Thematic Role Assignment in Context · 2019. 4. 1. · Thematic Role Assignment in Context Gerry T. M. Altmann University of York, Heslington, York YO10 5DD, United Kingdom Two experiments

Journal of Memory and Language41, 124–145 (1999)Article ID jmla.1999.2640, available online at http://www.idealibrary.com on

Thematic Role Assignment in Context

Gerry T. M. Altmann

University of York, Heslington, York YO10 5DD, United Kingdom

Two experiments are described in which participants read sentences of the form “He dranksome . . .” incontexts which either did or did not introduce something drinkable. Participants weremore likely to report that the sentence stopped making sense at the verb “drank” if nothing drinkablehad been introduced. When participants responded, in this case, that the verbdid make sense, theirreaction times tended to be elevated relative to when the context did introduce a suitable antecedent.The experiments were modeled on a series of studies reported by Boland, Tanenhaus, Garnsey, andCarlson (1995), who used the same stop-making-sense judgment task to investigate the processing offiller-gap dependencies of the form “I wondered which food the man drank . . . .” They observedincreased “no” responses on the verb when the prior filler was an implausible recipient of the patientrole associated with the verb and concluded that the wh-phrase is assigned a thematic role as soon asthe verb “drank” is encountered. In the studies reported here, with materials which did not containobligatory syntactic (filler-gap) dependencies, the equivalent phenomenon was observed—thematicinformation conveyed by a verb’s lexical entry was apparently evaluated, at the verb, with respect toits fit with whatever contextually introduced entities were available to receive the associated role.The data suggest that thematic roles may be assigned at a verb on the basis of thematic fit withcontext. © 1999 Academic Press

Key Words:thematic roles; parsing; context.

oshithethaor-ain, then

anyar—is-n a

blein

as-vail-

oredrivenen alessentivens—as-aticentsatthatrbal

thuninth

anyTheentonanExicew.

ersityl:

Much research into parsing has focusedattempts to understand better the relationbetween our knowledge of grammar andinterpretive processes that operate onknowledge. In this regard, Pritchett (1992) fmulated an important, and influential, constron sentence processing: broadly speakingprocessor attempts at each point in the sent

The work reported here was supported by grants fromNuffield Foundation and from the Medical Research Cocil (G9628472N). A pilot for Experiment 1 developedconversation with Judith Henstra, who helped createfiller items for that experiment and who provided minsightful comments on an earlier version of this paper.author thanks Anna Weighall for helping to run Experim1. The paper has benefited greatly from commentsprevious version by Keith Rayner, Martin Pickering, andanonymous reviewer. The experimental items used inperiments 1 and 2, although reproduced in the appendcan be obtained in electronic form from http://wwyork.ac.uk/;gtma1/thematic.html.

Address correspondence and reprint requests to GT. M. Altmann at the Department of Psychology, Univerof York, Heslington, York YO10 5DD, UK. E-mai
[email protected].
1240749-596X/99 $30.00Copyright © 1999 by Academic PressAll rights of reproduction in any form reserved.

np

t

te

ce

to assign as many thematic roles onto as marguments as are licensed by the gramm“when a theta role is available, it will be dcharged onto any available target, and whetarget is available, it will assume any availarole.” (p.13). This paper presents evidencesupport of the idea that thematic roles aresigned to targets as soon as both become aable. Whereas previous studies have explsuch assignments when they have been dby obligatory syntactic dependencies betweverb and the linguistic entities to which its roare assigned (cf. Boland et al., 1995), the prestudy explores thematic role assignments drinstead by optional contextual dependenciethey explore occasions when thematic rolesignments are made on the basis of the themfit between a verb and the discourse antecedwhich could plausibly receive a role from thverb, even when the anaphoric expressionrefers to (one of) those antecedents is postveand has yet to be encountered.

e-

e

a

-s,

ry

Pritchett’s claim that the processor attempts

ao

. Aesththas5;ick

m-enng

s:Ae d tH h-fi i-e em enm nedt al.lw elyT na edw as( anf that nc lts eea taxu aveb

o-l doen slys nom atm og thw b’sa

po-hatedrryof ation

ap-&

ng,

atedto

earl-&

pro-on-

noed

abi-n-ofs,k-to

wh-ncyngeuldeseter-

dentng

atd

on-thed asre-oresionm-lesnalced

odthe

125THEMATIC ROLE ASSIGNMENT IN CONTEXT

to assign roles to targets as soon as bothavailable is supported by experimental datathe processing of filler-gap dependencieslarge body of data has accrued which suggthat fillers are assigned roles as soon asrole-defining verb is encountered and beforesyntax unambiguously signals which role tosign to which filler (e.g., Boland et al., 199Garnsey, Tanenhaus, & Chapman, 1989; Pering & Barry, 1991). Boland et al., for exaple, explored the processing of filler-gap depdencies in the following structures (amoothers—see below).

(1) a. Which military base did Hank deliver the ma-chine guns to _ last week?

b. Which preschool nursery did Hank deliver themachine guns to _ last week?

In (1a), the verbdeliver signals three rolegent, Recipient, and Theme. Whendeliver isncountered, the Agent role will be assigneank. In addition, Boland et al. claim, the wller military basewill be assigned the Recipnt role also (not the Theme role becausilitary base is an implausible theme). Whachine gunsis encountered, they are assig

he Theme role. To test this, Boland etooked for an anomaly onmachine gunsin (1b),here the machine guns would be an unlikheme if preschool nurseryhad already beessigned the Recipient role. Boland et al. usord by word stop-making-sense judgment t

cf. Boland, Tanenhaus, & Garnsey, 1990)ound increased “no” responses (indicatinghe sentence had ceased to make sense) oma-hine gunsin (1b) but not in (1a). This resuuggests strongly that a thematic role has bssigned prior to the point at which the synnambiguously signaled which role should heen assigned.In the filler-gap constructions studied by B

and et al. (1995) and others, the processorot wait, then, until the syntax unambiguouignals the correct gap location. This doesean, however, that the role assignments thakes are not in accord with the principlesrammar; the wh-construction indicates thath-filler must receive a role, and the ver
rgument structure, coupled with knowledge o
ren

tsee-

-

-

o

a

akdt

n

s

titfe

the grammar, indicates the configurationalsitions in which the gap associated with tfiller is likely to be found. The results obtainby Boland et al. (1995) and Pickering and Ba(1991) can therefore be explained in termsprocessor which, in effect, uses the informaconveyed by a verb toproject structure—it isthis projected structure which triggers thepropriate role assignments (cf. GibsonHickok, 1993; Gorrell, 1993; but see Pickeri1993).

