Journal of Memory and Languagedoi:10.1006/jmla.2001.2824, available online at http://www.academicpress.com on

Constituent Structure Is Formulated in One Stage

Martin J. Pickering, Holly P. Branigan, and Janet F. McLean

University of Edinburgh, Edinburgh, United Kingdom

In four syntactic priming experiments, participants completed target fragments as “prepositional object” sen-tences (e.g., “The patient showed his leg to the doctor”) or “double object” sentences (e.g., “The patient showed

ted thement 1 prim-hifted”e pro- similarand thenulated in

H

knowsuika Brseco

AdPickburgdoming@

the doctor his leg”) or used another non-ditransitive form. The syntactic form of a prime sentence affecform of participants’ target completions. Experiments 1 to 3 used written sentence completion. Experidemonstrated that priming is a two-way process by comparing “prepositional object” and “double object”ing conditions with a baseline condition containing an intransitive verb. Experiments 2 and 3 found that “sprimes (e.g., “The racing driver showed to the helpful mechanic the problem with the car”) did not prime thduction of “prepositional object” sentences but instead behaved like baseline primes. Experiment 4 foundresults to those of Experiment 3 in spoken sentence production, where participants repeated the prime completed it. We interpret the results in terms of accounts that assume that constituent structure is formone stage. © 2002 Elsevier Science (USA)

Key Words:syntax; priming; production; formulation; constituent; dative.

ow do people make use of grammatical in-The Construction of Constituent Structure

Ar

t

g er n

In

hioo

ss-el isa-ithnsifi-gy.ed-9)elssi-he

hattheone-thet ityn-te-a

formation when they produce utterances? cording to most current models, language pduction involves three stages: conceptualizationor when the prelinguistic message is generaformulationor when the message is encodedlinguistic form, and articulation or when thisform is turned into sound or marks on a paFormulation encompasses both grammaticalcoding, whereby the syntactic content of apppriate lexical elements is retrieved and usedgenerate syntactic structure, and morphophological encoding, whereby the morphophonoloical content is assembled (Bock & Levelt, 199Levelt, 1989; Levelt, Roelofs, & Meyer, 1999). this article, we are concerned with the naturegrammatical encoding. It is uncontroversial tgrammatical encoding results in the constructof a constituent structure representation. Our ccern is with the stages that the processor gthrough in reaching this representation.

The order of the first two authors is arbitrary. We

1

with&lt

the ofrbsa-

ledge the assistance of Stuart Boutell, Robert Haer, and Andrew Stewart. This research was supporteditish Academy Postdoctoral Fellowship (awarded to tnd author) and ESRC Grant No. R000237418.dress correspondence and reprint requests to Ma

ering, Department of Psychology, University of Edinh, 7 George Square, Edinburgh EH8 9JZ, United Kin. Fax: 144 (0) 131 650 3461. E-mail: martin.pickered.ac.uk.

0749-596X/02 $35.00

c-o-

ed,in

e.n-

o-too-

g-4;

ofatonn-es

Most models agree that each level of proceing is independent in the sense that each levconcerned with its own characteristic informtion and its own representations for dealing wthat information. For example, representatioof grammatical encoding do not include speccations of conceptual structure or phonoloSuch independence holds in both strictly feforward models of production (Levelt, 198and models that allow feedback between lev(Dell, 1986). This means that it should be posble, in principle, to determine the nature of trepresentations at each level.

As in many current models, we assume tthe first stage of grammatical encoding is construction of a functional representati(Bock & Levelt, 1994; Garrett, 1980). The prcise nature of this representation is beyond scope of this article. However, we assume thacontains noun and verb lemmas, that is, the stactic information such as grammatical cagory, number, and gender that is associated an individual lexical concept (Kempen Hoenkamp, 1987). Following Bock and Leve(1994), we assume that each noun lemma infunctional representation is tagged for onethe grammatical functions specified by the velemma (e.g., subject–nominative, object accu

c-rt- byhe

rtin-g--

© 2002 Elsevier Science (USA)All rights reserved.

IG

oer n

lyl

uyrbbmcri

nin

ke P,

fo

aP D

tal

blih

o

o

edon-On.al-tleesntal

al-ckces lo-erip-ote

nd-cts ofe-r--y”ite

theel.ue

&hatn-

hatc- ac-ject,sub-icalrlk,

rdasingrs.ctsvel

t

2 PICKERING, BRAN

tive) and that the functional representation dnot specifically encode information about hiarchical relationships between nouns and veThe representation is also partial in that itconcerned simply with functions (verbs) atheir arguments (nouns); for example, it donot contain specifications for adjectives or dterminers. It serves as input to the processesultimately result in the construction of a fulspecified constituent structure that relates althe lexical elements in the sentence.

Strong evidence that a constituent structrepresentation is constructed comes from stactic priming. Under the guise of a memotest, Bock (1986) had speakers alternate tween repeating prime sentences and descrisemantically unrelated target pictures. She nipulated the syntactic forms of the sententhat speakers repeated. For example, the psentence might use the prepositional object(PO) form of an alternating dative verb in ocondition (e.g., “A rock star sold some cocato an undercover agent”) and the double object(DO) form in the other condition (e.g., “A rocstar sold an undercover agent some cocainThe target pictures could be described usingther form. Participants tended to produce a target picture description after a PO primeDO target picture description after a DO primand so on. Bock also found similar effects active/passive sentences.

It is less clear whether these effects are bal-ancedsuch that both alternative structures cbe primed. For example, presentation of a prime would lead to an increased tendencyproduce a PO target, and presentation of a prime would lead to an increased tendencyproduce a DO target. If priming is balancethen both PO and DO primes would produce get responses that differ from a neutral baseprime. Alternatively, priming could be biasedsuch that one construction could be primed the other could not. In that case, a baseprime would have a similar effect to either tPO prime (if POs are not primed) or the Dprime (if DOs are not primed). Perhaps the mlikely possibility is that a marked constructiocould be primed by another instance of that c

struction but an unmarked one could not. A

AN, AND MCLEAN

esr-bs.isdese-that

of

ren-ye-inga-esme

ee

”).ei-O

ae,r

nOtoO

tod,r-

ine

utneeOst

nn-

suming that the DO construction is mark(e.g., because many verbs do not admit this cstruction), we might expect priming for the Dconstruction but not for the PO constructioSome evidence suggests that priming is banced (Bock & Griffin, 2000, Experimen2),[AU1] but other effects have not been reliab(Bock, 1986) or have involved baseline primpresented at the beginning of the experimesession (Hartsuiker & Kolk, 1998).

Other experimental work helps to rule out ternative loci for these priming effects. Boand Loebell (1990) showed that PO sentencontaining prepositional phrases that specifycations (e.g., “The wealthy widow drove hMercedes to the church”) primed PO descrtions when the prepositional phrase did nspecify a location (e.g., “A rock star sold somcocaine to an undercover agent”). These fiings suggest that Bock’s (1986) priming effecould not be due to priming at an earlier stageproduction concerned with the encoding of thmatic (or event structural) relations. Furthemore, Bock and Loebell (1990) found no priming between “Susan brought a book to studand “The girl gave a brush to the man,” desptheir metrical similarities. This suggests that effects are not due to priming at a metrical levIn addition, these priming effects cannot be dto lexical repetition (Bock, 1989; Pickering Branigan, 1998) or discourse factors given tall of these experiments involve isolated setences. Another alternative explanation is tsyntactic priming is actually sensitive to funtional level representations. For example, antive sentence contains a subject and an obwhereas a full passive sentence contains a ject and a phrase with an oblique grammatfunction (the by-phrase). However, Hartsuikeand Westenberg (2000) and Hartsuiker, Koand Huiskamp (1999) found priming of woorder when no functional-level explanation wpossible. Hence, we can conclude that primof constituent structure representations occu

On one account, the processor construthese representations from the functional-leinput in a single stage. We call this a single-stage accountof the formulation of constituen

s-structure. Alternatively, the processor might

T

rse

ldrcr

l

oevrua T nya sorfcca oauninbuineti

tu0tivines

tarealdif-(in-an

herlti-d

ex-tive

-Theon-earrst

sist-verbonalerbt theonal nots be-Par-

at ahern., aase,ce-

sesepo-ro-st

ure5;

le,er:nts.g.,urk-nentsin-

CONSTITUEN

construct the final constituent structure repsentation via one or more intermediate repretations. We call accounts of this type multiple-stage accounts. Multiple-stage accounts hothat fully specified constituent structure repsentations are the ultimate output of syntaprocessing but that intermediate levels of repsentation are computed as well.

There is good evidence against one multipstage account, which we call the deep-structureaccount. This account draws on the assumptiof transformational grammar, namely that stences involve (at least) two fully specified leels of constituent representation: a surface stture, and a deep (or underlying) struct(Chomsky, 1965, 1981). It assumes that lguage production involves the constructionsyntactic structure in (at least) two stages. functional representation is first mapped ontounderlying structure representation, which isturn mapped onto a surface structure represetion. However, Bock, Loebell, and More(1992) found that speakers processed (hence represented) in a related manner theject of an active sentence and the subject passive sentence. They claimed that the ament associated with the subject function opassive sentence is directly assigned to the stituent structure position reserved for subjeTheir main criticism is of the assumption thfunctional relations are defined at two levelsrepresentation, as assumed within some theof linguistics, for example, relational gramm(Pearlmutter, 1983). In other words, they argthat production does not involve relation-chaing operations during functional processsuch that, for example, underlying objects come surface subjects. However, their resare similarly incompatible with a model which a fully specified representation of undlying structure is mapped onto a representaof surface structure.

