ref-erential theory (e.g., Altmann & Steedman, 1988). Inthese experiments, the propositional content of the rela-tive clause in ambiguous sentences satisfies referentialpresuppositions by uniquely specifying a referent in theprior discourse context. For example, given a contextthat introduces two businessmen, one who had receiveda loan and one who had not, a relative clause analysis ofthe businessmen loaned money at low interest will unique-ly identify one of them. However, both Ni et al.�s (1996)extension of the theory, and Sedivy�s (2002) investiga-tion of context effects, require that when only subdividesa noun into two sets of entities, then the reader antici-pates further information that specifies how the sets dif-fer, and on encountering modifying information, such asa relative clause, accommodates it as part of their dis-course model. The contextual manipulation used in thepresent experiments directly tests this latter claim, andin this respect is comparable to the contextual manipula-tions employed by Sedivy.

We made the following predictions. According toGarden path theory, readers initially should adopt themain clause analysis of an ambiguous reduced relativeclause sentence and we would therefore expect reading

EDPROOF

times at syntactic disambiguation to be longer forambiguous than unambiguous sentences with only, withno modulating influence of context. In line with this,Paterson et al. (1999) also predicted that reading timesat syntactic disambiguation would be longer for ambig-uous than unambiguous sentences, despite the presenceof only and irrespective of prior referential context, sincethe processing of ambiguous reduced relatives that tem-porarily permit an active transitive analysis is impervi-ous to referential manipulations. Paterson et al. arguedthat rather than guiding initial parsing decisions, theprocessing demands associated with the computationof contrastive focus instead might facilitate the semanticintegration of modifying information with the discoursemodel, in which case the relative clause sentences mightbe most easily integrated in which-contexts, that poten-tially will facilitate the reanalysis of the ambiguity.

Sedivy (2002) required that only and prior referentialcontext jointly influence initial parsing. Given that ourwhich-contexts predispose readers to anticipate modify-ing information, then they were similar to Sedivy�s ‘‘nocontrast’’ contexts. Thus, only and which should elimi-nate reading difficulty for ambiguous items. As ourwho-contexts do not create an expectation for modifyinginformation, then it seems reasonable to assume that Se-divy would predict that only and who would have a lessereffect on initial parsing decisions. Similarly, Constraint-satisfaction accounts predict an effect of referentialcontext on parsing, with the magnitude of this effectdependent on the relative strength of constraints favor-ing alternative analyses. Before conducting the eye-tracking experiment, we conducted a sentence comple-tion study to quantify the strength of these constraints,and to generate predictions on behalf of the Con-straint-satisfaction theory.

Method

Participants

Twenty-eight native English speakers with normal orcorrected vision from the University of Derbyparticipated.

Materials and design

We used the 36 target sentence stimuli listed in theAppendix of Paterson et al. (1999). Sentences were tem-porarily ambiguous reduced relatives or unambiguouscounterparts. The relative clause sentences followed acontext sentence that contained a who- or which-phrase.Thus, there were two independent variables: sentenceswere ambiguous or unambiguous, and appeared inwho- or which-contexts.

Items were divided into four lists, each including oneversion of each item, with equal numbers of items ineach of the four conditions. Each list also contained 16filler items, and 28 items from an unrelated experiment

T

477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510

511512513514515516517518519520521522523524525526527528529530531

532533534535536537538539

540

541

542543544

545

549

550551552553554555556557558559

560

561562563564565566567568569570571572573574575576577578579580581582583584585




UNCORREC

on pronoun resolution. The context and relative clausesentences appeared on the screen at the same time.Materials were double spaced across three lines oftext, with the critical region falling at the center of linetwo. A full set of experimental items is included inAppendix A.

Before conducting the eye-tracking study we collect-ed sentence completions from 40 participants from theUniversity of Derby (who did not participate in theeye-tracking study) for reduced relative clause materialsthat were truncated after the ambiguity and began withonly, e.g., Only builders paid a deposit. . . Items were pre-sented in which- or who-contexts. Thirty-six filler itemswere included with the 36 experimental items. Signifi-cantly more relative clause completions were producedfor sentence fragments in which- than who-contexts(46.1% vs. 32.3%), (t1(39) = 5.1, p < .001; t2(35) = 8.6,p < .001). Thus, the results indicated that only and which

created a significantly greater expectation for modifyinginformation than only and who, thereby promoting theproduction of relative clause completions. However,their combined influence was not sufficient to create anoverall preference for the relative clause completions(with less than 50% relative clause completions).

On the basis of these results we generated the follow-ing Constraint-satisfaction predictions. Neither contextcreated an overall preference for the relative clause anal-ysis; therefore, we would not expect only and an inter-rogative context to eliminate reading difficulty at theambiguity. Nevertheless, only and which created a stron-ger relative clause bias than did only and who, suggestingthat readers should experience less difficulty at syntacticdisambiguation when the sentences are in a which- ratherthan a who-context.

Procedure

A Fourward Technologies Dual Purkinje Generation6 eye-tracker monitored gaze location and participants�right eye movement during reading. The eye-trackerhas an angular resolution of 10 min of arc. A PC dis-played materials on a VDU 60 cm from participants�eyes. Tracker output was sampled to produce a sequenceof fixations recorded as x and y character positions, withstart and finish times. Before the start of the experiment,the eye-tracking procedure was explained and partici-pants were instructed to read normally and for compre-hension. Participants were seated at the eye-tracker andplaced on a bite-bar to minimize head movements. Par-ticipants then completed a calibration procedure.

Before the start of each trial, a fixation box appearedin the upper left half of the screen. Once participants fix-ated this box the experimenter prompted the computerto present a target text, with the first character of thetext replacing the fixation box. The experimenter re-ceived feedback on the estimated position of partici-pant�s fixation point. If this did not match with the

EDPROOF

fixation box then the experimenter re-calibrated theeye-tracker. Participants took breaks as required. Oncea participant completed reading each sentence, theypressed a key, and the computer displayed a comprehen-sion question such as Did some builders fit a bathroom?

Half of the questions had �yes� and half had �no� answers.Participants responded by pressing a key, with feedbackon their responses.

Results

Regions

Items were divided into scoring regions, as indicatedby vertical lines in (6).

6. Tom wondered who (which builders) fitted a kitch-en. region1ŒOnly builders (who were) region2Œ paid a

region3Œ deposit region4Œ fitted region5Œ a kitchen region6Œwithin the week. region7

Region 1 was the context sentence. Region 2 was thefirst noun-phrase (e.g., Only builders) for ambiguoussentences and the first noun, relative pronoun and aux-iliary for unambiguous sentences. Region 3 was the firstverb and the indefinite article. Region 4 was the follow-ing noun. Region 5 was the critical region, containingthe main verb, which disambiguated ambiguous sentenc-es. Region 6 was the following noun-phrase, and Region7 was the remainder of the sentence.

Analysis

An automatic procedure pooled short contiguous fix-ations. Fixations less than 80 ms were incorporated intolarger adjacent fixations within one character and fixa-tions of less than 40 ms that were not within three char-acters of another fixation were deleted. Fixations over1200 ms were truncated. Prior to analyzing the data weeliminated trials where participants failed to read thesentence or there had been tracker loss. Specifically,we removed those trials where two or more adjacent re-gions had zero first pass reading times, which accountedfor 3.6% of the data. Participants correctly answered90% of comprehension questions, with no significant dif-ference across conditions (F < 1).

We computed several standard reading time mea-sures (Liversedge & Findlay, 2000; Rayner, 1998; Ray-ner, Sereno, Morris, Schmauder, & Clifton, 1989).First pass reading time is the sum of all the fixationsmade in a region until the point of fixation exits the re-gion either to the left or to the right and provided a mea-sure of initial processing. Total reading time sums theduration of all fixations made within a region, and pro-vided a measure of overall comprehension difficulty atthis region. In addition to these standard measures, weexamined the regression path reading times for the crit-ical verb and for the following region of text. Regression

T

ROOF

586587588589590591592593594595596597598599600601602603

604605606607608609610611612613614615616617618619

620621622623624625626627628629630631632

633

634635636637638639

Table 1Mean first-pass and total reading times for Regions 2–6, and mean regression path reading times for Regions 5 and 6 of ambiguous andambiguous sentences with only in which- and who-contexts for Experiment 1, with standard errors in parentheses

Regions Measure (ms) Context

Which Who

Ambiguous Unambiguous Ambiguous Unambiguous

2 First-pass time 467 (29.0) 760 (39.1) 510 (27.4) 855 (45.4)Total time 731 (48.9) 1095 (75.8) 843 (56.5) 1341 (106.3)



5 First-pass time 243 (12.9) 224 (12.5) 255 (15.2) 232 (15.8)Regression path 380 (47.6) 269 (17.4) 371 (23.0) 309 (39.6)Total time 383 (24.1) 313 (20.0) 416 (26.2) 363 (27.9)

6 First-pass time 252 (20.5) 228 (11.7) 258 (17.6) 266 (17.6)Regression path 418 (33.7) 361 (41.1) 547 (58.4) 452 (41.5)Total time 353 (29.9) 294 (23.0) 353 (30.1) 342 (24.3)

2 It could be argued that the first verb (paid) is the first pointin the sentence at which it can be determined that a relativeclause analysis of the ambiguity does not uniquely refer to anentity in the discourse context, causing readers to abandon aninitial relative clause analysis in favor of the main clauseanalysis. Thus, early reanalysis of an initial relative clauseanalysis might account for the processing cost at this verb.However, it is not obvious how this explanation would accountfor the same effect in the Paterson et al. (1999) study, which didnot include a discourse context. Therefore, it seems more likelythat this is an ambiguity effect, with readers incurring anadditional processing cost when they are required to selectbetween possible analyses of an ambiguity.




UNCORREC

path reading time sums the duration of temporally con-tiguous fixations made on entering a region until a sac-cade transgresses its right boundary (Konieczny,Hemforth, Scheepers, & Strube, 1997; Liversedge, Pater-son, & Pickering, 1998; Rayner & Duffy, 1986). Thismeasure includes fixations made to re-inspect earlierportions of text and is usually interpreted as providingan indication of early processing difficulty along withtime spent re-inspecting the sentence in order to recoverfrom such difficulty. Data for each region were subjectedto two 2(context) · 2(ambiguity) ANOVAs, treatingparticipants (F1) and items (F2) as random variables.We also calculated min F 0 by combining results of theANOVAs based on participants and items means(Clark, 1973; Raaijmakers, Schrijnemakers, & Grem-men, 1999). Table 1 shows mean first pass, regressionpath, and total reading times for Regions 2–6, andTable 2 shows inferential statistics.