The results are also easily accommodwithin the constraint-satisfaction approachsentence processing (e.g., MacDonald, Pmutter, & Seidenberg, 1994; TrueswellTanenhaus, 1994), which views sentencecessing as the application of probabilistic cstraints in parallel as a sentence unfolds, withsingle constraint being more or less privilegthan any other except in respect of its problistic strength; the stronger one kind of costraint, the weaker the potential influenceothers (cf. Spivey-Knowlton & Tanenhau1994). It follows that if one constraint is lacing, other constraints could in principle comethe fore. For instance, the presence of afiller, and the associated syntactic dependewith the subsequent verb, is just one of a raof constraints on which specific entities shoreceive an upcoming thematic role. In thfiller-gap cases, grammatical constraints demine coreferentiality between some antece(the wh-filler) and a projected (null) referriexpression (the gap). However, in principleleast, contextual information (as embodiewithin the discourse model) could also cstrain, with some degree of probability,range of entities which should be considerepotentially coreferential with the projectedferring expression and which should therefreceive the role that the projected expreswould receive. In (2) and (3) below, for exaple, there is a high likelihood that thematic roassociated with the underlined verbs in the fisentences will be assigned to entities introduin the prior sentence. This high likeliho“qualifies” the context as a constraint on
finterpretation of the subsequent sentence.

hisenadneai

roleon

sorwitent

irsre-ceociblc-, at t

ndseri-dngpa

edd blisB

apoput

m,nlecingmaro

udg-ask.m-

pat-. Foredsen-.

f

l

m-im-

c-byalyes-theene-ie.t ofom-atedo-apst-but

ob-ass4a)n aatending-g of

p-itet

astme-sedere.n-hich

efert rges s aa ssw

126 GERRY T. M. ALTMANN

(2) The small boy reached up for the biscuits.He ate . . .

(3) Hank parked his van outside the local militarybase.He delivered the machine guns . . .

It is an empirical question, and one which tpaper addresses, whether any role assignmother than the Agent assignment would be mat the underlined verbs—if roles are assigsimply on the basis that a role becomes avable and a potential recipient exists for that(irrespective of the source of the constraintthe identity of that recipient), the processhould assign the Patient role associatedate in (2) above to the boy and the Recipirole associated withdelivered in (3) above tothe military base.

In Experiment 1 below, context–target pasimilar to that shown in (2) above were psented to participants. The contexts introdueither a plausible antecedent for a role assated with the subsequent verb or an implausantecedent.1 In Experiment 2, dative construtions such as that shown in (3) were usedwell as constructions that used an equivalenthe filled-gap logic introduced by Crain aFodor (1985) and by Stowe (1986) and ualso by Boland et al. (1995). In both expements, context–target pairs were presenteparticipants using Boland et al.’s stop-makisense paradigm. This paradigm was used inbecause the materials used here were basthe same range of structures as were useBoland et al., and the intention was to estabwhether the same phenomena observed byland et al. for grammatically forced filler-gdependencies would be observed also fortional contextual dependencies of the kind olined above. One criticism of this paradighowever, is that the judgment task may tap olate integrative processes which do not reflthe initial analysis that would be pursued durnormal language comprehension—the tasknot even reflect normal comprehension p

1 The term “antecedent” will be used henceforth to ro a discourse entity which may, in the subsequent taentence, receive a role from the main verb. In fact, it intecedent to the subsequent anaphoric referring expre
hose existence can be projected at that verb.
tsedl-

h

di-e

so

d

to-rtony

ho-

--

yt

y-

cesses, but may reflect instead reading and jment strategies that are unique to this tHowever, alternative paradigms which reseble normal reading have shown equivalentterns of data to the stop-making-sense taskexample, Pickering & Traxler (1998) monitorparticipants’ eye movements as they readtences such as those in example (4) below

(4) a. That’s the child that Mark reminded several othe women to watch this evening.

b. That’s the movie that Mark reminded severaof the women to watch this evening.

The postverbal region (“several of the”) unabiguously signals that there is no gap sitemediately following the verb (cf. “. . . the childthat Mark reminded _ to brush his teeth”). Acording to the filled-gap logic employedBoland et al. (1995) and others, an anomshould be observed in this region if the procsor commits itself at the verb to assigningrole associated with a postverbal gap (the Bfactive role) to either the child or the movAnd because a child is a plausible recipienthis role whereas a movie is not, such a cmitment could be made, and the associanomaly observed, in (4a) but not in (4b). Bland et al. did indeed find such a filled-geffect—increased “no” judgments in the poverbal region of sentences equivalent to (4a)not (4b). Pickering and Traxler (1998) alsoserved a filled-gap effect, with longer first preading times at the postverbal region in (compared with (4b). These data, based omore naturalistic reading task, thus replicBoland et al.’s (1995) filled-gap effect ahence validate their use of the stop-maksense judgment task to explore the processinthese filler-gap constructions.

Despite Pickering and Traxler’s (1998) relication of the Boland et al. (1995) effects,could still be argued that whereas Bolandal.’s judgment task tapped initial (or at leearly) processing in their cases, it may sohow tap some other level of processing if uin the contextualized cases investigated hThis possibility will be returned to in the Geeral Discussion, after data are presented w

tnion
do indeed show the same patterns of results as

va-haten-th

in-th

5)b

lisd im

tra-ichrth

olectebsinis-

lecnyen(7)

f

f

st),

cedoled-s t

be

n-as-are

eive” inis-

icaltiesect-the

o

ndi-ewas

befthat

eteely

ng

latt,a

nd

rsere

t off 1011re:

ingw vail-


reported in Boland et al., including the equilent of the filled-gap data, despite the fact tthe antecedents were introduced extrasentially, with no obligatory syntactic dependecies between the verb and the expressionsreferred to those antecedents.

EXPERIMENT 1

Boland et al. (1995) established thatcreased “no” judgments were observed onverb read in wh-filler constructions such as (

elow.

(5) Which food did the boy read . . . ?

The purpose of Experiment 1 was to estabwhether equivalent effects would be observethe recipient of the role associated with gramatical object position was introduced exsententially. Contexts were constructed wheither introduced a potential antecedent fosubsequent Theme or Patient role, givenselectional restrictions associated with that ror did not. Target sentences were construusing two kinds of verbs: “selecting” verwhose selectional restrictions would, dependon the context, either allow selection of a dcourse antecedent or would not; and “nonseing” verbs whose selectional restrictions, if aallowed selection of any discourse antecedThe basic design is exemplified in (6) andbelow.

(6) Antecedent conditionA car was driving downhill when it suddenlyveered out of control.In its path were some pigeons and a row obollards.It injured/missed . . .