Further evidence against the deep-strucaccount comes from Bock and Loebell (199who found that sentences containing a locaby-phrase such as “The foreigner was loiterby the broken traffic light” primed passive dscriptions involving an agentive by-phrase ju

as much as another agentive sentence did (but

STRUCTURE 3

e-n-

e-tice-

e-

nsn--

uc-ren-ofheaninta-

ndub-f agu- aon-ts.t

ofriesredg-ge-lts

r-on

re),eg-t

Potter & Lombardi, 1998). Apart from ruling outhematic accounts of priming, these resultsalso difficult to reconcile with transformationapproaches to syntax, which assume a veryferent representation for passive sentencesvolving transformed structures and traces) thfor locative sentences (Chomsky, 1981). In otwords, there is good evidence against a muple-stage account involving two fully specifieconstituent structure representations.

However, there are strong theoretical and perimental reasons to advocate an alternamultiple-stage account, which we call the domi-nance-only account. On this account, constituent structure is computed in two stages. first specifies the hierarchical aspects of cstituent structure but does not specify linorder. For “gave the hay to the horses,” the fistage would compute a representation coning of a verb phrase node that dominates a node, a noun phrase node, and a prepositiphrase node but would not specify that the vnode preceded the noun phrase node or thanoun phrase node preceded the prepositiphrase node. Note that this representation isa local tree because the precedence relationtween the daughter nodes are not specified (tee, ter Meulen, & Wall, 1990).

Following linguistic terminology, we say ththis intervening representation constitutesdominance-only level of representation. In otwords, it contains “dominance” informatioabout which phrases dominate others (e.gverb phrase node dominates verb, noun phrand prepositional phrase nodes) but not “predence” information about the order of phra(e.g., between the verb, noun phrase, and prsitional phrase nodes). This distinction is pposed within various linguistic theories, monotably modern theories of phrase structgrammar (Gazdar, Klein, Pullum, & Sag, 198Pollard & Sag, 1994). It allows, for exampparsimonious generalizations about word ordMany languages require particular constitueto occur at a specific point in a phrase (everbs come clause-finally in Japanese and Tish), and an autonomous precedence compoallows such a generalization to be made in a

cf.gle statement.

4 PICKERING, BRANIGAN, AND MCLEAN

hi &Ioan

arete

-

-

.

at

sio

au

wnu

s

enos

o

a

in-unast,

erb

nt,uc-mi-in-nlysistent theing ap-

in-ingtoin&-&n-ta-eceedn.a

ntsceivebyas

atem-l-ofre

-cein

n-eronSe-

-o-

Under the dominance-only account, tprocessor then converts this representation a second representation, which is specifiedorder. This second process has been calledlin-earization(Hartsuiker et al., 1999; Hartsuiker Westenberg, 2000; Vigliocco & Nicol, 1998). is compatible with computational models prposed by Kempen and Hoenkamp (1987) De Smedt (1990). To understand the differebetween the single-stage and dominance-oaccounts, let us consider the choices to be min the production of dative sentences. In fact,ternating dative verbs are compatible with thconstructions, namely PO, DO, and a shifconstruction illustrated in (1):

(1a) The racing driver showed the extremely dirty andbadly torn overall to the mechanic. (PO)

(1b) The racing driver showed the mechanic the extremely dirty and badly torn overall. (DO)

(1c) The racing driver showed to the mechanic the extremely dirty and badly torn overall. (shifted)

In a shiftedconstruction like (1c), the prepositional phrase precedes the noun phraseEnglish, such sentences occur consistentlycorpora and are acceptable but relatively rWasow (1997) estimated their occurrence in Brown corpus at 5.6% of eligible sentences. Toccurrence of shifted constructions appearsbe closely linked to the length and new informtion content of the direct object noun phra(hence the alternative name for the construct“heavy NP shift”). Indeed, when the nouphrase is longer than the prepositional phrand conveys new information, shifted constrtions may actually be produced more frequenthan PO constructions (Arnold, WasoLosongco, & Ginstrom, 2000). Shifted costructions have been the subject of both lingtic and psycholinguistic investigation (e.gArnold et al., 2000; Hawkins, 1994; StallingMacDonald, & O’Seaghdha, 1998).

According to standard analyses, all three stences have different (fully specified) costituent structures. In (1a), the verb phrase cprises a verb followed by a noun phrafollowed by a prepositional phrase (V NP PP);in (1b), it comprises a verb followed by twnoun phrases (V NP NP); and in (1c), it com-prises a verb followed by a prepositional phrfollowed by a noun phrase (V PP NP). But (1a)

entofor

t-ndcenlyadel-ed

- Inin

re.hehe toa-en,

nsec-tly,-is-.,,

n--m-e

se

and (1c) share dominance relations; both volve a verb phrase dominating a verb, a nophrase, and a prepositional phrase. In contr(1b) involves a verb phrase dominating a vand two noun phrases.

Figure 1a illustrates the single-stage accouwhere a simple choice among the three strtures is made. Figure 1b illustrates the donance-only account, where the first stage volves selecting between a dominance-orepresentation consistent with the DO analyand a dominance-only representation consiswith the PO and shifted analyses and wheresecond (linearization) stage involves selectbetween the PO and shifted analyses if thepropriate choice is made at the first stage.

We suggest that the dominance-only and sgle-stage accounts can be distinguished ussyntactic priming. Priming effects appearprovide evidence for levels of representationlanguage production (Branigan, Pickering,Cleland, 2000a; Branigan, Pickering, Liversedge, Stewart, & Urbach, 1995; PickeringBranigan, 1999). For one stimulus to prime aother, the stimuli must have related representions at some level of structure to which thcognitive system is sensitive. There is evidenthat priming is sensitive to representations usat both early and late stages in formulatioBock et al. (1992) found that after producingsentence with an animate subject, participawere more likely to produce another sentenwith an animate subject. For example, a passsentence with an animate subject was primedan active sentence with an animate subjectcompared to an active sentence with an animobject. These results provide evidence for priing of the mapping of conceptually specified eements to grammatical functions at a stageformulation that precedes constituent structugeneration.

Priming effects during single-word production have also provided substantial evidenabout the representations that are implicatedphonological encoding, following syntactic ecoding. For example, Roelofs and Mey(1998) demonstrated priming effects basednumber of syllables and stress pattern, andvald, Dell, and Cole (1995) demonstrated priming of the structure (as opposed to the phon

CONSTITUENT STRUCTURE 5

n

la

ren

i

r

onet itofc-ldofon-

ndofl. thesle

ac-NP,

logical content) of syllables (see also CostaSebastian-Galles, 1998). These experimesuggest the existence of a level of represention concerned with syllable structure (Sevaet al., 1995). Other evidence from priming hprovided support for the existence of represetations concerned with morphological structu(Roelofs, 1996) and perhaps grammatical gder (Jescheniak & Levelt, 1994; but cf. VaBerkum, 1997).

Thus, we propose that any level involvedconstituent structure encoding should be sentive to priming. More specifically, primingshould occur whenever an earlier level of rep

&ts

ta-dsn-en-

nsi-

e-

sentation can be mapped onto more thanlater level of representation. This means thashould be possible to prime the constructionconstituent structure in the single-stage acount. In the dominance-only account, it shoube possible to prime both the constructionthe dominance-only representation and the cstruction of the linearized representation.

Hartsuiker et al. (1999) and Hartsuiker aWestenberg (2000) found syntactic priming word order (in Dutch). In Hartsuiker et a(1999), participants tended to perseverate inproduction of “locative-inverted” sentencesuch as “Op de tafel ligt een bal” (“On the tab

FIG. 1. Single-stage and dominance-only accounts of constituent structure formulation. (a) Single-stage count. (b) Dominance-only account. PO, prepositional object; DO, double object; VP, verb phrase; V, verb; noun phrase; PP, prepositional phrase.

I

n

a

s

ngd

l

en

ve

h

n

lr

raters.9)on-ey

y of bechbut

)

nts.esanyionrstnsnds of

ediftedela-

re-ela-utla-ee-ime toeso aef-sesseshend, adPOnlyone-heg

effects due to repeating the same construction.

1 To make this more concrete, assume “baseline” weight-ings of .5 for PO/shifted and .5 for DO at the dominance-only level and .8 for PO and .2 for shifted at the linearizationlevel. This translates into .4 PO responses, .5 DO responses,and .1 shifted responses. If a shifted prime increases thePO/shifted weighting to .6 at the dominance-only level andthe shifted weighting to .4 at the linearization level, then theproportions become .36 PO responses, .40 DO responses,and .24 shifted responses. Thus, the shifted prime has de-creased the proportion of PO responses. It should be clearthat different assignments of weightings lead to different

6 PICKERING, BRAN

is a ball”) or their uninverted counterparts. Hartsuiker and Westenberg (2000), participatended to perseverate in the production of stences ending with a main verb followed by auxiliary or sentences ending with an auxiliafollowed by a main verb. Hartsuiker and Weenberg argued that the alternatives have same hierarchy of constituents and functiorelations and, in particular, that their findincannot be explained in terms of conceptual ferences in topic–comment order. They arguthat these word order priming effects imply ththe syntactic procedures associated with placwords in their appropriate positions are primeMore specifically, they claimed that their resudemonstrate priming from a dominance-onrepresentation to an ordered representation (during linearization).