First-pass reading time

See Fig. 1 for a graphical representation of first-passreading time effects, with error bars showing 95% confi-dence intervals (Masson & Loftus, 2003). Region 2 read-ing times were longer in who- than which-contexts (683vs. 614 ms), probably due to repetition priming of thehead noun in which-contexts. Longer reading times forunambiguous than ambiguous sentences (808 vs.489 ms), were due to these sentences containing addi-tional words. The interaction was not reliable. Region3 reading times were longer for ambiguous than unam-biguous sentences (318 vs. 256 ms). Paterson et al.(1999) had attributed the same effect to increased pro-cessing load when the verb was ambiguous than whenit was syntactically disambiguated by the preceding text.Advocates of Constraint-satisfaction theory might

EDP

interpret the effect as evidence for competition duringthe initial processing of the ambiguity.2 There were noother significant effects at Regions 3 or 4.

At Region 5, the critical region, there was no effect ofcontext (234 vs. 244 ms), but reading times were longerfor ambiguous than unambiguous sentences (249 vs.228 ms). Importantly, context and ambiguity did notinteract. Thus, reading difficulty was experienced at syn-tactic disambiguation regardless of prior context. At Re-gion 6, the effect of context was not reliable, withnumerically longer reading times in who- than which-contexts (262 vs. 240 ms), and no other significant ef-fects. There were no effects at Region 7.

Regression path reading times

See Fig. 2 for a graphical representation of regres-sion path effects. At Region 5, the critical region,ambiguous sentences had longer reading times thanunambiguous counterparts (378 vs. 313 ms). Therewere no other significant effects, indicating that contextdid not modulate the difficulty in processing the rela-

UNCORRECTEDPROOF

Table 2Statistical values for analyses of reading time data for Experiment 1

Source of variance F1 MSe F2 Min F 0

df F1 value df F2 value df Min F 0 value


Region 2Context 1,27 10.52** 12,479 1,35 12.00** 1,60 5.61**

Ambiguity 1,27 224.13*** 12,686 1,35 211.32*** 1,61 108.70***

Context · Ambiguity 1,27 3.64* 5,157 1,35 1.16 1,54 .88

Region 3Context 1,27 .03 3,052 1,35 .11 1,41 .02Ambiguity 1,27 14.62** 7,435 1,35 23.54*** 1,55 9.02**

Context · Ambiguity 1,27 .06 2,736 1,35 .08 1,58 .03

Region 4Context 1,27 .19 1,641 1,35 .01 1,39 .01Ambiguity 1,27 .02 3,219 1,35 .03 1,56 .01Context · Ambiguity 1,27 .92 1,491 1,35 .72 1,62 .40

Region 5Context 1,27 .85 3,448 1,35 .67 1,62 .37Ambiguity 1,27 4.56* 2,663 1,35 6.71* 1,56 2.71Context · Ambiguity 1,27 .07 2,307 1,35 .06 1,62 .03

Region 6Context 1,27 6.18* 2,153 1,35 3.28+ 1,60 2.14Ambiguity 1,27 .90 2,380 1,35 .40 1,58 .28Context · Ambiguity 1,27 1.97 3,616 1,35 1.15 1,61 .73


Regression path

Region 5Context 1,27 .49 14,331 1,35 .34 1,62 .20Ambiguity 1,27 16.21*** 12,886 1,35 6.20* 1,56 4.48*

Context · Ambiguity 1,27 .60 29,177 1,35 1.55 1,47 .43

Region 6Context 1,27 8.09** 41,664 1,35 14.91*** 1,52 5.24*

Ambiguity 1,27 7.28* 22,269 1,35 4.19* 1,61 2.66Context · Ambiguity 1,27 .27 37,817 1,35 .53 1,51 .18

Total reading time

Region 2Context 1,27 28.35*** 31,677 1,35 31.28*** 1,60 14.87***

Ambiguity 1,27 85.11*** 61,081 1,35 138.67*** 1,54 52.74***

Context · Ambiguity 1,27 6.30* 19,746 1,35 3.32+ 1,60 2.17

Region 3Context 1,27 .43 12,601 1,35 .77 1,53 .28Ambiguity 1,27 37.25*** 20,803 1,35 53.71*** 1,56 22.00***

Context · Ambiguity 1,27 2.73 6,467 1,35 .69 1,51 .55

Region 4Context 1,27 5.51* 5,003 1,35 2.97+ 1,60 1.93Ambiguity 1,27 6.59* 12 ,159 1,35 10.32** 1,55 4.02*


(continued on next page)




UNCORRECTEDPROOF

640641642643644

645646647

Table 2 (continued)



Region 5Context 1,27 6.24* 7,725 1,35 8.26** 1,58 3.55+

Ambiguity 1,27 25.33*** 4,231 1,35 15.06*** 1,61 9.44**


Region 6Context 1,27 3.38* 4,691 1,35 1.92 1,61 1.22Ambiguity 1,27 6.79* 5,134 1,35 4.15* 1,61 2.58Context · Ambiguity 1,27 2.42 6,395 1,35 1.95 1,62 1.08


* p < .05.** p < .01.*** p < .001.+ .1 > p > .05.

210

220

230

240

250

260

270

280

290

Region 4 (deposit) Region 5 (fitted) Region 6 (a kitchen)

which ambiguous who ambiguous

210

220

230

240

250

260

270

280

290

Region 4 (deposit) Region 5 (fitted) Region 6 (a kitchen)

which unambiguous who unambiguous

Panel 1

Panel 2

Fig. 1. Mean first pass reading times, with 95% within-subjectconfidence intervals based on the ambiguity effect, for Regions4–6 of ambiguous (Panel 1) and unambiguous (Panel 2)sentences used in Experiment 1.

200

250

300

350

400

450

500

550

600

Region 5 (fitted) Region 6 (a kitchen)

which unambiguous who unambiguous

200

250

300

350

400

450

500

550

600

Region 5 (fitted) Region 6 (a kitchen)

which ambiguous who ambiguousPanel 1

Panel 2

Fig. 2. Mean regression path reading times, with 95% within-subject confidence intervals based on the ambiguity effect, forRegions 5 and 6 of ambiguous (Panel 1) and unambiguous(Panel 2) sentences used in Experiment 1.




tive clause ambiguity. At Region 6, the regression pathreading times were longer for ambiguous than unam-biguous sentences (353 vs. 318 ms), and sentences inwho-contexts had longer reading times than those inwhich-contexts (348 vs. 324 ms), with no interaction.

Thus, it appeared that while the relative clause sentenc-es were easier to process in which- than who-contexts,this did not impinge on parsing decisions.

T

ROOF

648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703

704705

706707708709710711712

713

714715716717718719720721722723724725726727728729730731732733734735736737

Table 3The portion of variance explained (R2) in Experiments 1 and 2when the number of relative clause completions was regressedagainst the difference in reading time between ambiguous andunambiguous relative clause sentences in who and whichcontexts (Experiment 1) and whether and which contexts(Experiment 2)

Measure Region Experiment 1 Experiment 2

Which Who Which Whether

First pass 5 (critical) .02 .07 .00 .036 .07 .01 .07 .01

Regression path 5 .05 .02 .04 .026 .00 .05 .22** .07

Total time 5 .05 .22** .10+ .006 .02 .06 .11+ .01

* p < .05.** p < .01.*** p < .001.+.1 > p > .05.




UNCORREC

Total reading time

Region 2 reading times were longer in who-contexts(1092 vs. 913 ms). This effect also was obtained in firstpass times and attributed to repetition priming of thehead noun in which-contexts. Reading times also werelonger for unambiguous sentences (1218 vs. 787 ms),due to these sentences containing additional words.The interaction was unreliable.

The pattern of results for Regions 3-6 were similar,with longer reading times for ambiguous than unambig-uous sentences, and no modulating effect of context. AtRegion 3 there was an effect of ambiguity (623vs.456 ms), but no effect of context (533 vs. 546 ms)and no significant interaction. At Region 4, there wasan effect of ambiguity (399 vs. 345 ms), but the effectof context was unreliable (356 vs. 388 ms), and therewas no interaction. At Region 5, there was an effect ofambiguity (400 vs. 338 ms), and an effect of context, withlonger reading times in who- rather than which-contexts(390 vs. 348 ms), but no interaction. Finally, at Region 6there was an effect of ambiguity (353 vs. 318 ms), but theeffect of context was not significant (324 vs. 348 ms), andthe interaction was not significant.

Total reading time effects at each of these regions indi-cated that readers experienced difficulty as a consequenceof the syntactic ambiguity, with no interactions to indi-cate a modulating effect of context. However, contextdid have amain effect at the critical verb, without imping-ing on initial parsing decisions, reading times being short-er for both ambiguous and unambiguous sentences inwhich- than who-contexts. The critical verb disambigua-ted the relative clause in ambiguous sentences, andmarked the completion of the relative clause in unambig-uous counterparts. Therefore, this verb provided the firstopportunity for readers to integrate information provid-ed by the relative clause as part of their current discoursemodel. The reading time effects suggested that it was eas-ier to process the relative clause information when it wasanticipated in which-contexts but not in who-contexts.There were no effects at Region 7.

To further test the predictions of aConstraint-satisfac-tion account we used linear regression to determinewhether the frequency of relative clause completions toambiguous sentence fragments in an off-line task predict-ed the on-line reading cost for these sentences followingsyntactic disambiguation (see Table 3). The independentvariables were the number of relative clause completionsfor sentences beginning with only in either a which- or awho-context. The dependent variableswere the differencesin reading time between ambiguous and unambiguoussentences (inwhich- andwho-contexts) in first pass, regres-sionpath, and total timemeasures forRegions 5 and6.Noeffectswere obtained for either first pass or regressionpathreading times at Regions 5 or 6 (R2s < .06, Fs < 2). How-ever, there was a significant negative relationship betweenthe frequency of relative clause completions and total

EDPreading times for sentences in who-contexts (R2 = .22,

F = 9.32, p < .01), indicating that there was less process-ing difficulty for sentences with high than low rates of rel-ative clause completions. There were no other significanteffects (R2s < .08, Fs < 2.7). Thus, it appeared that theoff-line completions did not predict the cost of syntacticdisambiguation in measures of early sentence processing,and only predicted the overall reading difficulty for sen-tences in one of the contextual conditions.