(7) No-antecedent conditionA car was driving downhill when it suddenlyveered out of control.In its path were some dustbins and a row obollards.It injured/missed . . .

(“Bollard” is the British English for a short poused to impede the passage of cars.) In (6plausible antecedent (“pigeons”) is introduwhich could plausibly receive the Patient rassociated withinjured. In (7), no such anteceent is introduced. If the processor attempt
assign roles to available discourse antecedena
tn-

at

e

hf-

ae,d

g

t-,t.

a

o

evidence of a processing anomaly shouldobserved on the verbinjuredwhen it follows thecontext in (7) but not when it follows the cotext in (6)—in (6), the Patient role can besigned to the pigeons, whereas in (7) thereno plausible antecedents which could recthe Patient role (the basis for responding “nothis task is discussed in more detail in the Dcussion).

All target sentences referred in grammatobject position to one of the discourse entiintroduced in the prior context. Thus, the seling targets always became implausible atpostverbal noun phrase.

(8) Selecting targetIt injured several bollards that came close tobeing destroyed.

(9) Nonselecting targetIt missed several bollards that came close tbeing destroyed.

Whereas participants were all expected to icate an anomaly atbollards in response to thselecting target, at issue in Experiment 1whether evidence of an anomaly would alsoseen oninjured,but not onmissed,in the case othe contexts that did not introduce anythingcould plausibly be injured.

Method

Participants. Forty-eight members of thUniversity of York were paid £2 to participain the experiment, which lasted approximat30 min.

Apparatus.All stimuli were presented on aApple Macintosh with a 17-in. monitor runninPsyScope software (Cohen, MacWhinney, F& Provost, 1993).2 Participants responded onbutton box on which were marked a “yes” aa “no” button.

Materials. Thirty-two context–target paisuch as the items shown in (6) to (9) above wconstructed. See Appendix 1 for the full seitems. Each target sentence consisted owords, except for one target consisting ofwords. They each had the following structu

2 The PsyScope script for running word-by-word movindow displays with sentences of variable lengths is a

ts,ble from the author.

l)–ingd thgthh aerthegics”ncelec

ofoa

ly),igndth(cfon-en

ateul

ro-th

fill-erb

se-20

didhalet

ch) ananriaat

witedbudi

c-

thatof

ay ors theen-ety,ach

hathensould

ingtoto

f, atsedno”ey

thatolf”n-rac-ofces

hef-ererea-tou

aine to

e ofrdndeencu-sti-r off thear-int,sesnycor-


pronoun (it)–verb (injured)–adjective (severanoun (bollards)–continuation. The selectverbs averaged 6.2 characters in length, annonselecting verbs 6.3 characters in lenSeventy-five percent of items continued witrelative clause. The 32 experimental items wembedded among 68 filler items. Overall,fillers were designed to rule out any strateanticipation regarding either a “no” or a “yeresponse; overall, 50% of the target sentewere plausible and did not violate the setional restrictions of the verb.

The fillers were constructed as follows: 32the fillers used selecting verbs. Twenty-fourthese had “antecedent” contexts (containingentity to which the selecting verb could appand 8 had “no-antecedent” contexts. These efiller items with “no-antecedent” contexts (aselecting verbs) were rendered plausible intarget sentence by introducing a new entity“it injured several people that were standingthe sidewalk”). Of the 24 filler items with “antecedent” contexts (and selecting verbs), 8 mtioned in postverbal position the appropriantecedent, 8 introduced a new entity that coplausibly appear in that position, and 8 intduced a new entity that nonetheless violatedselectional restrictions of the verb. Sixteeners used nonselecting verbs. All of these wrendered implausible at the postverbal nounintroducing a novel entity that violated thelectional restrictions of that verb. There wereother fillers with contexts and targets thatnot resemble the experimental items at all;violated selectional restrictions in the targand half did not.

Design. There were two versions of eatarget sentence (selecting and nonselectingtwo versions of each context (antecedentno-antecedent). The design was a fully factorepeated measures design incorporating a LSquare. Four stimuli sets were constructedeach item represented in each set in just onits versions. Each subject was thus exposeall items and to all experimental conditions,never saw more than one version of any invidual item.

Procedure. The entire procedure—instru
tions, practice, and main session—was con
e.

e

s-

fn

ht

e.

-

d

e

ey

f,

ddlinhoftot-

trolled by PsyScope. Participants were toldeach story they would read would consistthree sentences and that the last sentence mmay not make sense. Their task was to pres“yes” button to see each of the first two stences, which were presented in their entirand to then press the “yes” button to reveal eword of the final sentence. To begin with, tsentence would appear as a series of hypacross the screen, and each button press wreveal the next word and “hide” the precedone. They were told to press the “yes” buttoneach word only if the sentence continuedmake sense and to press the “no” button iany point in the sentence, they felt that it ceato make sense. If a participant did respond “at any point, the trial would be terminated. Thwere given two examples in the instructionsdid not make sense (“he ate the books on gand “he read the fish on golf”). After the istruction phase, participants were given a ptice session consisting of four trials, twowhich contained implausible target senten(“he drank one of the books noisily” and “repaired the storeroom quite effortlessly”). Ater the practice sessions, participants wshown the practice materials again, and thesons for the implausibility were explainedthem (“you cannot drink a book” and “yocannot repair a storeroom”). During the mexperimental session, participants were frepause after any individual trial.

Results

Figure 1 shows the cumulative percentag“no” judgments for each of the first four wopositions. Following Boland et al. (1995), ain order to eliminate the dependence betwthe cumulative figure at one word and themulative figure for the preceding word, statical analyses were carried out on the numbe“no” responses calculated as a percentage oresponses still available at that point (if a pticipant had responded “no” at a previous pothe trial would be terminated, and no responfrom that subject would be available at asubsequent point in that sentence). The
-rected percentages (as a function of available

firscitrth

etsth12

2ral

urelethetheectereanfol

ns-f theon-gednal-ichlitydatametri-

(2hetely.onre

,

ge

s

21)88)24)

37)


responses) are given in Table 1. Only thefour positions were analyzed due to the pauof “yes” responses beyond the noun in fouposition in the selecting targets conditions.