An alternative explanation of these findingsthat the fully specified constituent structure reresentation is primed. Hartsuiker and Westberg (2000) argued that such an explanatioincompatible with priming effects obtained foactive/passive transitives (e.g., Bock, 1986). Atives and passives differ at the functional leso that if, for example, the conceptual represtation is of lightning striking a church, and if thsubject role is assigned to lightning and the ob-ject role is assigned to church, then the onlypossible sentence type is an active one suc“Lightning is striking the church.” In such caseconstituent structure is predetermined by futional relations. Thus, such priming effects apear to be due to the priming of functional retions rather than to constituent structuHowever, although their argument might eclude a constituent structure priming accoufor actives and passives, it does not demonstthat constituent structure priming never occuHence, the findings of Hartsuiker et al. (199and Hartsuiker and Westenberg (2000) are csistent with both a multistage account, as thargued, and a single-stage account. One wadistinguishing between these accounts wouldto determine whether two forms prime eaother when they share hierarchical relations differ in word order.

In other work, Hartsuiker and Kolk (1998

presented results that are compatible with b

GAN, AND MCLEAN

Ints

en-n

ryt-thealsif-edatingd.tslyi.e.,

isp-n- isrc-eln-

e

ass,c-p-a-e.x-nt

the single-stage and dominance-only accouThey employed PO, DO, and shifted prim(“medial datives,” in their terminology) and baseline condition that should not prime aform of the test sentence. The baseline conditused intransitive primes and constituted the fitrials in the experiment. Shifted constructioare more common in Dutch than in English, aso participants produced reasonable numberPO, DO, and shifted target responses.

According to the single-stage account, shifttarget responses should be increased by a shprime relative to a PO, DO, or baseline primbut should not be increased by a PO prime retive to the baseline prime; and PO target sponses should be increased by a PO prime rtive to a shifted, DO, or baseline prime bshould not be increased by a shifted prime retive to the baseline prime. According to thdominance-only account, shifted target rsponses should be increased by a shifted prrelative to a PO, DO, or baseline prime (duelinearization priming); and PO target responsshould be increased by a PO prime relative tshifted, DO, or baseline prime. However, the fects of a PO prime on shifted target responand of a shifted prime on PO target responcannot be predicted without knowledge of tstrength of priming at the dominance-only alinearization levels. Under some assumptionsPO prime will increase the proportion of shiftetarget responses and vice versa. But if a prime has a weak effect at the dominance-olevel but a very strong effect at the linearizatilevel, then the proportion of shifted target rsponses will actually be reduced relative to tbaseline.1 Thus, both accounts predict primin

othpatterns of responses.

T

ri

lksaoes

wess mw

aeO

Dhen tr anina

op a

n2caanmra

ticndiblearemayng-i-uc-esves.uners

n-x-O

r-m-ub-

edods

i-k--hated.lltac-ei-ask

asing thele-thearywithar-thele-uld re- the

CONSTITUEN

The single-stage account predicts no primingshifted responses by PO primes or vice veThe dominance-only account is compatible wpositive or negative priming or no priming.

In two experiments, Hartsuiker and Ko(1998) found increased shifted target responfollowing a shifted prime and increased PO tget responses following a PO prime (in accwith both accounts). In one experiment, thfound a numerical increase in shifted responfollowing a PO prime versus a baseline primbut in the other experiment they did not. In texperiments, they found a numerical increasPO responses following a shifted prime vera baseline prime. If these numerical increaare significant (no statistics were conductedthese comparisons), then this provides sosupport for the dominance-only account. Hoever, Hartsuiker and Kolk’s baseline primeswere produced at the beginning of the expment, and the combined proportions of PDO, and shifted target responses after PO, and shifted primes were considerably higthan after baseline primes (15% in Experim1 and 9% in Experiment 2). Thus, participamay have been primed through the coursethe experiment to produce one of the three get response categories (rather than othesponses), or they may have been primed stage in production in which PO, DO, ashifted responses do not differ relative to transitive primes to produce one of these cgories of response.

In summary, two multiple-stage accountsconstituent structure generation have been posed. The deep-structure account, whichcompatible with proposals in transformationgrammar, is difficult to reconcile with the sytactic priming findings of Bock et al. (199and Bock and Loebell (1990). The dominanonly account is also compatible with proposin theoretical linguistics, although its inspirtion comes from nontransformational accousuch as generalized phrase structure gramSome experimental and computational search is compatible with this proposal, though Hartsuiker and Kolk’s (1998) findingcan also be interpreted in terms of the sing

stage account.

STRUCTURE 7

ofsa.th

esr-rdyese,o inuseson

e-llri-,

O,ernttsof

ar-re-t ad-

te-

fro-isl

-)e-ls-tsar.

e-l-sle-

Rationale for Experiments

At this point, it appears hard to use syntacpriming to distinguish the dominance-only asingle-stage accounts. But it might be possto do so by considering constructions that very rarely used as target responses. This be the case with the shifted construction in Elish. Stallings et al. (1998) found that particpants produced shifted sentences in a prodtion task in which they were told which phrasto use but could choose the order themselHowever, their task used particularly long nophrases that are known to induce shifted ordfairly often (Hawkins, 1994). In tasks that ivolve freer production, shifted forms may be etremely rare in comparison to PO and Dforms.

Our experiments used free production for taget responses. Participants simply had to coplete sentences after being presented with a sject and verb (e.g., “The racing driver show. . .”). Previous studies using the same methhave not led to the production of shifted form(Branigan, Pickering, & Cleland, 1999; Brangan, Pickering, Stewart, & McLean, 2000b; Picering & Branigan,1998). However, we constrained production of prime sentences so tshifted sentences would regularly be producIt is possible that including a shifted prime wiinduce some shifted target responses via syntic priming, but probably fairly infrequently (seExperiments 2-4); note in particular that particpants undertaking the sentence completion tusually produce short completions.

If people produce shifted constructions prime responses but the probability of producshifted target responses is close to zero, thenpredictions of the dominance-only and singstage accounts become clearly distinct. In single-stage account, there is simply a binchoice between PO and DO target response, a PO prime increasing the proportion of PO tget responses and a DO prime increasing proportion of DO target responses. The singstage account predicts that shifted primes shohave no effects on the production of PO targetsponses relative to a baseline. In contrast,

dominance-only account predicts that both PO

IG

d

p

t

eloPt

nt1ee

li

nimtfthnin s

eeh

Pn

ina

tin

c-oysal.,eed an-ac-O,r-di-linet 3thanble

n et

ty

m-x).

heimeple-

-ts. Ine

se,, aun. thefor

8 PICKERING, BRAN

and shifted primes should increase the likelihothat the PO/shifted representation is activatethe dominance-only level to a similar extent. Bcause shifted responses are hardly producedsuch activation should proceed to the PO resentation at the linearized level.2 Therefore, bothPO primes and shifted primes should prime production of PO responses to a similar exten

Thus, evidence from syntactic priming expiments may allow us to distinguish the singstage and dominance-only accounts. If the dinance-only account is correct, then the prime (1a) and the shifted prime (1c) have same representation at the dominance-olevel. Thus, according to the dominance-oaccount, (1a) and (1c) both should prime production of PO sentences, whereas (should prime the production of DO sentencThis prediction holds whether priming is biasor balanced.

In contrast, the single-stage account impthat shifted sentences and PO sentences doshare a common representation. Hence, uthe single-stage account, (1c) should not prPO target responses (or, of course, DO targesponses). If priming is balanced, then the shiprime (1c) should behave differently from tPO prime (1a) and the DO prime (1b) ashould behave more similarly to a baselprime than either of these. Thus, if there areother (nonsyntactic) priming effects on theconstructions, then the shifted prime shouldindistinguishable from a baseline prime.

Note, however, that if priming is biased, thno difference is necessarily predicted betwthe shifted prime and the PO prime. If only tDO prime had any effect on target productiothen the shifted prime should behave like theprime, just as in the dominance-only accouThus, a prerequisite for using priming to distguish the single-stage and dominance-only counts is that the PO construction can primed, as predicted by the balanced accoun

Hence, our first purpose was to determwhether priming is balanced. Experiment

therefore, uses PO, DO, and (intransitive) ba

2 Formally, the weightings at the linearization level aclose to 1 for PO and 0 for shifted.

AN, AND MCLEAN

od ate-, allre-

het.r-e-m-O

henlylyhe

line primes to investigate whether priming ocurs for both PO and DO structures and emplwritten sentence completion (Branigan et 1999; Pickering & Branigan, 1998). On thbasis of Experiment 1, which shows balancpriming effects, Experiments 2 to 4 includeshifted prime condition, which allows us to cotrast the single-stage and dominance-only counts. Experiment 2 directly contrasts PO, Dand shifted primes in written production. Expeiment 3 adds a baseline condition to allow a rect comparison between shifted and baseprimes. Experiment 4 is similar to Experimenexcept that responses are spoken rather written and that participants repeat the preamas well as complete the sentence (Braniga

s on-sg-r

b)s.d

es notdere

re-edede

noe

be

nenen,Ot.-c-

be.e

1,e-

al., 2000b).

EXPERIMENT 1

Participants

A total of 27 participants from the Universiof Glasgow community took part.

Items

We constructed 24 sets of items. Each coprised two sentence fragments (see Appendi

(2a) The racing driver showed the torn overall . . .(PO-inducing prime)

(2b) The racing driver showed the helpful mechanic . . .(DO-inducing prime)

(2c) The racing driver fainted . . . (baseline prime)(3) The patient showed . . .

The first fragment (2a–c) was the prime; tsecond fragment (3) was the target. The prfragment was designed to induce a PO comtion, a DO completion, or an intransitive completion (the baseline condition). All fragmencontained a subject noun phrase and a verbthe PO- and DO-inducing conditions, the primfragment contained a postverbal noun phracomprising a determiner followed by a nounnoun compound, or an adjective and a noThis phrase always had the same verb andsame syntactic structure across conditions these two versions of an item. The baseline cdition employed a different verb, which wastandardly used intransitively. The target frae

ment consisted of a subject noun phrase and a

T

a

leisar

t

ab iutu

th

a

hdee eo

aoso bme

get etn,-

it isr

enlly

tri-

lethatpar-ny asedbler-r,

teri-ri-

tici-.

n.s a

hers,

, orivesey a

s ifn

by ae. Inf a

Af itasc-

CONSTITUEN

verb, which was the same verb as in the PO-DO-inducing conditions of the prime.