Discussion

We obtained two key findings in the reading timemea-sures. First, participants had difficulty in processingambiguous sentences despite only and regardless of theinterrogative context. There were longer first pass andregressionpath reading times at the critical verb, and long-er total reading times for Regions 4, 5, and 6, for ambigu-ous items. The magnitude of these effects did not differacrosswhich- andwho-contexts. Thus, initial parsing deci-sionswere impervious to the influence of only and context.

The second finding relates to the evidence that ourreferential manipulation had an influence on sentenceprocessing, albeit without modulating initial parsingdecisions. We found that total reading times at the crit-ical verb, and regression path reading times at thefollowing region, were shorter in which- than who-con-texts, and attributed this effect to readers finding it easierto semantically integrate sentential information when itsatisfied contextual expectations. Which-contexts re-quired modifying information that specified the natureof a contrast between two subsets of the head noun,and this information was supplied by the relative clauseof unambiguous items and the relative clause analysis ofambiguous items. By contrast, who-contexts required thespecification of a set of entities, this information being

T

738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787

788

789790

791792793794795796797798799800801802803

804805

806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840




NCORREC

provided by the head noun. Thus, relative clause infor-mation was more easily integrated with context whenit was anticipated in which-contexts than when it wasunanticipated in who-contexts.

Our results were consistent with previous demon-strations that ambiguous reduced relatives that tempo-rarily permit an active transitive analysis areimpervious to referential factors (e.g., Britt, 1994; Brittet al., 1992; Ferriera & Clifton, 1986; Frazier & Clif-ton, 1996; Murray & Liversedge, 1994; Patersonet al., 1999). They also are consistent with Patersonet al.�s finding that processing demands associated withcontrastive focus can facilitate the semantic integrationof relative clause information without impinging on ini-tial parsing decisions. Thus, the present results werecompatible with modular theories of parsing such asthe Garden path theory, which predicts that ambiguousreduced relatives will be harder to process than unam-biguous unreduced relatives, irrespective of the context.In contrast with Sedivy (2002), there was no evidencethat single-referent (or ‘‘no contrast’’) contexts (i.e.,which-contexts) eliminated reading difficulty for ambig-uous reduced relatives with only. The reading time re-sults can be accommodated by Constraint-satisfactiontheory if it attributes the absence of an effect to strongsyntactic preferences for adopting an active transitiveanalysis of the ambiguity being impervious to con-straints imposed by referential factors.

However, the absence of a relationship between off-line sentence completions and on-line parsing decisionsis problematic for advocates of this account who usesentence completion data to quantify the relativestrength of constraints supporting the rival analyses ofan ambiguity at a particular point in a sentence (e.g.,Spivey & Tanenhaus, 1998; Spivey-Knowlton & Sedivy,1995). Our sentence completion data suggested that onlywith a which-phrase would create a stronger bias foradopting the relative clause analysis of the ambiguitythan only and a who-phrase. However, the nature ofthe interrogative context did not modulate reading timesduring on-line ambiguity resolution. Instead, it affectedhow easily the relative clause information supplied byboth ambiguous and unambiguous sentences wassemantically integrated with the discourse context.Importantly, item-by-item regression analyses showedthat offline sentence completion data did not predictreading times. Thus, it appears that off-line sentencecompletions are not predictive of on-line parsing prefer-ences, for the present experiment at least (see also, Liv-ersedge et al., 2002, for similar findings).
U 841
842843844845846847

Experiment 2

Thus far we have demonstrated that only and priorcontext do not influence the initial parsing of ambiguous

EDPROOF

reduced relative clause sentences that temporarily permitan active transitive analysis. This finding is consistentwith other studies that have shown such decisions tobe impervious to referential factors (Britt, 1994; Brittet al., 1992; Clifton & Ferreira, 1989; Ferriera & Clifton,1986; Murray & Liversedge, 1994; Paterson et al., 1999).In Experiment 2 we examined the parsing of relativeclause ambiguities that do not temporarily permit atransitive analysis, but which instead are temporarilyambiguous between a relative clause analysis and anintransitive analysis. We used sentences from Liversedgeet al. (2002), such as Only motorists stopped in the car

park received a warning about their outdated permits.For these sentences, a prepositional phrase (e.g., in the

car park) ruled out the active transitive analysis priorto syntactic disambiguation. Liversedge et al. had ar-gued that with the most strongly preferred analysis ruledout, readers must select between two relatively dispre-ferred analyses, and that under these circumstances syn-tactic processing might be susceptible to extra-syntacticfactors, such as the processing of contrastive focus. Liv-ersedge et al. found that readers experienced short-liveddifficulty at the disambiguating verb when processingambiguous sentences without only, with no such effectfor ambiguous sentences with only. They took this asevidence that using only to create an expectation formodifying information could influence parsing decisionsfor this type of ambiguous sentence. Given Liversedgeet al.�s findings and given that Experiment 1 did not pro-duce effects of only and an interrogative context on ini-tial parsing decisions, then the likelihood of obtainingsuch effects might be greater if we use sentences that ruleout the possibility of assigning a transitive analysis. InExperiment 1, we examined the processing of a relativeclause ambiguity in a which-context and in a who-con-text. The who-context was effectively a null context, sinceit neither supported nor inhibited the relative clauseanalysis. Given the subtle nature of the ambiguity inves-tigated in the present experiment it was possible thatreaders would adopt a relative clause analysis of theambiguity regardless of whether the context supportedthis analysis or provided a null context. To avoid thispossibility, we strengthened the contextual manipulationin Experiment 2 by replacing the null who-context withone that was designed to inhibit the adoption of a rela-tive clause analysis. Sedivy (2002) had used contextscontaining two referents, with one repeated as the headnoun of the ambiguous sentence, to bias readers intoadopting a main clause analysis of an ambiguity, argu-ing that when context supplies an explicit contrast set,readers are not predisposed to adopting the relativeclause analysis. We also used a two-referent context toinhibit a relative clause analysis, but included an inter-rogative (i.e., whether) to emphasize that a contrastwas to be made between the two referents. An examplematerial is shown in (7):

T

84849850851852

853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904

905906907908909910911912913914915916917

918

919

920921922

923

924925926

927928929930931932

933934935936937938939940941942943944945946947948949950951952953

954955956957




UNCORREC

8 7. Simon wondered [which motorists/whether themotorists or the pedestrians] received a warning.Only motorists [who were] stopped in the car parkreceived a warning about their outdated permits.

The relative clause sentences were either temporarilyambiguous or were disambiguated by including a rela-tive pronoun and auxiliary verb (i.e., who were). Thewhich-phrase context indicated that a contrast was tobe made between two subsets of a noun and requiredfurther information about how these sets differed. Thesub-divided noun was repeated as the head noun of arelative clause sentence, and the relative clause suppliedmodifying information that specified how the sets dif-fered. We expected a combination of only and a which-phrase to bias readers into adopting the relative clauseanalysis of the ambiguity. The whether-phrase contextset up an explicit contrast between two sets of entities(e.g., motorists and pedestrians), creating an expectationfor further information that specified which entities per-formed a particular action. The head noun of the rela-tive clause sentences provided this information. Thus,whereas which-contexts created a need for further mod-ifying information, whether-contexts were designed toobviate this need, thereby inhibiting the adoption of arelative clause analysis of the ambiguity.

We made the following predictions. Garden path the-ory required that readers initially would adopt the syn-tactically least complex of the potential analyses,irrespective of the context. Since the intransitive analysisis syntactically less complex than the relative clause anal-ysis, the theory required that readers initially adopt thisanalysis of the ambiguity, experiencing difficulty when itis disambiguated as a reduced relative clause. By con-trast, Constraint-satisfaction theory requires that theanalysis that is assigned to a syntactic ambiguity will de-pend on the relative strength of constraints imposed thusfar. Thus, for only and which-contexts to create a prefer-ence for the relative clause analysis will depend on thestrength of the constraints that these factors imposeand the relative strength of competing constraints thatbecome available prior to disambiguation. As in Exper-iment 1, we collected sentence completion data to testthe off-line effectiveness of our manipulations and togenerate Constraint-satisfaction predictions. In line withthe predictions of a Constraint-satisfaction account,Sedivy (2002) proposed that readers adopt the relativeclause analysis of an ambiguity when prior context doesnot supply an explicit contrast set, and adopt the mainclause analysis of the ambiguity when it does. As ourwhich-contexts were functionally equivalent to Sedivy�sone-referent (‘‘no contrast’’) contexts, and did notsupply a contrast set, they should predispose readersto adopting a relative clause analysis of the ambiguity.By contrast, our whether-contexts were functionallyequivalent to Sedivy�s two-referent (‘‘contrast’’)

EDPROOF

contexts, supplying an explicit contrast set. These con-texts should obviate the need for modifying information,causing readers to initially adopt amain clause analysis ofthe ambiguity. Thus, following Sedivy, we would expectreaders to have difficulty in processing the ambiguoussentences in whether-contexts, but not in which-contexts.Liversedge et al. (2002) proposed that the processing ofcontrastive focus might affect the parsing of sentencesthat are ambiguous between an intransitive analysis anda relative clause analysis, since for these sentences thereis not a strong syntactic preference for adopting one ofthe alternative analyses, which is also in line with the pre-dictions of a constraint satisfaction approach.

Method

Participants

Thirty-two native English speakers with normal orcorrected vision from the University of Leicesterparticipated.

Materials and design

We used the 32 sentence stimuli listed in the Appen-dix of Liversedge et al. (2002). All sentences began withonly, included a prepositional phrase before the criticalverb, and were temporarily ambiguous reduced relativeclause sentences or unambiguous counterparts. Sentenc-es appeared in which- or whether-contexts. For which-contexts, the interrogative modified a noun that wasrepeated as the head of the relative clause sentence.Whether-contexts specified two referents. Thus, therewere two independent variables: relative clause sentenceswere ambiguous or unambiguous; and contexts includeda which - or whether-phrase.