Overall, 97% of experimental items’ targcontaining selecting verbs were judged, byend of the sentence, not to make sense, andof targets with nonselecting verbs (F1(1,47)5

266,P 5 .0001;F2(1,31)5 2196,P 5 .0001).There was no effect of context on the ovefigures nor any interaction with context (allF ,1). These figures are comparable to the figfor the fillers, with 97% of the implausibfillers being judged implausible, and 14% ofplausible fillers being judged implausible. Incase of the experimental items with nonseling verbs, 50% of the “no” responses wmade on the final word of the sentence (only 4.7% of available responses at nouns

FIG. 1. The cumulative percenta

TAB

Percentage of Remaining “No” Responses, fo

Target Context It

Selecting Antecedent 0.5 (412)No antecedent 0.5 (402)

Nonselecting Antecedent 0.5 (408)No antecedent 0 (404)

Note.Numbers in parentheses indicate the mean judgm

ty

e%

l

s

-

d-

lowing nonselecting verbs were “no” respoes), suggesting that some other aspect ocontext–target pairing (other than the relatiship between the verb and its object) was judimplausible in these cases. The statistical ayses reported below include trials on whsubjects “misjudged” the (intended) plausibiof the target sentences—in effect, thesewere treated as “noise,” and exactly the sastatistical patterns were observed with theseals excluded.

Separate two-way analyses of variancecontexts3 2 verbs) were carried out on tjudgment data at each word position separaThere were no “no” judgments in positione—the pronoun. At the following verb, thewere main effects of context (F1(1,47)5 4.3,P , .05; F2(1,31) 5 6.1, P , .02) and verb(F1(1,47) 5 5.2, P , .03; F2(1,31) 5 17.55

of “no” responses for Experiment 1.

1

ach of the First Four Word Positions in Experiment 1

Word position

Injured/missed Several Bollard

0.8 (572) 1.6 (435) 78.9 (66.3 (607) 2.1 (470) 76.1 (60.8 (504) 0 (442) 4.8 (6

0.5 (511) 0 (440) 4.6 (6

LE

r E

ent times (ms) when subjects responded “yes.”

en

re-tiotinthbl

eerb

nstha0%di

osere

.a

achca-ses”

,on-tha

rentes

bytmsene

d-ans v

achaner-

g-dentg-en

adethatent

re-ereadewerau-anoreeen

roletextForttedere

ro-renot

ina-enttheun-

h aci-on-es.

er-t isoterb,singnte-to,jectto

hanme

l re-the


P 5 .002) and a significant interaction betwethe two (F1(1,47)5 4.8, P , .04; F2(1,31)510.4, P 5 .003)—planned comparisonsvealed that these effects and the interacwere due to more “no” responses to selecverbs in the no-antecedent condition than inantecedent condition (6.3% vs. 0.8% availaresponses;F1(1,47)5 8.8,P , .005),F2(1,31)5 18.8, P 5 .0001). At position three, thadjective, there was a marginal effect of v(F1(1,47)5 19.3,P 5 .0001;F2(1,31)5 1.4,P . .2), with the selecting verb conditiotending to engender more “no” responsesthe nonselecting verb conditions (1.8% vs.available responses). There were no otherferences at this position (allF , 1.0). At thenoun, there was a significant main effectverb—77.5% of available responses to thelecting verbs were “no” responses, compawith 4.7% for nonselecting verbs;F1(1,47)5624, P 5 .0001;F2(1,31)5 609, P 5 .0001There were no other significant differencesthis position (allP . .2).

Similar analyses were carried out at eword position on the judgment times for ocsions when participants responded “yes” (Table 1). There were no differences in “yetimes at the pronoun (allF , 1). At the verbthere was a main effect of verb type, with nselecting verbs being responded to fasterselecting verbs (507 ms vs. 590 ms;F1(1,47)515.6, P 5 .0003; F2(1,31) 5 6.4, P , .02).Planned comparisons for selecting verbsvealed that the difference between the no-acedent case and the antecedent case (607 m572 ms) tended toward significance on thesubjects analysis (F1(1,47)5 3.9,P , .06), bufailed to reach significance on the by-iteanalysis (F2(1,31), 1). At the adjective, therwere no main effects nor interactions. Plancomparisons revealed, nonetheless, that theference for selecting verbs between the notecedent and antecedent contexts (470 m435 ms) was marginally significant (F1(1,47)54.7, P , .04; F2(1,31) 5 3.1, P , .09). Thedifferences at the final noun did not approsignificance. There were no other significmain effects, interactions, or pairwise diff
ences at these first four word positions.
ngee

n

f-

f-

d

t

e

n

--vs.-

dif--s.

t

As expected, there were more “no” judments at selecting verbs in the no-antececonditions. There were virtually no such judments in any of the other conditions. Whparticipants judged that the verbs did msense, there was some limited indicationjudgments were longer in the no-antecedcondition for selecting verbs only.

Discussion

Boland et al. (1995) observed more “no”sponses on the verb when their wh-fillers wimplausible recipients of the role that was mavailable by the verb and correspondingly fe“no” responses when the wh-filler was a plsible recipient of that role. In Experiment 1equivalent effect was found: there were m“no” responses on the verb when it had bpreceded by a context that didnot introduce aplausible antecedent for its Theme/Patientthan when it had been preceded by a conthat did introduce a plausible antecedent.verbs whose selectional restrictions admiplausible antecedents from either context, thwas no effect of context (and a negligible pportion of “no” responses). The filler items wedesigned to ensure that participants couldhave anticipated, on the basis of any combtion of context and target verb, which judgmwould subsequently prove appropriate whenpostverbal referring expression was encotered. Moreover, given the manner in whic“no” response would terminate a trial, partipants in fact experienced more plausible ctinuations after the verb than implausible on

The magnitude of the effect found in Expiment 1 was relatively small. In this regard, inoteworthy that even on trials which did nengender a “no” response at the selecting vthere was some limited evidence of a procescost associated with the failure to find an acedent which could plausibly be referredsubsequently, in postverbal grammatical obposition. It is unclear why judgment timesnonselecting verbs were, overall, faster tthose to selecting verbs—this may reflect soprocessing cost associated with selectionastriction of the referential domain. However,
difference should be interpreted with caution

arossbl

f-thereasr-

chcrlieasg-

t tht inati(oo

ntcedtiapoly

de-. Oolvt in

f

rolptIn

rticese

thein

ts tnaerh

s isses,theetedng-withtab-

res-such

an-n-

to,hatatedforatcedre-

uebutioncom-

ingeseopleinval-theird,ec-data

esti-egic

inr, aomthentshicheayhicht 2se-


given that it relies on a between-items compison (although the difference in length acrthe two types of verb of 0.1 characters probacannot account for this difference).

Boland et al. interpreted their plausibility efects to indicate that readers made use inexperiments of the verbs’ argument structuand their associated thematic information tosign roles to wh-fillers before syntactic infomation was available to uniquely identify whirole assignment was intended. If the sameteria regarding behavioral outcome are apphere as were applied in the Boland et al. c(regarding the sensitivity of the stop-makinsense task to thematic role assignments averb—see below), we would conclude thaExperiment 1 also, we had evidence of theminformation at the verb being used to assignat least evaluate alternative assignmentsroles to discourse antecedents. In Experimeunlike in the Boland et al. case, these anteents were not marked syntactically as potenrecipients of those roles—they constitutedtential recipients by virtue of being the onentities in the context to which the eventnoted by the subsequent verb could applycourse, this event need not necessarily invany previously introduced entity, as the tex(10) demonstrates.