For the PO- and DO-inducing conditions, wmanipulated the postverbal noun phrase in prime fragment to induce PO or DO comptions. In (2a), the postverbal noun phrase plausible patient but an implausible beneficifor the action denoted by the verb. Therefoparticipants should be likely to complete thefragments using the PO construction, wherepostverbal noun phrase is the patient of the v(e.g., “The racing driver showed the torn overto the team manager”). In (2b), the postvernoun phrase is a plausible beneficiary but anplausible patient. Therefore, participants shobe likely to complete these fragments using DO construction, where the postverbal nophrase is the beneficiary of the verb (e.g., “Tracing driver showed the helpful mechanic damaged tyre”). Sentences such as (2c) weresigned so that participants could produce intrsitive sentences, but obviously a range of copletions were possible (actual examples baseline completions include “The grandmotached all over” and “The car salesman snorehe slept”) The experimental items employseven verbs (see Appendix). Previous expments showed that participants were likelyproduce both PO and DO completions for thverbs without producing a high proportion other constructions.

The experimental items were placed inthree lists, each comprising eight items froeach condition, such that one version of eitem appeared in each list. In addition, we cstructed 96 filler fragments (48 noun phrafollowed by a verb and 48 noun phrases flowed by a verb and a noun phrase). None offiller fragments contained a verb that could completed with a PO or DO construction. Soverbs appeared in more than one filler fragm

We constructed 27 nine-page booklets of 1fragments consisting of 48 experimental framents (i.e., 24 items) and the 96 filler fragmenEach page (except the last page) containedfragments. The order of fragments was individally randomized for each booklet, with the costraint that at least 3 filler fragments interven

between experimental items. This means th

STRUCTURE 9

nd

ethe- arye,seheerbllalm-ldhen

hee

de-n-

m-ofer asdri-tosef

tomchn-esl-

theee

nt.44g-ts. 20u-n-ed

there were at least four trials between one tarcompletion and the next. Note that Braniganal. (1999), using written sentence completiofound no long-term priming when four trials intervened between prime and target, and so highly unlikely that any effects would carry ovefrom one trial to another in this paradigm. Evif they did, baseline trials were preceded equaoften by PO and DO (and previous baseline) als.

Instructions on the front page of the bookexplained that we were interested in seeing wsorts of sentences people produce and that ticipants should complete the fragments in away they liked, ensuring that they producedgrammatical sentence. The instructions strescompleting each fragment as quickly as possiwith the first completion that came to mind. Paticipants were told to fill in the booklet in ordewithout leaving out any fragments.

Procedure

Participants were given a booklet to compleand were told to hand it back to the expementer when they were finished. The expementer answered any questions that the parpants had. The experiment took about 25 min

Scoring

We always scored the first legible completioEach baseline prime completion was scored abaseline (i.e., none was excluded). For all otprime completions plus all target completionthe completions were scored as POs, DOs“others.” They were scored as POs if the datverb was immediately followed by a noun phrathat acted as the patient or theme and then bprepositional phrase beginning with to that actedas the beneficiary. They were scored as DOthe verb was immediately followed by a nouphrase that acted as the beneficiary and then noun phrase that acted as the patient or themaddition, the dative verb could not form part ophrasal verb (e.g., handed overin “The architecthanded the latest plan over to the builder”).prime completion was scored as a PO only icompleted a PO-inducing fragment and wscored as a DO only if it completed a DO-indu

ating fragment. For example, if a participant com-

I

Pt wats

mti

amlnio

no

a

elebinwe

e

inele-m- for ofseantses.

de-

ng ofthehe

uree-

ons

line ofm-%

getm-

teres.

This trend was not significant, F1(2,52) 5 2.61,

3 In fact, this is equivalent to computing the number of POtarget completions divided by the sum of the number of POtarget completions and the number of DO target comple-tions. Our discussion is in terms of proportions because themeans reported all are in terms of proportions. The use ofPO rather than DO is arbitrary because the proportions arecomplementary. Note that the (overall) proportion of PO tar-get responses5 PO target ratio3 (1 2 others) and that theproportion of DO target responses5 (1 2 PO target ratio)3

Note. PO, prepositional object; DO, double object.

10 PICKERING, BRAN

pleted a DO-inducing prime fragment as a (e.g., completing “The headmaster gives naughty pupils” with “to his assistant”), thenwas scored as an other. A target completion scored as a PO or DO only if it had a grammcal alternative in the other category where order of the patient and beneficiary was reverAll other completions were scored as others.

Design and Data Analysis

Every participant completed 24 target framents, 8 in each of the three priming conditiodefined by the three levels of the prime comption factor (PO vs DO vs baseline prime copletion). Every experimental item was presento all 27 participants, with 9 participants seeany one version of an item.

We first analyzed the results for the other tget completions to determine whether the cobined proportion of PO and DO target comptions was comparable across primiconditions. Thus, we compared the proportof other completions following PO prime completions, the proportion of other completiofollowing DO prime completions, and the prportion of other completions following baselinprime completions. We computed the relevproportions by dividing the number of other taget completions following PO prime compltions by the total number of PO prime comptions (i.e., PO prime completions followed other, PO, and DO target completions), dividthe number of other target completions folloing DO prime completions by the total numbof DO prime completions (i.e., DO prime completions followed by other, PO, and DO targcompletions), and dividing the number of othtarget completions following baseline primcompletions by the total number of baselprime completions (i.e., baseline prime comptions followed by other, PO, and DO target copletions). These proportions were calculatedeach participant and for each item. Analysesvariance (ANOVAs) were performed on thedata, with separate analyses treating particip(F1) and items (F2) as random effects. Thanalyses were within-subjects and within-item

We then computed a measure designed to

termine the relative proportions of PO versu

r--

e-gn-s-entr---

yg-r-etr

e

DO target completions in each of the primiconditions. This measure was the proportionPO target completions divided by the sum of proportion of PO target completions and tproportion of DO target completions.3 We callthis the PO target ratio. We employ this measbecause it allows us to compare priming btween conditions in cases where the proportiof other completions are not equivalent.

Results and Discussion

Participants produced POs, DOs, or basecompletions for the prime fragments on 92%trials (596 trials); of these, 30% were PO copletions, 34% were DO completions, and 36were baseline completions.

We considered the proportions of other tarcompletions first. Table 1 shows that such copletions are numerically more frequent afbaseline primes than after PO or DO prim

GAN, AND MCLEAN

Oheitasti-heed.

g-nsle--

edng

TABLE 1

Results

Experiment Prime completion Target completion

PO target ratio Other

1 PO .76 .31DO .51 .29Baseline .62 .37

2 PO .72 .27DO .46 .26Shifted .61 .32

3 PO .68 .28DO .44 .31Shifted .60 .32Baseline .61 .40

4 PO .70 .21DO .46 .18Shifted .59 .26Baseline .62 .27

s(1 2 others).

CONSTITUENT

a

sae

me

thasef

n

o

s

ty

m-x).

eri-2c)onm-heis

er-dernotori-

b. A in

ceptg

p 5 .08, MSe5 .052; F2(2,46) 5 1.97, p 5 .15,MSe5 .047.

ANOVAs on the PO target ratio revealed effect of prime completion, F1(2,52) 5 12.07, p , .001, MSe5 .036; F2(2,46) 5 12.21, p ,.001, MSe5 .034. Newman–Keuls tests, treaing both participants and items as random fects, revealed that all three ratios differed froeach other (all ps , .05).

Experiment 1 demonstrates that priming itwo-way effect. PO target completions (reltive to DO target completions) were more frquent following PO prime completions thafollowing baseline prime completions anwere less frequent following DO prime completions than following baseline prime completions. In other words, both PO prime copletions and DO prime completions affectthe likelihood of producing a PO target completion on the subsequent trial. Indeed, magnitude of priming was similar in botcases: The overall 25% priming effect (PO tget ratio following PO prime completionminus PO target ratio following DO primcompletions) was due to a combination o14% priming effect of PO prime completion(relative to baseline prime completions) aan 11% priming effect of DO prime completions.

Experiment 1 also shows an intriguing, bnonsignificant, trend toward a higher proportiof other target completions following baselinprimes than following either PO or DO prime

eoP

erdh

tione)

wasng.elset

uc- ex-low.

ted ofm-%

We return to this in Experiment 3.

EXPERIMENT 2

Experiment 2 investigated whether shiftprimes, containing a prepositional phrase flowed by a noun phrase, behaved like primes. Because Experiment 1 supported banced priming, such a finding would providstrong evidence for the dominance-only acount. However, it is also possible that shiftprimes do not behave like PO primes. Thefore, this experiment employed three contions, just like Experiment 1, but replaced tbaseline prime condition by the shifted primcondition. It also included PO and DO prim

conditions.

STRUCTURE 11

n

t-ef-m

a--

nd---d-

he

r-

asd-

utne.

dl-Oal-ec-de-i-eee

Participants

A total of 30 participants from the Universiof Glasgow community took part.

Items

We constructed 24 sets of items. Each coprised two sentence fragments (see Appendi

(2a) The racing driver showed the torn overall . . .(PO-inducing prime)

(2b) The racing driver showed the helpful mechanic. . . (DO-inducing prime)

(2d) The racing driver showed to the helpful me-chanic . . . (shifted-inducing prime)

(3) The patient showed . . .