Materials were divided into four lists, each includingone version of each sentence, and with equal numbers ofitems from each of the experimental conditions. Each listhad 48 additional filler items, including 28 that werematerials from an unrelated experiment concerning theprocessing of quantifier scope ambiguities. The contextand relative clause sentences appeared on the screen atthe same time. Materials were double spaced acrossthree lines of text, with the critical region falling at thecenter of line two. A full set of experimental items isincluded in Appendix B.

We collected sentence completions from 32 nativeEnglish speakers from the University of Derby (noneof whom participated in Experiment 1 or in the currenteye-tracking experiment) for 32 filler items and 32 exper-imental items that were truncated after the ambiguity.The relative clause sentences began with only, e.g., Only

motorists stopped in the car park. . . and were presentedin which- or whether-contexts. There was an overall highrate of relative clause completions, probably becauseparticipants were selecting between two relatively equal-ly dispreferred alternative analyses. Significantly more

T

958959960961962963964965966967968969970971972

973974975976977978

979

980981982983

98985986987988

989

990

991992993994995996997998

999

100010011002100310041005100610071008100910101011101210131014101510161017

1018

1019102010211022




C

relative clause completions were produced for sentencesin which- than whether-contexts (79.3% vs. 57.8%),(t1(31) = 5.46, p < .01; t2(31) = 4.83, p < .01). Fromthese off-line results we would predict on behalf of Con-straint-satisfaction theorists that which and only shouldcreate a strong preference to adopt a relative clause anal-ysis of the syntactic ambiguity. It should be sufficientlystrong to eliminate reading difficulty for the ambiguoussentences. Only and whether also created a preference forthe relative clause analysis; however, this preference wassignificantly weaker than that obtained for only andwhich. Thus, while we would also expect a whether-con-text to reduce the reading difficulty for ambiguous sen-tences it should not be as effective at doing so as awhich-context.

Procedure

We employed the same procedure as used in Experi-ment 1, using a Fourward Technologies Dual PurkinjeGeneration 6 eye-tracker located in the Leicester Psy-chology laboratories and operating the same data collec-tion and analysis software as used in Experiment 1.

Results

Regions

Sentences were divided into regions, as indicated byvertical lines in (8).

4 8. Simon wondered [which motorists/whether themotorists or the pedestrians] received a warning

region1Œ Only motorists [who were] region2Œ stoppedregion3Œ in the car park region4Œ received region5Œ awarn-ing region6Œ about their outdated permits region7.

UNCORRE

Table 4Mean first-pass and total reading times for Regions 2–6, and mean regrambiguous sentences with only in which- and whether-contexts for Ex

Regions Measure (ms)

Which

Ambiguous

2 First-pass time 444 (25.7)Total time 589 (35.8)



5 First-pass time 280 (13.6)Regression path 368 (26.6)Total time 409 (23.1)

6 First-pass time 324 (15.6)Regression path 380 (19.0)Total time 451 (20.1)

EDPROOF

Region 1 was the context sentence. Region 2 was thefirst noun-phrase for ambiguous items; and the firstnoun, relative pronoun and auxiliary for unambiguoussentences. Region 3 was the first verb. Region 4 wasthe prepositional phrase. Region 5 was the critical verbthat disambiguated ambiguous items. Region 6 was thefollowing noun or noun-phrase, and Region 7 was theremainder of the sentence.

Analysis

We employed the same automatic procedure as usedin Experiment 1 to pool contiguous fixations, to deleteshort fixations, and to truncate long ones. Prior to ana-lyzing the fixation time data we eliminated trials whereparticipants failed to read the sentence or there had beentracker loss. This involved removing trials where two ormore adjacent regions had zero first pass reading times,accounting for 5.3% of the data. Participants gave cor-rect responses to comprehension questions on 91% oftrials, with no significant differences between conditions(F < 1.3).

We computed the same measures as used in Experi-ment 1. Data for each region were subjected to two2(context) · 2(ambiguity) ANOVAs, treating partici-pants (F1) and items (F2) as random variables. We alsocalculated min F 0. Table 4 shows mean first pass, regres-sion path, and total reading times for Regions 2–6, andTable 5 shows inferential statistics.

First pass reading time

At Region 2, first pass reading times were longer forunambiguous sentences, due to the these sentences hav-ing additional words at this region (725 vs. 441 ms), withno other effects. At Region 3, there was no significant

ession path reading times for Regions 5 and 6 of ambiguous andperiment 2, with standard errors in parentheses

Context

Whether

Unambiguous Ambiguous Unambiguous

722 (35.6) 438 (21.4) 728 (38.6)946 (58.1) 636 (37.6) 911 (46.6)

319 (19.8) 325 (22.2) 294 (21.2)456 (25.9) 468 (30.8) 475 (30.0)

640 (34.1) 663 (33.4) 585 (21.9)954 (46.2) 939 (44.6) 891 (30.0)

276 (8.6) 283 (10.0) 285 (12.5)377 (38.3) 345 (21.1) 330 (20.0)374 (20.1) 384 (18.8) 404 (18.9)

313 (15.3) 332 (15.7) 308 (15.2)385 (30.1) 497 (42.1) 366 (22.1)423 (22.3) 459 (22.8) 455 (25.5)

UNCORRECTEDPROOF

Table 5Statistical values for analyses of reading time data for Experiment 2




Region 2Context 1,31 0.00 9,471 1,31 0.08 1,39 .01Ambiguity 1,31 186.02** 13,909 1,31 181.54*** 1,62 91.88***

Context · Ambiguity 1,31 0.08 12,489 1,31 0.28 1,47 .06

Region 3Context 1,31 1.60 2,624 1,31 1.17 1,61 .68Ambiguity 1,31 2.63 3,642 1,31 2.25 1,62 1.21Context · Ambiguity 1,31 2.06 3,232 1,31 0.85 1,53 .60

Region 4Context 1,31 1.96 13,260 1,31 1.26 1,59 .77Ambiguity 1,31 4.83* 17,954 1,31 8.49** 1,58 3.08+


Region 5Context 1,31 0.52 3,001 1,31 0.22 1,53 .15Ambiguity 1,31 0.11 1,954 1,31 0.49 1,44 .09Context · Ambiguity 1,31 0.18 2,693 1,31 0.32 1,57 .12

Region 6Context 1,31 0.02 2,991 1,31 0.22 1,37 .02Ambiguity 1,31 2.04 4,630 1,31 1.98 1,62 1.00Context · Ambiguity 1,31 0.36 3,542 1,31 0.65 1,57 .23

Regression path


Region 6Context 1,31 3.15+ 24,262 1,31 2.24 1,60 1.31Ambiguity 1,31 5.31* 23,858 1,31 4.50* 1,62 2.44Context · Ambiguity 1,31 6.64* 22,520 1,31 9.14* 1,60 3.85+

Total reading time

Region 2Context 1,31 0.07 18,554 1,31 0.03 1,53 .02Ambiguity 1,31 76.46*** 41,791 1,31 123.84*** 1,59 47.27***



Region 4Context 1,31 3.77+ 25,473 1,31 3.66+ 1,62 1.86Ambiguity 1,31 1.28 40,188 1,31 3.97+ 1,49 .97Context · Ambiguity 1,31 0.08 26,209 1,31 0.07 1,62 .04

Region 5Context 1,31 0.03 5,958 1,31 0.00 1,50 .01Ambiguity 1,31 0.15 12,052 1,31 1.33 1,38 .13Context · Ambiguity 1,31 2.08 11,519 1,31 3.73+ 1,57 1.34

(continued on next page)




TEDPROOF

10231024102510261027102810291030103110321033103410351036

10371038103910401041104210431044104510461047104810491050105110521053

1054105510561057105810591060106110621063

1064106510661067106810691070107110721073107410751076

Table 5 (continued)




* p < .05.** p < .01.*** p < .001.+ .1 > p > .05.

300

350

400

450

500

Region 5 (received) Region 6 (a warning)

which ambiguous whether ambiguous

300

350

400

450

500

Region 5 (received) Region 6 (a warning)

which unambiguous whether unambiguous

Panel 1

Panel 2

Fig. 3. Mean regression path reading times, with 95% within-subject confidence intervals based on the ambiguity effect, forRegions 5 and 6 of ambiguous (Panel 1) and unambiguous(Panel 2) sentences used in Experiment 2.




UNCORREC

main effect of context (321 vs. 310 ms), no significantmain effect of ambiguity (324 vs. 307 ms), and no signif-icant interaction. Region 4 reading times were longer forambiguous than unambiguous sentences (664 vs.612 ms), with no other significant effects, suggesting thatreaders had more difficulty in processing the preposi-tional phrase when it was structurally ambiguous, irre-spective of whether the context supported a relativeclause analysis. At Region 5, the critical region, therewere no significant effects. At Region 6, the main effectof ambiguity was not significant (328 vs. 311 ms), withno other effects. Thus, first pass reading times providedno evidence of readers having difficulty in processing thesentences following syntactic disambiguation.

Regression path reading times

There were no significant effects at Region 5. At Re-gion 6, the post-critical region, the main effect of contextwas not significant (383 vs. 432 ms). However, readingtimes were longer for ambiguous than unambiguous sen-tences (439 vs. 376), and there was a significant interac-tion (see Fig. 3). Reading times were longer forambiguous than unambiguous sentences in whether-con-texts (497 vs. 366 ms), with no such difference in which-contexts (380 vs. 385 ms). The interaction was exactly ofthe form that Sedivy (2002) would have predicted, withreaders having difficulty in processing the ambiguoussentences in two-referent contexts (i.e., whether-con-texts) but not in one-referent contexts (i.e., which-con-texts). Thus, the effect was consistent with thecombined influence of only and the interrogative contextaffecting how the ambiguous sentences were processed.

Total reading time

At Region 2, total reading times were longer forunambiguous than ambiguous sentences (929 vs.613 ms), due to these sentences having more words atthis region, with no other significant effects; and no sig-nificant effects at Region 3. At Region 4, the effect ofcontext approached significance, with numerically long-er reading times in which- contexts (970 vs. 915 ms). Theeffect of ambiguity was not reliable (963 vs. 923 ms),with no interaction. At Region 5, there were no signifi-

cant main effects and the interaction was not significant.Finally, no significant effects were found at Region 6.The absence of total time effects at Regions 5 and 6 sug-gests that the effect obtained in the regression path read-ing times was short-lived.