(10) A car was driving downhill when it suddenlyveered out of control.In its path were some dustbins and a row obollards.It injured several people who were standingnearby.

Thus, the processor need not attempt, atinjuredin that final sentence, to assign the Patientto a previously introduced entity (an attemwhich would fail in the case of this example).which case, why, and on what basis, did paipants respond “no” at the target verb in thexamples?

One interpretation of these data is thatprocessor projects, at the verb, the upcompostverbal referring expression and attempestablish, given the thematic criteria (selectiorestrictions) imposed by the verb, whether thare any discourse antecedents with whic
could be coreferential. When there is no suc
-

y

irs-

i-de

e

crf)1,-l-

fe

e

-

goleit

antecedent, participants respond “no.” Thinot to say that the processor is, in such caunable to proceed—the ease with whichtarget sentence in (10) above can be interpris testimony to that. Rather, the stop-makisense judgment task is sensitive to the easewhich anaphoric dependencies can be eslished between the projected referring expsion and the discourse antecedents (if anyantecedents exist—if there are no discoursetecedents, as ina runaway car yesterday ijured . . . , no anaphoric dependencies havebe considered, andinjured would, presumablyprove unproblematic for the processor). Tsuch sensitivities exist has been demonstrelsewhere (see Garrod & Sanford, 1994,review); for instance, Murphy (1984) found thestablishing a new discourse referent produlonger sentence comprehension times thanferring to an already existing referent. At issthen is not whether such sensitivities exist,rather at what point during the comprehensprocess these anaphoric dependencies areputed.

If the judgment task employed here is tappprocessing during the initial analysis of thsentences, the data would suggest that pecan, and do, fill thematic roles predictivelythe manner outlined earlier (or at least can euate alternative antecedents with respect tothematic fit with the verb). If, on the other hanthe judgment task taps more strategic, refltive, task-dependent processes, then thesuggest that peoplecan predictively evaluatthe anaphoric and thematic relations invegated here, but only when there are stratreasons for doing so. This point is revisitedthe General Discussion. There is, howevefurther reason for interpreting the data frExperiment 1 with caution. It concerns nottask itself, but rather the fact that participawere expected to process target sentences whad a high likelihood (50%) of violating thselectional restrictions of the verb. This mitself have engendered reading strategies wdo not reflect normal reading. Experimenrectifies this problem by avoiding any such
hlectional violations; all the postverbal continu-

ati

ntce

phap

pege

theet

entin

l

ithy

( denc oma n-t ineg exto DiscTc tr ant ntb be-t t ind

then theo dat ene tedt

w hep erba lter-n tedf ct att thei n.C t thev ont uldn thea

thefi eta tar-g thea ap”i

d inalltwo

n athe“heafteru-lk-ndi-ibleo bentheil-as-

ho-eir-ut-nge


ations were plausible in respect of the themfit between the verb and its arguments.

Experiment 2 also differed from Experime1 in that it used a variety of different sententypes to explore the same role assignmentnomenon, including the dative and filled-gconstructions described earlier, but as in Eximent 1, the potential recipients of the tarroles were introduced in the context.

EXPERIMENT 2

Experiment 2 used modified versions ofBoland et al. (1995) stimuli. Context–targpairs were devised for a number of differtarget constructions. Each will be describedturn.

(11) DativesPlausible antecedent context and target

Hank parked his van outside the local militarybase.He delivered some machine guns to the militarybase and left.

Implausible antecedent context and targetHank parked his van outside the preschoonursery.He delivered some machine guns to the militarybase next door.

If, in (11), the Recipient role associated wdelivered is assigned topreschool nurserwhen preceded by the implausible anteceontext), there should be evidence of an anly whenmachine gunsis subsequently encou

ered. Note that although delivering machuns may be slightly implausible in the contf parking outside a preschool nursery (seeussion), the thematic fit ofmachine gunsto theheme role associated withdeliveredis entirelyompatible withmilitary basein the Recipienole. Thus, subjects could not strategicallyicipate, in Experiment 2, a “no” judgmeased on any anticipated incompatibility

ween the direct object and the subsequenirect object.Boland et al. used verbs that permitted ei

onalternating dative constructions (as inne above) or which permitted alternating

ive constructions also (e.g., “she granted grous maternity leave to them”/“she gran

hem generous maternity leave”). All the verbs

c

e-

r-t

t-

-

-

-

r

--

ere used in their nonalternated form. Tresent study also included both types of vnd also presented them only in their nonaated form. The possibility of an alterna

orm means that the processor cannot predihe verb which role will be assigned atmmediately following postverbal positioonsequently, any role assignments made aerb (other than Agent) could not be madehe basis of which role is due next, but woeed to take into account the plausibility oflternative assignments.In addition to the dative constructions,

ller-gap constructions employed by Bolandl. (1995) were modified to create context–et pairs such as the following (and inbsence of any true gaps, the term “filled-g

s replaced by “filled-role”).

(12) Filled-rolesPlausible antecedent context

The twins listened to their father talking to theirmother.

Implausible antecedent contextThe twins listened to their father talking abouttheir mother.TargetHe asked them to be especially nice to her.

The same discourse entities are introduceboth contexts in (12); in structural terms,should be equally accessible (across thecontext conditions, but not necessarily withicontext—see below). However, whereasmother is a plausible candidate for whoeverasked” in the plausible antecedent case (all, he was talking to the mother, so it is plasible that the asking took place within the taing event), the mother is a less plausible cadate for whoever was asked in the implausantecedent case (the father was less likely ttalking to the mother). If contextually drivethematic role assignments are sensitive toreal-world (or rather, discourse world) avaability of candidates, the processor shouldsign to the mother, at the verbasked in thetarget sentence, the role associated with wever was being asked in the talking-to-thmother context, but not in the talking-abotheir-mother context. Thus, the subtle cha
between the two contexts, betweento and

o”fa

nn-thetheeth

r toder-nterges-

b g.,G 989G -uc tht ,i f aa nsa

al.( tod topm tieo ada ses

heo ),

alsth-

er5)redthee

d vaa chm

i ntiala thatt akec er-w thiss ass

thesi-

the5)

pair-rvedon-

eteely

as

ingich

ur-da-on-geteter-5thundialandeen

de-ineseun–

ctshese


about, should be reflected in increased “njudgments on the target pronoun when thether was heard talkingto the mother than whehe was heard talkingabout her (because, cotrary to expectation, the direct object into-their-mother case does not refer tomother). The twins were introduced into thscenarios described in (12) to ensure thatpostverbal pronoun could successfully refesomebody other than the target antece(“their mother”)—in the talking-about-theimother context, there is thus a plausible acedent for the being-asked role when the taverbaskedis encountered. According to Gern

acher’s “Advantage of First Mention” (e.ernsbacher, Hargreaves, & Beeman, 1ernsbacher, 1990),the twinsshould be particlarly salient as an antecedent forthemin bothases. Of course, this salience could meanhey are assigned the Benefactive role ataskedn which case there should be no evidence onomaly in either of the two context conditiot them.Finally, Experiment 2 followed Boland et

1995) in including also structures devisedemonstrate, among other things, that the saking-sense task is sensitive to the subtlef the alternative role assignments that are mvailable at the target verb. Boland et al. utructures as follows.