The items were the same as those in Expment 1 except that the baseline condition (was replaced by the shifted prime conditi(2d), which was intended to induce shifted copletions (e.g., “The racing driver showed to thelpful mechanic the serious problem with hcar that had developed”). The critical charactistic of these primes is that they induce the orverb, prepositional phrase, or noun phrase, that the noun phrase is especially long “heavy” (see scoring below). In all three condtions, prime and target shared the same vertotal of 96 filler fragments were constructedthe same way as in Experiment 1.

Procedure, Scoring, and Design and DataAnalysis

These were the same as Experiment 1 exin the following respects. A shifted-inducinprime was scored as a shifted prime compleif the completion contained a patient (or themnoun phrase (and other otherwise). There no requirement for this noun phrase to be loThe prime completion factor had three lev(PO vs DO vs shifted prime). Note that targcompletions that employed the shifted constrtion were scored as others. Such cases weretremely rare and are discussed separately be

Results and Discussion

Participants produced POs, DOs, or shifcompletions for the prime fragments on 94%trials (680 trials); of these, 31% were PO copletions, 35% were DO completions, and 34

were shifted completions. Two cells on the par-

I

o

f

f

eh

r

ta

i

s

ihe

o

m-x).

ri-nsotrb.as8

unse,

oa

dseds-dthatn-

a(orThevsok

12 PICKERING, BRAN

ticipants analysis in the DO prime completicondition were empty and, therefore, were placed by the grand mean.

The proportions of other completions did ndiffer across conditions (both Fs , 1.2). ANOVAson the PO target ratio revealed a significant efof prime completion, F1(2, 58) 5 9.59, p , .001,MSe5 .052; F2(2, 46) 5 17.64, p , .001, MSe5.032). Newman–Keuls tests revealed that all thratios differed from each other (p , .05) exceptthat the difference between the PO and shiprime conditions was marginal on the participaanalysis (p 5 .065). Participants produced shifttarget responses on three occasions, all of tfollowing a shifted prime (with two being fromthe same participant).

These results demonstrate that shifted primdo not produce the same proportion of PO tacompletions as do PO primes. In other wordsleast some component of the representationPO sentences that brings about syntactic primis not replicated in the shifted prime sentencTherefore, this is compatible with the claim thno dominance-only representation is construcHowever, it is possible that both the intermedirepresentation and a final ordered representaproduce priming. In that case, we would expecfind a priming effect in the shifted prime condtion but reduced relative to the PO prime contion. This explanation is unlikely for two reasonFirst, the PO target ratio in the shifted condit(61%) was roughly halfway between the PO tget ratios in the PO and DO conditions (72% a46%, respectively). Second, the pattern was vsimilar to that in Experiment 1, where the baline condition also fell roughly halfway betweethe other conditions. Thus, the shifted conditin Experiment 2 behaved very similarly to tbaseline condition in Experiment 1. But to detmine whether baseline primes and shifted primproduce equivalent PO target ratios, it is necsary to compare shifted and baseline conditi

t

s,ns astedellsime

within a single experiment.

EXPERIMENT 3

Participants

A total of 60 participants from the Universi

of Glasgow community took part.

GAN, AND MCLEAN

nre-

ot

ect

ree

tedntsdem

esget, at ofinges.ated.te

tiont toi-di-s.onar-nderye-noner-es

es-ns

y

Items

We constructed 32 sets of items. Each coprised two sentence fragments (see Appendi

(2a) The racing driver showed the torn overall . . .(PO-inducing prime)

(2b) The racing driver showed the helpful mechanic. . . (DO-inducing prime)

(2c) The racing driver sneezed very . . . (baselineprime)

(2d) The racing driver showed to the helpful mechanic. . . (shifted-inducing prime)

(3) The patient showed . . .

The items were based on those in Expements 1 and 2 and employed all four conditiowithin a single experiment. In (2a,b,d) but nin (2c), prime and target shared the same veThe baseline prime included an adverb suchvery after the verb. There were 128 fillers (4noun phrases followed by a verb, 56 nophrases followed by a verb and a noun phra8 noun phrases followed by a verb in anit-cleftconstruction [e.g., “It was the park warden whspotted . . .”], 8 noun phrases followed byverb in awh-cleft construction [e.g., “What thekid hated . . .”], and 8 noun phrases followeby a verb and a noun phrase in an extrapoconstruction [e.g., “The waiter insulted the cutomer yesterday who . . .”]). We includethese somewhat unusual sentence types sothe shifted-inducing primes would be less cospicuous.

Procedure, Scoring, and Design and DataAnalysis

A shifted-inducing prime was scored asshifted if the completion contained a patient theme) noun phrase (and other otherwise). prime completion factor had four levels (PO DO vs shifted vs baseline). The experiment toabout 35 min.

Results and Discussion

The prime was completed as POs, DOshifted, or baseline on 92% of all completio(1,759 trials); of these, 22% were completedPO primes, 25% as DO primes, 26% as shifprimes, and 27% as baseline primes. Two con the participants analysis, one in the PO pr

completion condition and one in the DO prime

T

r

re

la

n

iod

thO

e-c

rn

aenhf

ael

liohhfoce

b rl

on-mees,

nslinets.edthisets.thisingmsn-

m-in atlym-ns

s. aderb as

g-oneasine-ncein-

thin-ndd)ti-de-er, theer

m-in ated

e-

CONSTITUEN

completion condition, were empty and, thefore, were replaced by the grand mean.

For the other target analyses, ANOVAs vealed a main effect of prime completion, F1(3,177) 5 6.83, p , .001, MSe5 .024; F2(3,93) 56.05, p , .001, MSe 5 .015. In accord withTable 1, Newman–Keuls tests showed that pticipants produced more other target comptions following baseline prime completions thfollowing PO, DO, or shifted prime completion(all ps , .02) but that the other three conditiodid not differ (all ps . .19).

Table 1 also shows that the PO target ratdiffer across conditions. ANOVAs confirmethis observation,F1(3,177)5 14.18,p , .001,MSe5 .042;F2(3,93)5 13.88,p , .001,MSe5.026. Planned comparisons demonstratedshifted prime completions differed from Pprime completions,F1(1,59) 5 4.44, p , .05,MSe5 .039;F2(1,31)5 4.83,p , .05,MSe5.027, but they did not differ from baseline primcompletions (bothFs , 1.0). Participants produced a shifted target response on two ocsions: one following a shifted prime and one folowing a DO prime. In accord with the pattefound in Experiments 1 and 2, shifted primcompletions did not appear to prime in the wthat PO prime completions did, and shiftprime completions behaved similarly to baseliprime completions. Thus, there was no sign tshifted prime completions served as a primePO target completions.

The analysis of other completions demostrated that participants produced more PO DO target completions (combined) after thhad produced PO, DO, or shifted prime comptions than after they had produced baseprime completions. This appears to be a secpriming effect that is not directly related to tsyntactic priming effect demonstrated by ttendency to produce PO target completions lowing PO prime completions and to produDO target completions following DO primcompletions. Instead, some property sharedPO and DO target completions is primed by PDO, and shifted prime completions but not baseline prime completions. It is impossiblebe certain of the source of the priming. Clea

the PO, DO, and shifted primes share ma

STRUCTURE 13

e-

-

ar-e-nss

s

at

a-l-

eydeator

n-ndye-nendeel-e

byO,ytoy,

properties (related to the fact that they can cstitute alternative ways of describing the saevent). This contrasts with the baseline primwhich describe very different kinds of events.

Thus, PO, DO, and shifted prime completioinvolved three arguments, whereas basecompletions did not involve three argumenThe PO and DO target completions involvthree arguments. Hence, it is possible that second priming effect reflects priming of thproduction of a particular number of argumenHowever, the current analyses do not test precisely because the definitions for scorprime completions were not expressed in terof number of arguments. Therefore, we coducted argument analyses, in which we re-scored a prime completion as a PO if it copleted a PO-inducing fragment and resulted sentence containing one main verb with exacthree arguments. For example, we allowed copletions containing phrasal verbs, completiothat involved prepositions apart from to, andcompletions involving different thematic roleSimilarly, a prime completion was scored asDO if it completed a DO-inducing fragment anresulted in a sentence containing one main vwith exactly three arguments and was scoreda shifted if it completed a shifted-inducing frament and resulted in a sentence containing main verb with exactly three arguments. It wscored as a baseline if it completed a baselinducing fragment and resulted in a sentecontaining a single main verb that did not volve three arguments.

Second, the priming conditions differed wirespect to the number of entities that were volved in the event described. PO, DO, ashifted prime completions (as originally scoretypically described events involving three enties, whereas baseline completions typically scribed events involving one entity. Howevthere were some exceptions to this becauseoriginal scoring was not in terms of the numbof entities involved. In the entity analyses, wescored a prime completion as a PO if it copleted a PO-inducing fragment and resulted sentence containing one main verb associawith exactly three entities. An entity was d

nyfined in terms of a head noun. For example,

IG

lt

.

n

et

n

edi

mif l

g

ta

r

e a

,

ashethe

es it ishe

3led,

e.ec-ca- beck intici-ingme-s toavefi-ldci-anri-

beetewe.

ntwasro-imeureat ad 4kenro- this0).

be-ani-it-nceshey

14 PICKERING, BRAN

“The racing driver showed the torn overall thcost him the race during the interview” woucount as a PO because the torn overall that coshim the raceinvolves one head noun and duringthe interviewcontains the head noun interview.Note that the entities could be abstract (etemporal expressions). The completion wscored as a DO if it completed a DO-inducifragment and resulted in a sentence containone main verb associated with exactly three tities and was scored as a shifted if it complea shifted-inducing fragment and resulted insentence containing one main verb associawith exactly three entities. It was scored asbaseline if it completed a baseline-inducifragment and resulted in a sentence containinsingle main verb that was not associated wexactly three entities.