As with Experiment 1, we used linear regression todetermine whether the frequency of relative clause com-pletions to ambiguous sentence fragments predicted thereading time costs for the ambiguous sentences follow-ing syntactic disambiguation (see Table 3). The indepen-dent variables were the number of relative clausecompletions for fragments with only occurring in eithera which- or a whether-context. The dependent variables

T

1077107810791080108110821083108410851086108710881089109010911092109310941095109610971098

1099

1100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130

113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154

11551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186




UNCORREC

were the differences in reading time between ambiguousand unambiguous sentences (in which- and whether-con-texts) in first pass, regression path, and total time mea-sures for Regions 5 and 6. No effects were obtained infirst pass reading times (R2s < .07, Fs < 2.2). However,there was a significant positive relationship betweenthe completion data and regression path reading timesfor sentences in which-contexts at Region 6 (R2 = .22,F = 9.32, p < .01). As the relationship was positive, itwas in the opposite direction to that predicted by Con-straint-satisfaction theorists, indicating that readershad more processing difficulty for sentences that re-ceived high rather than low rates of relative clause com-pletions. A similar, but non-significant effect, wasobtained in total reading times at Region 5 for sentencesin which-contexts (R2 = .10, F = 3.48, .05 < p < .10),with no other effects (R2s < .03, Fs < 1). Thus, the sen-tence completion data do not enable us to predict thereading time cost following syntactic disambiguationas Constraint-satisfaction theorists have proposed (e.g.,Spivey & Tanenhaus, 1998; Spivey-Knowlton & Sedivy,1995).

Discussion

This experiment showed that readers had difficulty inprocessing ambiguous sentences in whether-contexts butnot in which-contexts. It should be noted that the effectobtained in whether-contexts was short-lived (foundonly in the regression path measure) and occurred atthe region following the disambiguating region. Thetransitory nature of the effect was not surprising, sincethe effect that Liversedge et al. (2002) obtained for thisform of ambiguity also occurred for re-reading time,but not in first pass or total reading times. Re-readingtime is a measure that captures effects in fixations madewhen text is being re-inspected. In both cases, the effectmay have been short-lived because it is relatively easyfor the parser to recover from an initial misanalysiswhen one of two relatively dispreferred analyses turnsout to be incorrect compared with when a single stronglypreferred analysis of an ambiguity turns out to beincorrect.

It may be of concern to some that we did not obtainfirst pass effects of context in Experiment 2, given ourargument that context influences initial parsing deci-sions. However, we do not believe that the failure to ob-tain first pass reading time effects is in any wayproblematic for our account. When a reader encountersa portion of a sentence that causes them processing dif-ficulty, there are three possible oculomotor responsesthat they may make: (1) remain fixating the difficultword, (2) make a leftward regression to reinspect textto assist in processing the difficulty, (3) make a right-ward saccade to inspect novel text that could also facil-itate processing. The regression path reading time

EDPROOF

measure (Konieczny et al., 1997; Liversedge, Paterson,et al., 1998, 1998), which showed our effects, is sensitiveto both the first and the second of these possible behav-iors. It provides an index of the time the participantspent initially detecting a problem and then re-readingthe text (presumably in an attempt to overcome process-ing difficulty) prior to fixating novel linguistic material.A strong, and in our view reasonable, argument hasbeen made that that the existence of a regression path ef-fect unambiguously indicates that disruption to earlyprocessing has occurred even if a proportion of that ef-fect reflects recovery rather than initial detection of thedifficulty (Clifton et al., 2003). Note also, that since weobtained a regression path reading time effect, but nocorresponding first pass effect, this suggests that readersinitiated a regressive saccade to re-inspect text immedi-ately upon detection of the syntactic misanalysis. Thus,such regressive saccades served to truncate first passreading times for the sentence region in which readersfirst detected the difficulty.

To be clear, we believe that the disruption to process-ing that we observed occurred because readers initiallyadopted a main clause analysis of the ambiguity inwhether-contexts, and then experienced difficulty whenthe sentence was disambiguated as a reduced relativeclause. An alternative possibility is that the effect mayhave been due to readers initially adopting the relativeclause analysis of the ambiguity, regardless of the con-text, and having difficulty in integrating this analysiswith the whether-context, since the whether-context didnot set-up an expectation for modifying information.We considered this latter explanation to be somewhatunlikely, however. If the effect was due to readers havingdifficulty in integrating relative clause information withcontext, then we would have expected it to occur forboth ambiguous and unambiguous sentences, and notexclusively for the ambiguous ones. Crucially, irrespec-tive of the origins of the whether-context effect, no corre-sponding effect was obtained in which-contexts. Thus, itappeared that readers had initially adopted the relativeclause analysis of the ambiguity in this context, sincethey had no difficulty in processing the sentence whenit was disambiguated as having a reduced relative clausestructure.

The present results are consistent with Liversedgeet al.�s (2002) claim that contrastive focus might influ-ence the parsing of a relative clause ambiguity whenthe preferred active transitive analysis is unavailable.The reading time data are also consistent with Sedivy�s(2002) claim that only and prior referential context canjointly affect the processing of syntactic ambiguities.However, the sentence completion data were not consis-tent with her account of how only and discourse contextinteract during language processing. According to Sedi-vy, contexts containing two referents (i.e., ‘‘contrast’’contexts) should bias the readers into adopting the main

T

11871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221

1222

12231224122512261227122812291230123112321233123412351236123712381239

1240124112421243124412451246124712481249

1250125112521253125412551256125712581259126012611262126312641265

1266126712681269

12701271127212731274




EC

clause analysis of an ambiguity. However, the sentencecompletions for fragments with only in whether-contexts(which were comparable to Sedivy�s two-referent con-texts) indicated that readers did not have a preferencefor the main clause analysis of the ambiguity (partici-pants producing 58% relative clause completions). Onepossibility is that the two-referent contexts eliminatethe expectation that only creates for modifying informa-tion, without creating a preference for adopting themain clause analysis of an ambiguity. That is to say,when the reader encounters only in a sentence, it causesthem to instantiate focus and contrast sets in their dis-course model, and to seek information that specifieshow these sets differ. When this information is not al-ready available then readers will anticipate further mod-ifying information, such as that supplied by a relativeclause. However, when context specifies the nature ofthe contrast, then readers will not anticipate furthermodifying information, but neither will they be biasedinto adopting the main clause analysis of an ambiguity.Thus, two-referent contexts may stop readers fromanticipating further modifying information, rather thaninhibiting them from adopting the relative clause analy-sis of an ambiguity.

The results are problematic for the Garden path the-ory, since they suggest that contrastive focus can affectthe processing of some forms of syntactic ambiguity;although advocates of the theory might argue that effectsof context in cases like this could be a result of rapidreanalysis, with context affecting reanalysis processesrather than guiding initial parsing decisions (e.g., Fra-zier, 1995). The results are readily accommodated bytheories that allow referential factors to influence initialparsing decisions, including the Constraint-satisfactiontheory.

127512761277127812791280128112821283128412851286128712881289129012911292129312941295

UNCORR

General discussion

We have reported two eye-tracking experiments thatare informative about the conditions in which using only

and an interrogative to indicate contrastive focus can af-fect the processing of a relative clause ambiguity. In con-ducting these experiments, we distinguished betweenmain clause/relative clause ambiguities which do or donot permit an active transitive analysis. Previous re-search (Liversedge et al., 2002; Paterson et al., 1999) sug-gested that whereas the parsing of ambiguities that allowthe transitive analysis might be impervious to the pro-cessing demands of contrastive focus, the parsing ofambiguities that do not permit this analysis might besusceptible to its influence. The results of the presentstudies suggest that this is indeed the case.

In Experiment 1 we found that readers had difficultyin parsing relative clause ambiguities that temporarilypermitted the active transitive analysis, despite the pres-

EDPROOF

ence of only, and irrespective of an interrogative context.In Experiment 2, the structure of the ambiguous sen-tences ruled out the possibility of a transitive analysis,with the sentences being ambiguous between a relativeclause analysis and an intransitive analysis. Under theseconditions, contrastive focus did affect how the ambigu-ity was processed. Readers had no difficulty when only

and a which-phrase created an expectation for modifyinginformation. But they did have difficulty when awhether-phrase specified a contrast between two sets,thereby obviating the need for modifying information,as observed in increased regression path reading timesfor the post-critical region of ambiguous sentences.

A second key finding from Experiment 1 was thatalthough only and an interrogative context did not affectsyntactic processing, it did affect how easily sentenceswere semantically integrated with the reader�s currentdiscourse model. Sentences were processed more easilywhen they provided information that was congruentwith expectations generated by wh-word contexts, irre-spective of whether the sentences were ambiguous orunambiguous. Thus, we found that regression path read-ing times for the post-critical region, and total readingtimes for the critical region were shorter for both ambig-uous and unambiguous sentences when these occurred inwhich-contexts rather than who-contexts. We took thisto show that the semantic integration of relative clauseinformation was easier in which-contexts that had creat-ed an expectation for modifying information than inwho-contexts that had not. This finding demonstratedthat our interrogative manipulation had influenced theprocessing of the relative clause sentences used in Exper-iment 1, but, crucially, that it had done so without affect-ing initial parsing decisions.

It should be clear that, when considered together, theresults from Experiments 1 and 2 are problematic formodular processing accounts, such as Garden path the-ory, which require that the initial parsing of syntacticambiguities is unaffected by extra-syntactic factors. Con-trary to this theory, our results suggest that referentialfactors can affect the processing of syntactic ambiguities,particularly when the most strongly preferred analysis isunavailable and the parser must select between dispre-ferred alternatives. We noted, however, that proponentsof Garden path theory (e.g., Frazier, 1995) might arguethat these effects are due to context triggering rapidreanalysis of the ambiguity rather than guiding initialparsing decisions.