(13) a. Which child did your brother remind _ towatch the show?

b. Which movie did your brother remind _ towatch the show?

In (13a)child is a plausible Benefactive for tbject control verbremind, whereas in (13b

movieis not. However, the verbremindsignalsmore than just the Benefactive role; it signalso a “generalized Theme” associated withverb in the (optional) upcoming infinitival complement (cf. “Which movie did your brothremind you to watch _ ?”). Boland et al. (199found that in (13b) no anomaly was registeuntil to and beyond. They proposed thatimplausibility of movie in the Benefactive rol

oes not cause an anomaly because of the ability at the verb of an alternative role whi
oviecan instead be assigned. This finding is
-

e

nt

-t

;

at

n

-se

d

e

il-

mportant because it is an example of a potenomaly which is not detected (suggesting

he task does not simply force readers to mommitments earlier than they would othise). Experiment 2 attempts to replicateame effect with context–target pairingshown in (14).

(14) Object controlContext

Mike quickly chose a video.Transitive target

He ate his sandwich in the TV room.Object control target

He reminded his brother to get some popcorn.

For both targets, the video introduced incontext is an implausible Patient (in the trantive target) or an implausible Benefactive (inobject control target). If the Boland et al. (199results generalize to these context–targetings, an anomaly at the verb should be obsein the transitive case but not in the object ctrol case.

Method

Participants. Twenty-four members of thUniversity of York were paid £2 to participain the experiment, which lasted approximat30 min.

Apparatus.The same apparatus was usedin Experiment 1.

Materials.Sixteen context–target items usnonalternating dative verbs were devised whmirrored the items in (11) above. Sixteen fther items were devised using alternatingtives. All the items were presented in their nalternated form. The first 5 words of each tarconsisted of the sequence pronoun–verb–dminer–adjective/noun–noun (the 4th andwords could either be a noun–noun compoor an adjective followed by a noun); the cruceffects were expected on the adjective/nounnoun. These 32 targets ranged in length betw8 and 15 words. A further 16 items werevised which mirrored the filled-role items(12) above. The first 5 words of each of thtargets consisted of the sequence: pronoverb–pronoun–“to”–verb (the crucial effewere expected on the second pronoun). T
target sentences were all 9 words long. Sixteen

oll 8geha

bedu-rbsnt

chlledntethensie

ureitheexonn o

tow

thaat

ex-parp-e ide

k.

t

ch

r

theuc-

e-e re-inin

nsexts4%con-tingted,also

eresi-nsere

emre-tici-osi-

sns.

thestal-

nal-

re-

ingvs.

the:

tosi-

erethebleny

ed


transitive/object control items were devisedthe form shown in (14) above. They were awords long. The object control verbs avera7.8 characters, and the transitive verbs 6.2 cacters. The 64 experimental items were emded amongst a further 48 fillers. All were plasible. Sixteen of these used object control veThe fillers varied in similarity to the differeexperimental items.

Design. There were two versions of eaexperimental passage: for the datives and firole constructions, these were “plausible acedent” and “implausible antecedent.” Forobject control constructions, these were “trative target” and “object control target.” Thdesign was a fully factorial repeated measdesign. Two stimuli sets were constructed weach item represented in each set in just onits two versions. Each participant was thusposed to all items and to all experimental cditions, but never saw more than one versioany individual item.

Procedure.The procedure was identicalthat used in Experiment 1. There was, hoever, some modification to the exampleswere given to participants to explain whwas meant by “not making sense”—theamples (and explanation as given to theticipants) were as follows (the first two apeared in the instructions and the next threboth the five-trial practice phase and thebriefing that followed).

(15) Mary was talking to her mother.She told him she’d been unhappy.

(The word “him” is anomalous)John was talking to his mother.He told her it had been unhappy.

(“It” is introduced out of the blue)Jake parked his car in the supermarket car parHe locked the banana and went in.

(Bananas are unlikely things to be locked)A lecturer was talking to some students abouthe course.He encouraged her to ensure she came to ealecture.

(The “her,” and subsequent “she,” appeafrom nowhere, so to speak)The politician defended himself during thedebate.He denied the lights were true.

(The “lights” appear as if from nowhere . . .)

f

dr--

.

--

-

s

of--f

-t

-

n-

Results

Results will be reported separately fordative constructions, the filled-role constrtions, and the object control constructions.

Datives. The mean numbers of “no” rsponses, expressed as a percentage of thmaining available responses, are givenTable 2. The cumulative data are shownFig. 2. Overall, 75% of the dative constructiopreceded by implausible antecedent contwere judged implausible, compared with 2when preceded by plausible antecedenttexts. Only the data across both the alternaand nonalternating dative verbs are reporalthough the patterns reported below applyto the individual subsets.

Separate one-way analyses of variance wperformed at each of the first eight word potions on the percentage remaining respodata. Although the majority of targets wemore than eight words long, to include thwould have confounded unavailability ofsponses at a particular position due to parpants having responded “no” at previous ptions and unavailability due to fewer itembeing available at successively later positioThere were no significant differences innumber of “no” judgments in any of the firthree word positions (“he delivered some”),though at word position three the by-items aysis approached significance (F2(1,31) 5 3.5,P , .08). At the next word (“machine”) thewere more “no” judgments following the implausible antecedent condition than followthe plausible antecedent contexts (12.7%2.4%:F1(1,23)5 20.5,P 5 .0002;F2(1,31)511.2, P , .003). The same was true atfollowing word (“guns”)—30.1% vs. 5.1%F1(1,23)5 61.8,P 5 .0001;F2(1,31)5 23.7,P 5 .0001. This plausibility effect persisted inthe following words (but not the postprepotional determiner in position seven).