Both argument and entity analyses producalmost identical results to the main analysThe proportions of other responses barely fered from the main analyses, and the statistsignificance of all effects remained the samThus, Experiment 3 demonstrated two differetypes of priming. The effect of the form of thprime on the PO target ratio is a syntactic priing effect. Within this general priming effect, showed that a shifted prime had the same efas a baseline prime and, therefore, that it didfacilitate the production of PO target comptions. But the experiment also showed that pticipants tended to produce PO or DO tarcompletions more often after PO, DO, or shiftprime completions than after baseline primcompletions. Further analyses suggested thaproduction of a prime either containing three guments or referring to three entities facilitatthe production of a PO or DO target, which cotained three arguments and referred to threetities. This priming effect appears to reflect pcessing before choice of syntactic analysis wmade, although it is impossible to localize teffect precisely. Therefore, it provides some idence for the suggestion that participantsHartsuiker and Kolk’s (1998) experiments mhave been primed to produce three-argumenthree-entity responses by a PO, DO, or shifprime versus a baseline prime. It is possible

course, that some other difference between

AN, AND MCLEAN

atd

g.,asgingn-

ed ated agg aith

eds.if-cale.nte

-tectnot

baseline condition and the other conditions wresponsible for the effect. For example, tbaseline primes used different verbs from targets, whereas the PO, DO, and shifted primused the same verb as the targets. However,unclear why this difference should affect tproportion of others.

A possible concern with Experiments 2 andis that participants might sometimes have faito notice the preposition to after the verb andthus, read the fragment as a DO-inducing primIf this were the case, then they would have efftively produced a DO completion on those ocsions and, hence, the DO structure wouldprimed. Therefore, it is conceivable that the laof a priming effect for the shifted sentencesExperiment 3 might have been because parpants sometimes completed the shifted-inducprime fragment as a shifted sentence and sotimes completed it as a DO sentence. For thihave happened, participants would have to hwritten a DO completion to a shifted prime sufciently often that the two priming effects wouhave “canceled out.” The possibility that partipants systematically misinterpret the prime cstraightforwardly be tested by using an expemental task in which the preposition to has toreproduced. Because writing out 160 complsentences is too laborious for participants,

e-ar-et

ede

ther-

edn- en-o-as

hev-iny

t orted of

turned to spoken production for Experiment 4

EXPERIMENT 4

Experiment 4 was a replication of Experime3 except that spoken sentence completion employed, participants were required to repduce the preamble as well as complete the prsentence, and a computerized timed procedwas used (as in Branigan et al., 2000b). Note thcomparison of the results of Experiments 3 anallows us to determine the extent to which spoand written sentence completion methods pduce similar results. Some evidence suggeststo be the case (Hartsuiker & Westenberg, 200Other evidence suggests that the relationshiptween the methods may be more complex. Brgan et al. (1999) found strong priming with wrten sentence completion so long as no senteintervened between prime and target, but t

thealso found that priming rapidly decayed if even

T

neso

ar

ty

l

inetsinhnecen

henioth e

eTfrlle

t

b

if the

Thevs

Os,876POtedll onionre-

re-le-

ol-erse-er

fferb-

c-onsdif-

r

tsr toid,

arlys nos a

nte-ons:

CONSTITUEN

one sentence intervened between prime andget. In contrast, Branigan et al. (2000b) fouthat priming persisted over an intervening stence in spoken sentence completion. This reis in accord with Bock and Griffin (2000), whfound that syntactic priming persisted overmany as 10 trials in the (spoken) picture desction method (Bock, 1986).

Participants

A total of 32 participants from the Universiof Edinburgh community took part.

Items

The items in this experiment were identicathose in Experiment 3 (see Appendix).

Procedure

This experiment was presented usPsyScope software (Cohen, MacWhinnFlatt, & Provost, 1993).[AU4] Participanwere told that we were interested in seewhat sorts of sentences people produce. Twere instructed to read out loud the sentefragments and then complete the sentencany way they liked, ensuring that they produa grammatical sentence. Each experimetrial consisted of a fixation point (“1”) appear-ing at the side of the screen for 1,000 ms. Twas then replaced with a sentence fragmThe first letter of the first word of the fragmeappeared in the location where the fixatpoint had been. The fragment remained in screen for 7,000 ms. The screen was tcleared, and after a 1,000-ms delay, a beepcurred. A further delay of 1,000 ms occurrbefore the next item was presented.

Participants took part in a short practice periment before the experimental session. practice session consisted of 10 sentence ments similar in structure and length to the fiitems used in the experimental session. Thetire experiment took about 25 min and containtwo breaks, the duration of which was under participants’ control.

Scoring

The tape for each participant was transcri

and scored. Scoring was as in Experiment 3 e

STRUCTURE 15

tar-dn-ult

sip-

to

gy,

geyce ind

tal

isnt.tn

heenoc-d

x-heag-r

en-edhe

ed

cept that a response was scored as an other fragment was incorrectly repeated.

Design and Data Analysis

These were the same as in Experiment 3. prime completion factor had four levels (PO DO vs shifted vs baseline).

Results and Discussion

The primes were completed as POs, Dshifted, or baseline on 86% of all responses (trials); of these, 23% were completed as primes, 24% as DO primes, 26% as shifprimes, and 28% as baseline primes. One cethe items analysis, in the PO prime completcondition, was empty and, therefore, was placed by the grand mean.

For the other target analyses, ANOVAs vealed a marginal main effect of prime comption, F1(3,93) 5 2.45, p 5 .07, MSe 5 .027;F2(3,93) 5 2.54, p 5 .06, MSe 5 .027. Al-though there were numerically more others flowing baseline primes than following the othconditions, a planned comparison of the baline prime condition and the mean of the othconditions was not significant (both ps . .10).

Table 1 shows that the PO target ratios diacross conditions. ANOVAs confirmed this oservation, F1(3,93) 5 8.49, p , .001, MSe5.040; F2(3,93) 5 6.94, p , .001, MSe5 .035.Hence, the usual syntactic priming effect ocurred in this experiment. Planned comparisdemonstrated that shifted prime responses fered from PO prime responses, F1(1,31) 54.99, p , .05, MSe5 .040; F2(1,31) 5 4.41, p , .05, MSe5 .042, but they did not diffefrom baseline prime responses (both Fs , 1.0).In accord with the pattern found in Experimen1 to 3, shifted prime responses did not appeaprime in the way that PO prime responses dand shifted prime responses behaved similto baseline prime responses. Thus, there wasign that shifted prime responses served aprime for PO target responses.

The preposition to in the shifted prime wasmissed on five prime trials (with one participaaccounting for four of these trials). A shifted rsponse was given in the target on five occasi

x-two after a DO prime response (both from the

IG

rs

eihreigrnrrb

line noh

ysisre-edre-nsese-es. ofuc-ro-ncemi-or-

16 PICKERING, BRAN

same participant), two after a shifted prime sponse, and one after an other prime respon

In conclusion, the results of Experiment 4 acomparable to those of Experiment 3. The fects in this experiment cannot be due to partpants failing to notice the preposition in tshifted prime condition. The results, therefoprovide further support for the claim that shiftprimes behave similarly to baseline primes wrespect to the proportions of PO and DO tarresponses produced. They also lend suppothe claim that written sentence completion aspoken sentence completion (together with production of the preamble) produce compable results and, therefore, are likely to equally valid methods for using syntactic prim

aohEngeea

li

t

auo uosac

O

thaten-thison).tedto

andntsle-se-

hat ac-ins,

ofrlyle-earssesheynly

e, thely

ro-l ofodere- notes, Inne

ing to investigate language production.

GENERAL DISCUSSION

Experiment 1 demonstrated that priming wbalanced so that it was possible to prime bPO and DO forms in relation to a baseline (twas not completed as either a PO or a DO). periment 2 showed that shifted primes did behave like PO primes; production of PO tarcompletions was less common following shiftprime completions than following PO primcompletions. Experiment 3 confirmed thshifted prime completions behaved like baseprime completions with respect to the prodution of PO versus DO target completions. Eperiment 4 found similar results to those of Eperiment 3 when participants orally repeatedpreamble and completed the sentence.

Shifted target responses were extremely rpresumably because the construction is fairly common in English. They are more commwhen the final noun phrase is particularly long“heavy,” but such completions were not encoaged by the experimental methods used (in cparison, e.g., to the methods used by Stallingal., 1998). Hence, syntactic priming does not pear to be strong enough to cause this tendenbe systematically overruled. In Experiments 24, there were a total of 10 shifted target respon(produced by eight different participants). these 10 responses, 1 followed an other primesponse. Of the other 9 responses, 6 follow

shifted prime responses, 3 followed DO prime r

AN, AND MCLEAN

e-e.ref-

ci-ee,dthet

t tode-a-e-

sthatx-

otetd

tnec-x-x-he

re,n-norr-m- etp-y totosesf

re-ed

sponses, and none followed either PO or baseprime responses (but recall that there wasbaseline condition in Experiment 2). Althougthese small numbers make statistical analrisky, the data suggest that shifted target sponses are more common following shiftprime responses than following PO prime sponses and perhaps that shifted target respodo not occur more often following PO prime rsponses than following DO prime responsThese findings suggest that the productionshifted responses is affected by the prior prodtion of shifted responses but not by the prior pduction of PO responses. This is in accordawith the single-stage account but not the donance-only account. Note that the small proption of shifted target responses may suggest participants do not normally produce shifted stences in conditions comparable to those in experiment (e.g., isolated sentence completiIf so, then participants may be producing shifprime completions in a way that is dissimilar sentence production in the other conditions (elsewhere). However, the fact that participadid produce (grammatical) shifted prime comptions just as often as they did PO, DO, and baline completions provides good evidence tshifted sentences are unusual but completelyceptable (as argued in other work, e.g., Hawk1994; Stallings et al., 1998).