The present results are also problematic for Ni et al.�s(1996) account in terms of Referential theory. Ni et al.claimed that contrastive focus can eliminate parsing dif-ficulty for syntactic ambiguities; however, our resultsindicate that this depends on the ambiguity, some formsof ambiguity being more susceptible to its influence thanothers. Similarly, whereas Sedivy (2002) claimed thatonly and prior referential context jointly determine

T

12961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351

13521353135413551356135713581359136013611362136313641365136613671368136913701371

137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407




UNCORREC

which analysis of an ambiguity is initially selected, ourresults demonstrate that this is not always the case,but depends on which analyses are available to the pars-er, and whether one is much more strongly preferredthan the others. Instead, the present results are mostreadily accommodated by a theoretical account requir-ing that the alternative syntactic analyses of an ambigu-ity are constructed in parallel, and that referentialfactors, such as the processing of contrastive focus,can influence which analysis is selected. Constraint-satis-faction theory (e.g., MacDonald, 1994; MacDonaldet al., 1994; Tanenhaus & Trueswell, 1995) is the bestknown of these accounts, but other theories also allowfor the parallel construction and evaluation of alterna-tive syntactic analyses (e.g., Unrestricted Race theory:Van Gompel & Pickering, 2001; Van Gompel, Pickering,Pearson, & Liversedge, 2005). Constraint-satisfactiontheory differs from the other accounts by requiring thatthe rival analyses of an ambiguity compete for adoption,with various sources of linguistic and non-linguisticinformation imposing constraints that support one orother of the possible analyses.

Constraint-satisfaction theory can accommodate thepresent results by requiring that there are strong syntac-tic constraints supporting the construction of an activetransitive analysis of an ambiguity, whenever it is avail-able. Referential factors will also impose constraints onambiguity resolution. Thus, whereas only and a which-context supports a relative clause analysis of the ambi-guity, only and either a who- or a whether-context doesnot. To explain the Experiment 1 results, the theorywould require that syntactic constraints supporting thetransitive analysis over-ride referential processing pref-erences, causing readers initially to adopt this analysisof the ambiguity irrespective of the interrogative con-text. Having adopted the transitive analysis, readersmust reanalyze if it subsequently proves to be deficient.

The theory can account for the Experiment 2 resultsby requiring that, with the transitive analysis ruled out,syntactic constraints supporting either the intransitive orrelative clause analyses are much weaker, and moreequally balanced; and that under these conditions, refer-ential factors influence the competition between them.Thus, using only and a which-context to create an expec-tation for modifying information biases the processorinto adopting the relative clause analysis of the ambigu-ity, thereby avoiding difficulty at disambiguation. Bycontrast, only and a whether-context does not requiremodifying information. Consequently, readers are lesslikely to adopt this analysis, and to incur a processingcost at syntactic disambiguation.

More problematic for advocates of the Constraint-satisfaction theory was our finding that off-line sentencecompletions were not predictive of on-line parsing pref-erences. Its proponents (e.g., Spivey & Tanenhaus, 1998;Spivey-Knowlton & Sedivy, 1995) often use sentence

EDPROOF

completion data to estimate the relative strength of con-straints favoring the alternative analyses of an ambiguityat particular points in a sentence. However, we foundthat although sentence completions were informativeabout the effectiveness of only and an interrogative con-text in causing readers to anticipate modifying informa-tion, they were not necessarily informative aboutwhether this affected the initial parsing of the ambiguity.Thus, in Experiment 1, the sentence completion dataindicated that only and which more strongly predisposedreaders to adopting the relative clause analysis of theambiguity than did only and who. However, this affectedsemantic integration rather than parsing decisions, sinceboth ambiguous and unambiguous relative clause sen-tences were easier to integrate with which- than who-con-texts. In Experiment 2, the sentence completionsindicated that only with a which-context were reliablymore likely to predispose readers to adopting the relativeclause analysis of the ambiguity than only with awhether-context. However, in this case the relative pref-erence did affect parsing decisions, with readers havingdifficulty in processing the ambiguity in whether-con-texts, but not in which-contexts. Thus, although sentencecompletions may be informative about contextual influ-ences on completion preferences, it is not always the casethat these preferences reflect initial parsing decisions.We note that several other researchers have found a mis-match between sentence completion data and on-linereading times (Clifton, Kennison, & Albrecht, 1997; Liv-ersedge, Paterson, et al., 1998, 1998; Liversedge et al.,2002; Murray & Liversedge, 1994; Paterson et al.,1999; Pickering, Traxler, & Crocker, 2000; Van Gompel& Pickering, 2001). Although such findings do not con-stitute evidence against Constraint-satisfaction theoryper se, they do suggest that sentence completion meth-odology does not provide a reliable index of on-lineparsing decisions.

We attributed the different pattern of results inExperiments 1 and 2 to post-verbal information affectingthe availability of a transitive analysis prior to syntacticdisambiguation. However, other explanations are possi-ble. One such explanation might be that the ambiguousverbs used in Experiment 1 were more biased towards atransitive analysis of the ambiguity than those in Exper-iment 2, and that the differing transitivity biases wereresponsible for our effects. In other words, readers mighthave had more processing difficulty when attempting toovercome a strong transitivity bias in Experiment 1 thana weak transitivity bias in Experiment 2. Alternatively, itmay have been harder for readers to assign a relativeclause analysis to the ambiguity for sentences in Exper-iment 1 than in Experiment 2, irrespective of the avail-ability of a transitive analysis. Assigning a relativeclause analysis in Experiment 1 involved goal extraction,whereas it involved theme extraction for the sentences inExperiment 2. Relative clause analyses involving some

TD

14081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463

1464146514661467146814691470147114721473147414751476147714781479148014811482

1483

1484

1485

1486148714881489

1490149149

1493149149

1496149149

1499150

1501150150

1504150150

1507150150

1510151151

1513151151




UNCORREC

forms of extraction are more difficult to process thanothers (e.g., Gibson, 1998; Gibson & Hsiao, 2005).Therefore, it may be that extraction from a goal positionwas hard in Experiment 1 while extraction from a themeposition was easy in Experiment 2. Future researchcould clarify this issue.

At least one other piece of evidence is consistent withour account. MacDonald (1994) also explored the effectof post-verbal evidence for a transitive analysis of anambiguity by varying the first word that followed anambiguous verb so that it either was or was not consis-tent with a transitive analysis. For example, includingthe preposition in in the sentence The dictator fought in

the coup was hated immediately rules out the possibilityof a transitive analysis of fought. However, if the wordjust is included after the verb, as in The dictator fought

just after dawn was hated, it prolongs the possibility ofa transitive analysis (e.g., the sentence could continueas just one soldier). Readers had more difficulty in pro-cessing sentences when the possibility of a transitiveanalysis was prolonged than when it was short-lived.The effect cannot be attributed to any transitivity biasesassociated with individual verbs, since both forms of thesentence included the same ambiguous verb. Further-more, it cannot be attributed to differences in the pro-cessing of different relative clause types, since thesentences ultimately were disambiguated as the sametype of relative clause construction. Instead, the effectappears to be due to the post-verbal information (i.e.,in or just) either ruling out or prolonging the possibilityof a transitive analysis of the ambiguity.

Finally, like Sedivy (2002), we used materials inwhich contrastive focus and context either did or didnot create an expectation for a relative clause, but thepropositional content of the relative clause did notuniquely specify a referent in the prior discourse context.In this respect, our manipulation was not of the typethat often is used when testing the Referential theory(e.g., Altmann & Steedman, 1988), since this theory stip-ulates that readers prefer to adopt the relative clauseanalysis of an ambiguity when the propositional contentof this analysis uniquely identifies a discourse referent.Neither Sedivy nor Ni et al. (1996) included the samestipulation in their account of the influence of contras-tive focus on parsing. Nevertheless, our results suggestthat Sedivy was correct in supposing that a combinationof context and contrastive focus would bias readers intoanticipating relative clause information, regardless ofwhether it uniquely identified a discourse referent. How-ever, we also found that the contextual manipulation didnot impinge on initial parsing decisions for the tempo-rarily transitive ambiguity investigated in Experiment1. It is yet to be determined whether further strengthen-ing the contextual manipulation by additionally requir-ing that the relative clause does uniquely identify adiscourse referent will be sufficient to over-ride the pref-

ROOF

erence for a transitive analysis of a relative clauseambiguity.

In conclusion, our studies contribute to resolving thecontroversy concerning the influence of contrastive fo-cus on the parsing of relative clause ambiguities (Cliftonet al., 2000; Liversedge et al., 2002; Ni et al., 1996; Pat-erson et al., 1999; Sedivy, 2002), by showing that syntac-tic processing preferences modulate the influence ofcontrastive focus on parsing decisions. The results sug-gest that when there is a strongly preferred syntacticanalysis, it will be adopted regardless of the influenceof contrastive focus. However, if the parser is selectingbetween relatively dispreferred analyses, then underthese conditions, the choice may be influenced by theprocessing demands associated with the computationof contrastive focus. We have suggested that these re-sults are consistent with a processor that integrates mul-tiple sources of information, including referentialfactors, when resolving syntactic ambiguities.

PUncited reference

Pearlmutter and MacDonald (1995).

EAppendix A

Sentence materials used in Experiment 1. Contexts wereinterrogatives formed from a which- or who-phrase, and fol-lowed by either an ambiguous reduced or an unambiguousunreduced relative clause sentence.

1. Harry wondered [which actors/who] received an apology.Only actors [who were] refused an audition received an apol-ogy within the week.

2. Liz wondered [which actresses/who] gave a curtsy. Onlyactresses [who were] passed a bouquet gave a curtsyimmediately.

3. Tom wondered [which builders/who] fitted a kitchen. Onlybuilders [who were] paid a deposit fitted a kitchen withinthe week.

4. Bronwen wondered [which children/who] ate an egg. Onlychildren [who were] passed a spoon ate an egg straightaway.

5. Phil wondered [which clerks/who] wanted a job. Only clerks[who were] issued a work permit wanted a job before thesummer.

6. Alice wondered [which directors/who] sent a reply. Onlydirectors [who were] faxed a message sent a reply the nextday.

7. Jeremy wondered [which editors/who] contacted a lawyer.Only editors [who were] served a writ contacted a lawyerstraightaway.

8. Caroline wondered [which executives/who] found a confi-dential file. Only executives [who were] typed a letter founda confidential file later that day.

9. John wondered [which farmers/who] ploughed a field. Onlyfarmers [who were] sold a tractor ploughed a field thatafternoon.