Separate one-way analyses of variance walso performed at each word position onjudgment times for “yes” responses (see Ta2). There were no significant differences in aof the first four word positions (“he deliver
some machine”). At the fifth word (“guns”) the

usith

ms

s a-tex

-peseatan-coithed

een

o”si-tts

ionon-

. ewv

if-

s”onlydi-”),ent

re-texts

se)

er

.6109).768)


difference between the plausible and implable antecedent contexts was significant onby-subjects analysis only (669 ms vs. 763F1(1,23) 5 5.0, P , .04; F2 , 1). The onlydifference in subsequent word positions waword eight (“military”), with faster “yes” responses in the plausible antecedent con(500 ms vs. 789 ms:F1(1,23) 5 46.3, P ,.0001;F2(1,31)5 20.8,P , .0001).

Filled-role constructions.The mean numbers of “no” responses, expressed as acentage of the remaining available responare given in Table 2, with the cumulative dshown in Fig. 3. Overall, 66% of target setences preceded by plausible antecedenttexts were judged implausible, compared w42% when preceded by implausible antecent contexts.

Separate one-way analyses of variance wperformed at each word position on the perc

TAB

Percentage of Remaining “No” Responses, foand Object Control Co

Target Context

He delivered some

Datives Plausible 0 0.5 0.8antecedent (441) (525) (497

Implausible 0.3 0.5 2.9antecedent (434) (532) (503

He asked them

Filled- Plausible 1.0 0.5 25.4role antecedent (503) (586) (86

Implausible 0 0.5 8.5antecedent (505) (579) (581

He reminded his

Object Object control 0 2.6 4.7control verb (440) (550) (532

He ate his

Transitive 0 14.6 12.2verb (446) (660) (508

Note.Numbers in parentheses indicate the mean jud

age remaining response data. There were n

-e:

t

ts

r-s,

n-

-

ret-

significant differences in the number of “njudgments in either of the first two word potions (“he reminded”); allF , 1. At the nexword (“them”) there were more “no” judgmenfollowing the plausible antecedent conditthan following the implausible antecedent ctexts (25.4% vs. 8.5%:F1(1,23)5 20.6, P 50001;F2(1,15)5 20.2,P 5 .0004). The samas true at the following word (“to”)—14.4%s. 3.1%:F1(1,23)5 9.4,P , .006;F2(1,15)5

16.2,P , .002. There were no significant dferences in later word positions.

Analyses of the reaction time data (for “yeresponses—see Table 2) revealed that thesignificant difference between the two contions was at the postverbal pronoun (“themwith responses in the implausible antecedcontexts being significantly faster thansponses in the plausible antecedent con(581 ms vs. 860 ms:F1(1,23) 5 15.1, P ,

2

ach of the Word Positions in the Dative, Filled-Role,tructions of Experiment 2

Word position

machine guns to the military (ba

2.4 5.1 0.6 4.1 2.1(609) (669) (462) (551) (500)12.7 30.1 13.6 7.1 20.2(647) (763) (528) (507) (789)

to be especially nice to h

14.4 6.9 4.8 5.4 2.4 35(520) (545) (474) (659) (540) (13.1 3.3 0.6 8.1 4.7 23(516) (514) (544) (678) (470) (8

brother to get some popcorn

3.6 1.1 1.9 2.7 14.3(719) (460) (490) (491) (844)

sandwich in the TV room

16.3 9.2 2.1 3.7 20.5(672) (504) (479) (525) (847)

ent times (ms) when subjects responded “yes.”

LE

r Ens

)

)

0)

)

)

)

gm

o.0008;F2(1,15)5 8.3, P , .02).

-cenata

veom

een

nhed”erthe

ject

eer-

dvs.

ar-

seethatith

sp


Object control constructions.The mean numbers of “no” responses, expressed as a perage of the remaining available responses,given in Table 2, with the cumulative dashown in Fig. 4. Overall, 54% of transititarget sentences were judged implausible, cpared with 27% of object control targets.

Separate one-way analyses of variance wperformed at each word position on the percage remaining response data. There were“no” judgments made to the first word of ttarget sentences. At the next word (“reminde“ate”), planned comparisons revealed thwere more “no” judgments in response to

FIG. 2. The cumulative percentage of “no” re

FIG. 3. The cumulative percentage of “no” res

t-re

-

ret-o

/e

transitive verb than in response to the obcontrol verb (14.6% vs. 2.6%:F1(1,23)5 20.3,P 5 .0002;F2(1,15)5 7.2,P , .02). The sampattern held at the following word (the detminer—12.2% vs. 4.7%:F1(1,23)5 6.4, P ,.02; F2(1,15)5 4.7, P , .04) and at the worafter that (the postverbal noun)—16.3%3.6% (F1(1,23)5 12.5,P , .002;F2(1,15)510.4, P , .003). There was some hint ofdifference in the following word, but no diffeences at all in subsequent word positions.

Reaction time data for “yes” responses (Table 2) revealed a difference at the verbwas significant on the by-subjects analysis, w

onses for the dative constructions in Experiment 2.

ses for the filled-role constructions in Experiment 2.
pon

thas

the

tothith(aad

beeAnjecer---

ntstheretexbedgof

heil-

litylau

alyw.

l

the

d asit

c (cf.M ne-t onala elya rth-c ac-t rth-c anN 95)o ntsm er-e nta-t nts,a as

ns( utt

ons


faster responses to the object control verbsto the transitive verbs (550 ms vs. 660 mF1(1,23) 5 12.5, P , .002; F2(1,15) 5 2.1,P . .1). There were no other differences atother word positions.

Discussion

The pattern of “no” judgments in responsethe dative constructions suggests that bytime the sentential position associated wTheme was encountered, the Recipient rolesociated with a later sentential position) halready been assigned to an entity that hadintroduced in the prior sentential context.anomaly was observed at the indirect ob(receiving the Theme role) only when its refent was an implausible Themegiven that discourse entity as Recipient.There was some indication also that even when participaresponded “yes” at the indirect object whencontext introduced an implausible Recipient,action times were slower than when the conintroduced a plausible Recipient. It couldargued, however, that the increased “no” juments on “machine guns” in the contextparking a van at a school nursery and tdelivering the machine guns (albeit to the mitary base next door) reflects the implausibiof the scenario as a whole and not the imp

FIG. 4. The cumulative percentage of “no” resp

sibility of any specific role assignment. Thus, asc

n:

e

s-

n

t

-t

-

n

-

one reviewer pointed out, the same anommight be found on “machine guns” in (16) belo

(16) Hank parked his van outside the preschoonursery. He saw some machine guns . . .