However, our data show that the likelihoodproducing a shifted target response is neazero. The data are compatible with the singstage account because the shifted prime bno special relationship with PO target responany more than with DO target responses. Tare hard to reconcile with the dominance-oaccount because the effect of a shifted primlike that of a PO prime, should be to increaseactivation of the PO/shifted dominance-onnode (Fig. 1a). Because the likelihood of pducing a shifted response is (nearly) zero, althe increased activation at the PO/shifted nshould benefit the production of PO target sponses. Thus, PO and shifted primes shoulddiffer in their effects on PO target responsand both should differ from baseline primes.fact, shifted primes behaved like baseli

e-primes rather than like PO primes.

T

hae

ch

tldr

l

tr

isim

oolsoca

uhth

ai

d

a

u-n-teds

b),n,

bi-h

itsge.-

ionanthisre-cticthe

ense,rb

ap-

ber

tedtic8)

nve

neyhees theiza-en-ate ad-tedecf.

ep-

CONSTITUEN

Together, the experiments reported in tpaper provide evidence for the single-stage count, whereby language production involvthe mapping of a pre-syntactic representationa representation that is fully specified syntacally. Our experiments specifically suggest tproduction does not involve the computationa dominance-only level of representation. Athough other multiple-stage accounts areprinciple possible, the results provide no reasfor adopting such models.

Our conclusion appears to run counter to conclusion of Vigliocco and Nicol (1998). Folowing Bock and Miller (1991), there is consierable evidence that the verb sometimes eneously agrees with non-head nouns withincomplex subject such as helicopters in “theflight of the helicopters.” Vigliocco and Nico(1998) found that such errors were just as comon when the subject noun phrase followed verb (e.g., “Were the flight of the helicoptesafe?”) as when it preceded the verb (e.g., “Tflight of the helicopters were safe”). From ththey inferred two separate stages in grammatencoding, with the first stage involving the coputation of hierarchical structure and functionrelations in the absence of ordering informatiOn their account, subject–verb agreement curs during the first stage. However, it is apossible that agreement may have been cputed before the computation of hierarchistructure, for instance, at the functional level sumed by Bock and Levelt (1994).

Combinatorial Nodes in the Lemma Stratum

So far, we have interpreted our results terms of the levels of representation employduring the production of utterances. The resalso shed light on production from a somewdifferent perspective, namely the nature of lexical representations accessed during prodtion. Roelofs (1992, 1993) and Levelt et (1999) argued that this lexical information represented at three levels or strata: a concep-tual stratum(containing semantically specifieconcepts), a lemma stratum(encoding syntacticinformation), and a word form stratum(encod-ing morphological and phonological inform

tion). Pickering and Branigan (1998) expande

STRUCTURE 17

isc-s

toti-atofl-inon

he--ro- a

m-heshe,

cal-

aln.c-om-als-

inedltsate

uc-l.s

-

on the model of the lemma stratum, in particlar, by characterizing the way in which it ecodes syntactic information that is associawith lexical entries. This information include(major) category information (e.g., noun, verfeatural information (e.g., number, persotense), and combinatorial information. Comnatorial information specifies the way in whica word can combine with other linguistic unto form possible expressions of the languaThus, a verb such as givecan combine with arguments (e.g., the man, the book, to the boy)that correspond to the participants in the actdenoted by the verb. Pickering and Branigproposed that the lemma stratum encodes information by means of lemma nodes, repsenting the base form of words, and syntaproperty nodes, which are connected to lemma nodes via labeled links.

For example, the lemma give is connected tothe syntactic category node verb and variousfeatural nodes such as present tense, past tsingular number, and plural number. The venode is activated whenever giveis activated, andthe relevant feature nodes are activated aspropriate. Thus, when the form givesis used in asentence, the present tense and singular numnodes are activated. The lemma give is alsolinked to combinatorial nodes that are activawhen the verb is used in a particular syntacconstruction. Pickering and Branigan (199proposed that the NP,NPnode is activated whegive is used in the DO construction (e.g., “githe boy a book”) and that the NP,PPnode is ac-tivated when give is used in the PO constructio(e.g., “give a book to the boy”). However, thdid not commit to a precise specification of tcircumstances under which particular nodwere activated. One issue concerns whethernodes correspond to traditional subcategortion frames or whether they are activated whever the verb is associated with appropriphrases, which can be either arguments orjuncts of the verb. Pickering and Branigan nothat Bock and Loebell’s (1990) findings providsome evidence for this latter account (but Potter & Lombardi, 1998).

The evidence against a dominance-only r

dresentation suggests that the nodes are speci-

IG

th

edio ue

.g.,e am-arsesast,m-i-

18 PICKERING, BRAN

fied for the order of phrases. This suggests there are three different nodes: NP,NP; NP,PP;and PP,NP. The processor appears to makstraightforward choice among the three noat a single stage during production. Selectof a verb lemma is associated with selectiona combinatorial node that mandates constrtion of a fully specified constituent structur

Hence, it appears that once people have c

4. The cricket player showed the ball/the umpire. Th

AN, AND MCLEAN

at

aesn

ofc-.

structed earlier levels of representation (efunctional structure), they are ready to makchoice about which construction to use. Priing affects the choice of structure. It appethat construction of syntactic structure takplace in a single stage. In this respect at lethe number of levels of representation eployed during sentence formulation is min

on-mized.

ter theprime,1).

little girl

ed.

. sailor

famous

d.

d.

ter theord

The

howed..

APPENDIX

Items for Experiments 1 and 2

For each item, the first sentence contains the PO-inducing prime before the slash and the DO-inducing prime afslash. The second sentence contains the baseline prime from Experiment 1. To reconstruct the shifted-inducing add the word to after the verb in the DO-inducing fragment (e.g., “The mother gave to the hungry baby ” in Item The final sentence contains the target fragment.

1. The mother gave the expensive toy/the hungry baby. The mother sneezed. The air hostess gave.2. The architect gave the latest plans/the cheerful engineer. The architect sneezed. The teacher gave.3. The lecturer gave the book/the professor. The lecturer coughed. The shopkeeper gave.4. The hostess handed the dessert/the guests. The hostess hiccupped. The news agent handed.5. The efficient secretary handed the long fax/the grumpy businessman. The efficient secretary sneezed. The

handed.6. The grandmother handed the big present/the little girl. The grandmother ached. The tennis fan handed.7. The millionaire loaned the valuable painting/the struggling artist. The millionaire coughed. The explorer loan8. The swimmer loaned the towel/the diver. The swimmer ached. The draftsman loaned.9. The woman loaned the rusty bike/the new neighbour. The woman fainted. The librarian loaned.

10. The man lent the lawnmower/the neighbour. The man fainted. The actor lent.11. The fashion designer lent the pink jacket/the famous journalist. The fashion designer itched. The diver lent.12. The car salesman lent the mini/the couple. The car salesman snored. The forest ranger lent.13. The booking clerk posted the last ticket/the young fan. The booking clerk hiccupped. The serial killer posted14. The blackmailer posted the incriminating photos/the sleazy journalist. The blackmailer coughed. The lonely

posted.15. The captain gave the spare life jacket/the old sailor. The captain ached. The bus driver gave.16. The disgruntled employee sent the long letter/the managing director. The disgruntled employee snored. The

novelist sent.17. The secretary sent the invoice/the manager. The secretary coughed. The boyfriend sent.18. The woman sent insurance claim/the insurance company. The woman hiccupped. The fan sent.19. The racing driver showed the torn overall/the helpful mechanic. The racing driver fainted. The patient showe20. The youngster showed the toy/the teacher. The youngster itched. The private detective showed.21. The lifeguard showed the lifebelt/the surfer. The lifeguard sneezed. The inventor showed.22. The cricket player showed the ball/the umpire. The cricket player ached. The car mechanic showed.23. The bank manager handed the cheque/the customer. The bank manager snored. The junior surgeon hande24. The builder lent the drill/the surveyor. The builder snored. The hairdresser lent.

Items for Experiment 3 and 4

For each item, the first sentence contains the PO-inducing prime before the slash and the DO-inducing prime afslash. The second sentence contains the baseline prime. To reconstruct the shifted-inducing prime, add the wtoafter the verb in the DO-inducing fragment (e.g., “The racing driver showed to the helpful mechanic” in item 1). final sentence contains the target fragment.

1. The racing driver showed the torn overall/the helpful mechanic. The racing driver sneezed very. The patient s2. The youngster showed the toy/the teacher. The youngster clapped extremely. The private detective showed3. The lifeguard showed the lifebelt/the surfer. The lifeguard yelled quite. The inventor showed.

e cricket player yelled very. The car mechanic showed.

remely.

ded.ws agent

geon

rer

osted.. The

The diver

ovelist

t.

daughter

aned. personnel

ld.

CONSTITUENT STRUCTURE 19

5. The efficient secretary handed the long fax/the grumpy businessman. The efficient secretary sneezed extThe little girl handed.

6. The grandmother handed the big present/the little girl. The grandmother yawned very. The tennis fan han7. The hostess handed the dessert/the guests. The news agent handed. The hostess snored rather. The ne

handed.8. The bank manager handed the cheque/the customer. The bank manager yelled extremely. The junior sur

handed.9. The captain gave the spare lifejacket/the old sailor. The captain snored extremely. The bus driver gave.