T

O

1516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576

157715

15791515

15821515

15851515

1588151590

1591

1592

1593159415951596

1597151516

16011616

16041616

1607161616

1611161616

1615161616

1619161616

1623161616

1627161616

1631161616




UNCORREC

10. Sally wondered [which footballers/who] scored a goal. Onlyfootballers [who were] offered an orange scored a goalstraightaway.

11. Peter wondered [which foreigners/who] drank a coffee. Onlyforeigners [who were] bought a hamburger drank a coffee atthe same time.

12. Louise wondered [which gamblers/who] placed a bet. Onlygamblers [who were] lent a tenner placed a bet the next day.

13. Keith wondered [which nurses/who] insulted a patient. Onlynurses [who were] bought a present insulted a patient today.

14. Nicky wondered [which girls/who] solved an equation. Onlygirls [who were] sold a calculator solved an equation thesame afternoon.

15. Bernard wondered [which thugs/who] surrendered a weap-on. Only thugs [who were] granted an amnesty surrendereda weapon straightaway.

16. Gillian wondered [which wives/who] developed a rash. Onlywives [whowere] knitted a scarf developed a rash within a week.

17. Terry wondered [which inspectors/who] wore a disguise.Only inspectors [who were] assigned a case wore a disguisethe next day.

18. Hayley wondered [which journalists/who] wrote a novel.Only journalists [whowere] asked a favourwrote a novel thatyear.

19. Miles wondered [which policemen/who] caught a criminal.Only policemen [who were] sent a bribe caught a criminalimmediately.

20. Rita wondered [which lecturers/who] got a promotion.Only lecturers [who were] awarded a fellowship got a pro-motion within the year.

21. Mick wondered [which pensioners/who] sent a letter. Onlypensioners [who were] baked a cake sent a letter yesterday.

22. Gloria wondered [which politicians/who] gave an answer.Only politicians [who were] asked a question gave an answerstraightaway.

23. George wondered [which reviewers/who] wrote an article.Only reviewers [who were] sent a book wrote an articlewithin the week.

24. Nicola wondered [which salesman/who] bought a car. Onlysalesmen [who were] paid a commission bought a car thenext day.

25. Reg wondered [which schoolboys/who] wrote an essay.Only schoolboys [who were] lent a book wrote an essaythe next day.

26. Fiona wondered [which spectators/who] made a complaint.Only spectators [who were] told a joke made a complaintthat evening.

27. Barry wondered [which stockbrokers/who] called a relative.Only stockbrokers [who were] made an offer called a rela-tive within the hour.

28. Lorna wondered [which suspects/who] signed a confession.Only suspects [who were] refused a lawyer signed a confes-sion within the hour.

29. Geoff wondered [which teenagers/who] invited a juggler.Only teenagers [who were] allowed a party invited a jugglerstraightaway.

30. Rachel wondered [which toddlers/who] lost a rattle. Only tod-dlers [who were] peeled a banana lost a rattle under the sofa.

31. Josh wondered [which tradesmen/who] bought a drink.Only tradesmen [who were] offered a job bought a drinkstraightaway.

F

32. Debbie wondered [which visitors/who] bought a gift. Only vis-itors [who were] told a story bought a gift when they arrived.

33. Jake wondered [which widows/who] married a toyboy.Only widows [who were] left a fortune married a toyboythat same year.

34. Louisa wondered [which women/who] phoned a salesman.Only women [who were] delivered a catalogue phoned asalesman the same day.

35. Stan wondered [which workers/who] smoked a cigarette.Only workers [who were] allowed a tea-break smoked a cig-arette that morning.

36. Jade wondered [which youths/who] bought a jacket. Onlyyouths [whowere] posted a cheque bought a jacket yesterday.

EDPROAppendix B

Sentence materials used in Experiment 2. Contexts wereinterrogatives formed from a which- or whether-phrase, andfollowed by either an ambiguous reduced or an unambiguousunreduced relative clause sentence.

1. Emma wondered [which soldiers/whether the soldiers or thepilots] suffered psychological problems. Only soldiers [whowere] fought in the jungle suffered psychological problemslater.

2. Lee wondered [which toddlers/whether the toddlers or theschoolchildren] sulked. Only toddlers [who were] washedin the playroom sulked quietly.

3. Kellywondered [whichCubScouts/whether theCubScoutsortheGirl Guides] blushed. Only Cub Scouts [whowere] salutedin the parade blushed a lot if their friends had seen them.

4. Simon wondered [which motorists/whether the motorists orthe pedestrians] received a warning. Only motorists [whowere] stopped in the car park received a warning if theyhad parked illegally.

5. Joanne wondered [which visitors/whether the visitors or thelocals] admired the scenery. Only visitors [who were]sketched in the city admired the scenery if they had notbeen before.

6. Ed wondered [which paratroopers/whether the paratroop-ers or the naval officers] received a medal. Only paratroop-ers [who were] attacked in the morning received a medal ifthey had acted heroically.

7. Louise wondered [which defenders/whether the defendersor the goalkeepers] complained to the referee. Only defend-ers [who were] fouled in extra time complained to the refer-ee if they were sent off.

8. Mark wondered [which heavyweights/whether the heavy-weights or the middleweights] failed to win the match. Onlyheavyweights [who were] punched in the early rounds andfailed to win the match if they were injured.

9. Sally wondered [which teenagers/whether the teenagers orthe parents] enjoyed going to the concert. Only teenagers[who were] pushed in the crowd enjoyed going to the con-cert if they had got to the front.

10. Dave wondered [which artists/whether the artists or the dra-ma students] became famous. Only artists [who were] filmedin the local park became famous if they had found a goodagent.

T

1635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695

1696169169169

1700170170170

1704170170170

1708170171171

1712171171171

1716171171171

1720172172172

1724

1725

1726

172717281729

173017311732

17331734173517361737

173817391740

174117421743

174417451746

174717481749

1750175117521753

17541755




UNCORREC

11. Robert wondered [which politicians/whether the politiciansor the public] celebrated. Only politicians [whowere] cheeredin the houses of parliament celebrated at home if they hadwon the vote.

12. Hayleywondered [whichmusicians/whether themusicians orthe singers] impressed the manager. Only musicians [whowere] tutored in the studio impressed the manager if theyhad played well.

13. Steve wondered [which teachers/whether the teachers or thestudents] worried. Only teachers [who were] interviewed inthe common room worried all day if they had been threa-tened with redundancy.

14. Stella wondered [which women/whether the women or themen] purchased a portrait. Only women [who were] paintedin the studio purchased a portrait later if they wereimpressed.

15. Geoff wondered [which teachers/whether the teachers or theheadmaster] thanked the guest speaker. Only teachers [whowere] assisted in the fete thanked the guest speaker after-wards if they had enjoyed the speech.

16. Lorna wondered [which doctors/whether the doctors or thechemists] relaxed that afternoon. Only doctors [who were]studied in the laboratory relaxed that afternoon if theyhad completed the experiment.

17. Chris wondered [which students/whether the students orthe lecturers] became drunk. Only students [who were]served in the wine bar became drunk that evening if theyhad finished their shift.

18. Claire wondered [which postgraduates/whether the post-graduates or the research assistants] continued their duties.Only postgraduates [who were] instructed in the collegecontinued their duties if they had time.

19. Richard wondered [which footballers/whether the football-ers or the cricketers] triumphed in a game. Only footballers[who were] coached in the stadium triumphed in a game ifthey had been training.

20. Rachel wondered [which gymnasts/whether the gymnasts orthe dancers] improved their routine. Only gymnasts [whowere] directed in the gymnasium improved their routine ifthey had practised.

21. Tom wondered [which pensioners/whether the pensionersor the housewives] drank in the pub. Only pensioners[who were] helped in the charity shop drank in the pub ifthey had found a bargain.

22. Rebecca wondered [which postgraduates/whether thepostgraduates or the undergraduates] succeeded in theirexams. Only postgraduates [who were] taught in thesummer school succeeded in their exams if they hadrevised.

23. Phil wondered [which mechanics/whether the mechanics orthe students] became qualified engineers. Only mechanics[who were] trained in the workshop became qualified engi-neers if they had worked hard.

24. Alison wondered [which visitors/whether the visitors or thecaterers] complimented the host. Only visitors [who were]entertained in the bar complimented the host if they hadstayed over.

25. George wondered [which mountaineers/whether the moun-taineers or the ramblers] succeeded in the climb. Onlymountaineers [who were] trained in the Peaks succeededin the climb if they had taken their time.

DPROOF

26. Caroline wondered [which artists/whether the artists or themusicians] enjoyed the lessons. Only artists [who were]taught in the college enjoyed the lessons if they had madenew friends.

27. Keith wondered [which gangsters/whether the gangsters orthe thugs] joined the rival gang. Only gangsters [who were]fought in the alleyway joined the rival gang if they had lostthe battle.

28. Sue wondered [which mercenaries/whether the mercenariesor the snipers] concealed themselves during the day. Onlymercenaries [who were] hunted in the mountains concealedthemselves during the day if they had been afraid.

29. Gary wondered [which managers/whether the managers orthe directors] expected a wage rise. Only managers [whowere] telephoned in the day expected a wage rise if theyhad met their targets.

30. Julia wondered [which youths/whether the youths or theadults] visited a psychiatrist. Only youths [who were] bul-lied in the playground visited a psychiatrist if they weredepressed.

31. Tony wondered [which fighter pilots/whether the fighterpilots or the soldiers] parachuted into the desert. Only fight-er pilots [who were] attacked in the day parachuted into thedesert if they had been shot.

32. Helen wondered [which spies/whether the spies or the ter-rorists] threatened the manager. Only spies [who were]filmed in the seedy motel threatened the manager on theirway out.
E
References

Altmann, G. T. M., & Steedman, M. (1988). Interaction withcontext during human sentence processing. Cognition, 30,191–238.

Altmann, G. T. M., Garnham, A., & Dennis, Y. (1992).Avoiding the garden path: Eye-movements in context.Journal of Memory And Language, 31, 685–712.