10. The mother gave the expensive toy/the hungry baby. The mother laughed quite. The air hostess gave.11. The architect gave the latest plans/the cheerful engineer. The architect laughed very. The teacher gave.12. The lecturer gave the book/the professor. The lecturer giggled rather. The shopkeeper gave.13. The millionaire loaned the valuable painting/the struggling artist. The millionaire yawned quite. The explo

loaned.14. The swimmer loaned the towel/the diver. The swimmer yelled quite. The draftsman loaned.15. The woman loaned the rusty bike/the new neighbour. The woman giggled very. The librarian loaned.16. The booking clerk posted the last ticket/the young fan. The booking clerk laughed quite. The serial killer p17. The blackmailer posted the incriminating photos/the sleazy journalist. The blackmailer laughed extremely

lonely sailor posted.18. The car salesman lent the mini/the couple. The car salesman grumbled very. The forest ranger lent.19. The man lent the lawnmower/the neighbour. The man snored rather. The actress lent.20. The fashion designer lent the pink jacket/the famous journalist. The fashion designer clapped extremely.

lent.21. The builder lent the drill/the surveyor. The builder sneezed quite. The hairdresser lent.22. The chairman sent the long letter/the managing director. The chairman yawned extremely. The famous n

sent.23. The accountant sent the invoice/the client. The accountant giggled quite. The boyfriend sent.24. The customer sent the insurance claim/the insurance company. The customer clapped quite. The fan sen25. The nurse showed the X-ray/the doctor. The nurse yawned extremely. The jeweler showed.26. The courier handed the parcel/the receptionist. The courier sneezed rather. The child handed.27. The chief librarian sent the reminder/the student. The chief librarian grumbled very. The thoughtful grand

sent.28. The young woman loaned the necklace/the teenager. The young woman grumbled rather. The motorist lo29. The researcher posted the detailed questionnaire/the eager journalist. The researcher snored rather. The

manager posted.30. The spy sold the stolen documents/the foreign diplomat. The spy grumbled rather. The shop assistant so

31. The florist gave the huge bouquet/the startled butler. The florist giggled rather. The pharmacist gave.

is t.

Rtrt

p

e

nna

pct

c-er

).or

a).

art,s-

n,n:

32. The receptionist lent the spare key/the busy ass

REFERENCES

Arnold, J. E., Wasow, T., Losongco, A., & Ginstrom, (2000). Heaviness vs. newness: The effects of stural complexity and discourse status on constituendering. Language, 76,28–55.

Bock, J. K. (1986). Syntactic persistence in language duction. Cognitive Psychology, 18,355–387.

Bock, J. K. (1989). Closed class immanence in sentproduction. Cognition, 31,163–186.

Bock, J. K., & Griffin, Z. M. (2000). The persistence of sytactic priming: Transient activation or implicit learing? Journal of Experimental Psychology: Gener;129,177–192.

Bock, J. K., & Loebell, H. (1990). Framing sentences. Cog-nition, 35,1–39.

Bock, J. K., Loebell, H., & Morey, R. (1992). From concetual roles to structural relations: Bridging the syntacleft. Psychological Review, 99,150–171.

Bock, J. K., & Miller, C. A. (1991). Broken agreement. Cog-

nitive Psychology, 23,45–93.

tant. The receptionist clapped very. The Coast Guard len

.uc-or-

ro-

nce

--l

-ic

Bock, J. K., & Levelt, W. J. M. (1994). Language prodution: Grammatical encoding. In M. A. Gernsbach(Ed.), Handbook of psycholinguistics(pp. 945–984).San Diego: Academic Press.

Branigan, H. P., Pickering, M. J., & Cleland, A. A. (1999Syntactic priming in written production: Evidence frapid decay. Psychonomic Bulletin & Review, 6,635–640.

Branigan, H. P., Pickering, M. J., & Cleland, A. A. (2000Syntactic coordination in dialogue. Cognition, 75B,13–25.

Branigan, H. P., Pickering, M. J., Liversedge, S. P., StewA. J., & Urbach, T. P. (1995). Syntactic priming: Invetigating the mental representation of language. Journalof Psycholinguistic Research, 24,489–506.

Branigan, H. P., Pickering, M. J., Stewart, A. J., & McLeaJ. F. (2000b). Syntactic priming in spoken productioLinguistic and temporal interference. Memory & Cog-nition, 28,1297–1302.

Chomsky, N. (1965). Aspects of the theory of syntax. Cam-

bridge, MA: MIT Press.

20 PICKERING, BRANIGAN, AND MCLEAN

Chomsky, N. (1981). Lectures on government and binding.Dordrecht, Netherlands: Foris.

nc

:

s

o

e

r

y

Partee, B. H., ter Meulen, A., & Wall, R. E. (1990). Mathe-matical methods in linguistics. Dordrecht, Netherlands:

ta- in

m--

-

in

ma

y of

of

in

nd

bles or

.ro-

vy-

ech

la-im-

Cohen, J. D., MacWinney, B., Flatt, M., & Provost, (1993). Psyscope: A new graphic interactive enviroment for designing psychological experiments. Behav-ioral Research Methods, Instruments, and Compute,25,257–271.

Costa, A., & Sebastian-Galles, N. (1998). Abstract phological structure in language production: Evidenfrom Spanish. Journal of Experimental PsychologyLearning, Memory, and Cognition, 24,886–903.

Dell, G. S. (1986). A spreading activation model of retrievin sentence production. Psychological Review, 93,283–321.

De Smedt, K. J. (1990). IPF: An incremental parallel formlator. In R. Dale, C. Mellish, & M. Zock (Eds.), Cur-rent research in natural language generation (Cogntive Science Series, pp. 167–192). San DiegoAcademic Press.

Garrett, M. (1980). Levels of processing in speech prodtion. In B. Butterworth (Ed.), Language production(Vol. 1, pp. 177–220). London: Academic Press.

Gazdar, G., Klein, E., Pullum, G. K., & Sag, I. A. (1985Generalized phrase structure grammar. Cambridge,MA: Harvard Univ. Press.

Hartsuiker, R. J., & Kolk, H. H. J. (1998). Syntactic persience in Dutch. Language and Speech, 41,143–184.

Hartsuiker, R. J., Kolk, H. H. J., & Huiskamp, P. (1999Priming word order in sentence production. QuarterlyJournal of Experimental Psychology, 52A, 129–147.

Hartsuiker, R. J., & Westenberg, C. (2000). Persistenceword order in written and spoken sentence productiCognition, 75B,27–39.

Hawkins, J. A. (1994). A performance theory of order andconstituency. Cambridge, UK: Cambridge Univ. Press

Jescheniak, J. D., & Levelt, W. J. M. (1994). Word frquency effects in speech production: Retrieval of sytactic information and of phonological form. Journal ofExperimental Psychology: Learning, Memory, anCognition, 20,824–843.

Kempen, G., & Hoenkamp, E. (1987). An incremental pcedural grammar for sentence formulation. CognitiveScience, 11,201–258.

Levelt, W. J. M. (1989). Speaking: From intention to articu-lation. Cambridge, MA: MIT Press.

Levelt, W. J. M., Roelofs, A., & Meyer, A. (1999). A theorof lexical access in speech production. Behavioral andBrain Sciences, 22,1–75.

J.n-

rs

o-e

:

al

u-

i-

uc-

).

t-

).

ofn.

.-n-

d

o-

Kluwer.Pearlmutter, D. M. (1983). Studies in relational grammar 1.

Chicago: Univ. of Chicago Press.Pickering, M. J., & Branigan, H. P. (1998). The represen

tion of verbs: Evidence from syntactic persistencewritten language production. Journal of Memory andLanguage, 39,633–651.

Pickering, M. J., & Branigan, H. P. (1999). Syntactic priing in language production. Trends in Cognitive Sciences, 3, 136–141.

Pollard, C., & Sag, I. A. (1994). Head-driven phrase structure grammar. Chicago: Univ. of Chicago Press.

Potter, M. C., & Lombardi, L. (1998). Syntactic priming immediate recall of sentences. Journal of Memory andLanguage, 38,265–282.

Roelofs, A. (1992). A spreading-activation theory of lemretrieval in speaking. Cognition, 42,107–142.

Roelofs, A. (1993). Testing a non-decompositional theorlemma retrieval in speaking: Retrieval of verbs. Cogni-tion, 47,59–87.

Roelofs, A. (1996) Serial order in planning the productionsuccessive morphemes of a word. Journal of Memoryand Language, 35,854–876.

Roelofs, A., & Meyer, A. S. (1998). Metrical structure planning the production of spoken words. Journal ofExperimental Psychology: Learning, Memory, aCognition, 24,922–939.

Sevald, C. A., Dell, G. S., & Cole, J. S. (1995). Syllastructure in speech production: Are syllables chunkschemas? Journal of Memory and Language, 34,807–820.

Stallings, L. M., MacDonald, M. C., & O’Seaghdha, P. G(1998). Phrase ordering constraints in sentence pduction: Phrase length and verb disposition in heaNP shift. Journal of Memory and Language, 39,392–417.

Van Berkum, J. J. A. (1997). Syntactic processes in speproduction: The retrieval of grammatical gender. Cog-nition, 64,115–152.

Vigliocco, G., & Nicol, J. (1998). Separating hierarchical retions and word order in language production: Is proxity concord syntactic or linear? Cognition, 68B,13–29.

Wasow, T. (1997). Remarks on grammatical weight. Lan-guage Variation and Change, 9, 81–105.

(Received December 5, 2000)(Revision received April 30, 2001)