Altmann, G. T. M., Garnham, A., & Henstra, J. A. (1994).Effects of syntax in human sentence parsing: Evidenceagainst a structure-based proposal mechanism. Journal ofExperimental Psychology: Language, Memory and Cogni-

tion, 20, 209–216.Altmann, G. T. M., van Nice, K. Y., Garnham, A., & Henstra,

J. A. (1998). Late closure in context. Journal of Memory and

Language, 38, 459–484.Binder, K. S., Duffy, S. A., & Rayner, K. (2001). The effects of

thematic fit and discourse context on syntactic ambiguityresolution. Journal of Memory and Language, 4, 297–324.

Britt, M. A. (1994). The interaction of referential ambiguity andargument structure in the parsing of prepositional phrases.Journal of Memory and Language, 33, 251–283.

Britt, M. A., Perfetti, C. A., Garrod, S., & Rayner, K. (1992).Parsing in discourse: Context effects and their limits. Journalof Memory and Language, 31, 293–314.

Clark, H. H. (1973). The language-as-fixed-effect fallacy: Acritique of language statistics in psychological research.Journal of Verbal Learning and Verbal Behavior, 12,335–359.

Clifton, C., Bock, J., & Rado, J. (2000). Effects of the focusparticle only and intrinsic contrast on comprehension of

T

1756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816

18171818182

1821182182

1824182182

1827182182183

1831183183

1834183183183

1838183184184

1842184184184

1846184184184

1850185185185

1854185185185

1858185186186

1862186186

1865186186186

1869187187187187

1874187187187




UNCORREC

reduced relative clauses. In A. Kennedy, R. Radach, D.Heller, & J. Pynte (Eds.), Reading as a perceptual process.Amsterdam: Elsevier.

Clifton, C., Jr., Kennison, S. M., & Albrecht, J. E. (1997).Reading the words her, him, his: Implications for parsingprinciples based on frequency and on structure. Journal ofMemory and Language, 36, 276–292.

Clifton, C., Traxler, M. J., Mohamed, M. T., Williams, R. S.,Morris, R. K., & Rayner, K. (2003). The use of thematicrole information in parsing: Syntactic processing autonomyrevisited. Journal of Memory and Language, 49, 317–334.

Crain, S., & Steedman, M. (1985). On not being led up thegarden path: The use of context by the psychological syntaxprocessor. In D. R. Dowty, L. Kartunnen, & A. M. Zwicky(Eds.), Natural language parsing: Psychological, computa-

tional, and theoretical perspectives. Cambridge: CambridgeUniversity Press.

Ferriera, F., & Clifton, C. (1986). The independence of syntacticprocessing. Journal of Memory and Language, 25, 348–368.

Frazier, L., & Clifton, C. (1996). Construal. Cambridge, MA:MIT Press.

Frazier, L., & Clifton, C (2002). Processing �d-linked� phrases.Journal of Psycholinguistic Research, 31, 633–659.

Frazier, L. (1995). Constraint satisfaction as a theory ofsentence processing. Journal of Psycholinguistic Research,

24, 437–468.Frazier, L., & Rayner, K. (1982). Making and correcting errors

during sentence comprehension: Eye movements in thestudy of structurally ambiguous sentences. Cognitive Psy-

chology, 14, 178–210.Gibson, E. (1998). Linguistic complexity: locality of syntactic

dependencies. Cognition, 68, 1–76.Gibson, E., & Hsiao, F. (2005). Processing relative clauses in

Chinese. Cognition, 90, 3–27.Hanna, J. E., Spivey-Knowlton, M. J., & Tanenhaus, M. K.

(1996). Integrating discourse and local constraints inresolving lexical thematic ambiguities. In Proceedings of

the sixteenth annual cognitive science society meeting

(pp. 266–271). Hillsdale, NJ: Erlbaum.Heim, I. (1982). The semantics of definite and indefinite noun

phrases. Doctoral dissertation, University of Massachusetts,Amherst.

Konieczny, L., Hemforth, B., Scheepers, C., & Strube, G.(1997). The role of lexical heads in parsing: evidence fromGerman. Language and Cognitive Processes, 12, 307–348.

Liversedge, S. P., & Findlay, J. M. (2000). Saccadic eyemovements and cognition. Trends in Cognitive Science, 4,6–14.

Liversedge, S. P., Paterson, K. B., & Clayes, E. (2002). Theinfluence of only on syntactic processing of ‘‘long’’ relativeclause sentences. Quarterly Journal of Experimental Psy-

chology, 55A, 225–240.Liversedge, S. P., Paterson, K. B., & Pickering, M. J. (1998).

Eye movements and measures of reading time. In G.Underwood (Ed.), Eye guidance in reading and scene

perception. Oxford: Elsevier Science.Liversedge, S. P., Pickering, M. J., Branigan, H. P., & Van

Gompel, R. P. G. (1998). Processing arguments andadjuncts in isolation and in context: The case of by-phraseambiguities in passives. Journal of Experimental Psychology:

Learning, Memory and Cognition, 24, 461–475.

EDPROOF

Liversedge, S. P., Pickering, M. J., Clayes, E. L., & Branigan,H. P. (2003). Thematic processing in adjuncts: Evidencefrom an eye tracking experiment. Psychonomic Bulletin and

Review, 10, 667–675.MacDonald, M. C. (1994). Probabilistic constraints and syn-

tactic ambiguity resolution. Language and Cognitive Pro-

cesses, 9, 157–201.MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S.

(1994). The lexical nature of syntactic ambiguity resolution.Psychological Review, 101, 676–701.

McRae, K., Spivey-Knowlton, M. J., & Tanenhaus, M. K.(1998). Modelling the influence of thematic fit and otherconstraints in on-line sentence comprehension. Journal ofMemory and Language, 38, 283–312.

Masson, M. E. J., & Loftus, G. R. (2003). Using confidenceintervals for graphically based data interpretation. CanadianJournal of Experimental Psychology, 57, 203–220.

Murray, W. S., & Liversedge, S. P. (1994). Referential contexteffects on syntactic processing. In C. Clifton, Jr., L. Frazier,& K. Rayner (Eds.), Perspectives on sentence processing.Hillsdale, NJ: Erlbaum.

Ni, W., Crain, S., & Shankweiler, D. (1996). Sidesteppinggarden paths: The contribution of syntax, semantics andplausibility in resolving ambiguities. Language and CognitiveProcesses, 11, 283–334.

Paterson, K. B., Liversedge, S. P., & Underwood, G. (1999).The influence of focus operators on parsing of short reducedrelative clause sentences. Quarterly Journal of Experimental

Psychology, 52A, 717–738.Pearlmutter, N. J., & MacDonald, M. C. (1995). Individual

differences and probabilistic constraints in syntactic ambi-guity resolution. Journal of Memory and Language, 34,521–542.

Pesetsky, D. (1987). Wh-in-situ: movement and unselectivebinding. In E. Reuland & A. ter Meulen (Eds.), The

representation of (in)definiteness (pp. 98–129). CambridgeMA: MIT Press.

Pickering, M. J., Traxler, M. J., & Crocker, M. W. (2000).Ambiguity resolution in sentence processing: Evidenceagainst frequency-based accounts. Journal of Memory and

Language, 43, 447–475.Raaijmakers, J. G.W., Schrijnemakers, J.M. C., &Gremmen, F.

(1999). How to deal with ‘‘the language-as-fixed-effects-fallacy’’: Common misconceptions and alternative solutions.Journal of Memory and Language, 41, 416–426.

Rayner, K. (1998). Eye movements in reading and informationprocessing: 20 years of research. Psychological Bulletin, 124,372–422.

Rayner, K., & Duffy, S. A. (1986). Lexical complexity andfixation time in reading: Effects of word frequency, verbcomplexity, and lexical ambiguity. Memory and Cognition,

33, 527–544.Rayner, K., Carlson, M., & Frazier, L. (1983). The interaction

of syntax and semantics during sentence processing – eyemovements in the analysis of semantically biased sentences.Journal of Verbal Learning and Verbal Behaviour, 22,358–374.

Rayner, K., Sereno, S. C., Morris, R. K., Schmauder, A. R., &Clifton, C. (1989). Eye-movements and on-line languagecomprehension processes.Language and Cognitive Processes,4, 21–50.

187818791880188118821883188418851886188718881889189018911892189318941895

189618971898

1899190019011902

190319041905

1906190719081909

1910191119121913




Rooth, M. (1992). A theory of focus interpretation. Natural

Language Semantics, 1, 75–116.Sedivy, J. C. (2002). Invoking discourse-based contrast sets and

resolving syntactic ambiguities. Journal of Memory and

Language, 46, 341–370.Spivey, M., & Tanenhaus, M. K. (1998). Syntactic ambiguity

resolution in discourse: Modelling the effects of referentialcontext and lexical frequency. Journal of Experimental

Psychology: Learning, Memory, and Cognition, 24,1521–1543.

Spivey-Knowlton, M., & Sedivy, J. (1995). Resolving attach-ment ambiguities with multiple constraints. Cognition, 55,227–267.

Spivey-Knowlton, M. J., & Tanenhaus, M. K. (1994).Referential context and syntactic ambiguity resolution.In C. Clifton, L. Frazier, & K. Rayner (Eds.), Perspec-

tives on sentence processing (pp. 415–439). Hillsdale, NJ:Erlbaum.

UNCORRECT

1914

OOF

Spivey-Knowlton, M. J., Trueswell, J. C., & Tanenhaus, M. K.(1993). Context and syntactic ambiguity resolution. Cana-dian Journal of Psychology, 47, 276–309.

Tanenhaus, M. K., & Trueswell, J. C. (1995). Sentence compre-hension. In J. Miller & P. Eimas (Eds.), Handbook of

Perception and Cognition. Speech, Language and Communi-

cation (Vol. 11, pp. 217–262). New York: Academic Press.Townsend, D. J., & Bever, T. G. (2001). Sentence comprehension:

The integration of habits and rules. Cambridge, MA: MITPress.

Van Gompel, R. P. G., & Pickering, M. J. (2001). Reanalysis insentence processing: Evidence against current Constraintsatisfaction and two-stage models. Journal of Memory and

Language, 45, 225–258.Van Gompel, R. P. G., Pickering, M. J., Pearson, J., &

Liversedge, S. P. (2005). Evidence against competition duringsyntactic ambiguity resolution. Journal of Memory and

Language, 42, 284–307.

EDPR

parsing with focus particles in context: eye ... · 14 and interrogative contexts jointly...

Documents