Such a finding would not be problematic forpresent account because although the verbsawdoes not assign a Recipient role tonursery inthe same way thatdeliveredwould, it could inprinciple assign a Locative role tonursery.Ofcourse, because the locative is not specifiepart of the verbsaw’s argument structure,annot be considered an implicit argumentauner, Tanenhaus, & Carlson, 1995). No

heless, the grammar does specify an optidjunct, and if the processor can predictivctivate representations corresponding to fooming arguments, it may also predictivelyivate representations corresponding to fooming adjuncts (cf. Altmann, Garnham, vice, & Henstra, 1998). As Mauner et al. (19bserved, the encoding of implicit argumeay support important aspects of text cohnce—the predictive activation of represe

ions corresponding to forthcoming argumes well as forthcoming adjuncts, may playimilar role.With regard to the filled-role constructio

“The twins heard their father talking to/aboheir mother. He asked them . . .”): When the

es for the object control constructions in Experiment 2.

ontext introduced a plausible recipient for the

aproereursterimthe

erthsigthouereghr t

erbr eu-antana.then-

hocan

res(cfbeorursonrobe

ro-disro-theanheth

eirchento

Hee

sen-thebeen-not

rolethe”).

TVsi-e of

getnem-tionru-eend-to

1.5the-nseing

theuld

gingn-at-be

dis-s-llerich

r anrseon-tionths

therurse


role associated with the postverbal object,anomaly was observed when the postverbalnoun did not refer to that discourse entity. Thwas no equivalent anomaly when that discoentity was not a plausible recipient. One inpretation of these data, then, is that by the tthe postverbal pronoun was encountered intarget sentence, the processor had “assumthat the mother in the talking-to-their-mothcontext was the person being asked andwhen an alternative role assignment wasnaled by the pronoun, an anomaly arose. Inimplausible antecedent case (“talking abtheir mother”), the slight anomaly seen th(8.5% “no” responses on the pronoun) mireflect the expectation that the pronoun refesome third party that was implied by the vbut was not explicitly expressed (cf. Mauneal.’s (1995) evidence that the “implicit argments” of a verb can be represented semcally and can be available for subsequentphoric reference even when not expressed)

It is possible that the effect observed onpronoun is not indicative of the prior assigment at the verb of the Experiencer role (wever was being talked to), but rather is inditive of a difficulty in resolving the pronouwhen it is encountered—that is, it is indicativeinstead of the sensitivities of the anaphoricolution process to prior discourse structureGarrod & Sanford, 1994). Thus, it mayharder to establish the appropriate anaphdependency if there exists some other discoentity which could plausibly, given the situatidescribed in the context, receive the sameas the actual anaphor. However, for this tothe case, either this alternative role assignmmust already be encoded by the time the pnoun is encountered or else it has to be “covered” during the pronominal resolution pcess. If the latter is true, it would appear thatplausibility of the alternative assignment cinterfere with the resolution process even wnumber mismatch, for example, betweenpronoun (“them”) and the alternative (“thmother”) might be expected to rule out suinterference. Nonetheless, these data are opalternative interpretations, and we return
these in the General Discussion.
n-

e-eed”

at-et

to

t

i--

--

-.

ice

leent

-

ne

to

The transitive/object control structures (“ate/reminded his . . .”) demonstrate that thstop-making-sense judgment task is indeedsitive, as Boland et al. (1995) found, tosubtleties of the alternative roles that mightmade available at the verb. Both target stences were preceded by a context which didintroduce a plausible antecedent for theassociated with the grammatical object oftarget verb (“Mike quickly chose a videoWhereas an anomaly was observed atate in thetarget sentence “He ate his sandwich in theroom” (replicating the effect found with trantives in Experiment 1), there was no evidencany equivalent anomaly atremindedin the tar-get sentence “He reminded his brother tosome popcorn,” becauseremind introduces aadditional role which a video could fill—thgeneralized Theme role in a potentially upcoing complement. Of course, a degree of cauis required in interpreting these data—the ccial comparisons in these cases are betwdifferent verbs (although the reaction time avantage of the object control verbs is unlikelybe a simple length effect, as they werecharacters longer than the transitives). Noneless, if participants in the stop-making-setask were adopting a strategy of respond“no” simply because there was nothing incontext that the verb could apply to, we wonot expect this result.

GENERAL DISCUSSION

Experiments 1 and 2 present data converon the view that, in sufficiently constrained cotexts, the thematic role associated with whever is about to appear in object position willassigned, at the verb, to whatever priorcourse entity can plausibly fill that role. In esence, it does not matter whether it is a wh-ficonstruction that places a constraint on whentity should be considered a candidate foupcoming role or whether it is a prior discoucontext—nothing distinguishes the two cstraints, according to the constraint-satisfacapproach, except their probabilistic streng(relative to each other and relative to the oconstraints that are applied as the disco
unfolds). By what mechanism does the context

haon-nthisragstxtser

ichxtsacpoomat

s bithThheakith

lf a&eris

totruthpaheire

ntiaht

ib-ndar-ingxistruchinplextse-theeenuchsi

the

ionsEl-

te,dis-

hetherco-

ndi-vescti-ingthe

ipular. Itt acanndsign-ct,thene

rep-en-

ngeb-

the.g.,).

ul-de-si-gle

aintion

stn athelish

ctions to

rna-

Theh a


influence the assignment process? And wwould happen if, unlike in the constrained ctexts used here, more than one potential acedent were introduced in the context? Tsecond question is related to a third, more pmatic, one: Are the effects reported here jupeculiarity of the highly constrained contethat were employed in these studies or is thsome underlying principle at work here whshould generalize to less constrained conte

It was suggested earlier that in order tocommodate these data one need simply prothat the processor projects, at the verb, upcing argument structure and immediatelytempts to establish anaphoric dependencietween the noun phrases that are projected wthat structure and the discourse context.present data suggest that in establishing tanaphoric dependencies, the processor tinto account thematic criteria associated wthe verb. Projection of structure is not in itsenovel idea (cf. Fodor & Frazier, 1980; GibsonHickok, 1993; Gorrell, 1993); however, undthe current interpretation, what is projectedinterpretedstructure.

Within a probabilistic constraints approachsentence processing, a verb’s argument sture represents the predictive contingencieshold between a verb and other elements atticular positions relative to that verb within tsentence. If such contingencies are acquthrough experience of the contexts (senteand otherwise) in which individual verbs migoccur (cf. the kind of predictive learning exhited by Elman’s (1990) recurrent network), aif the referring expressions in the differentgument positions commonly refer to existdiscourse entities, it follows that there may econtingencies between a verb’s argument sture and the extrasentential contexts witwhich a verb may tend to occur. For examverbs of eating may tend to occur in conte(linguistic or real-world) which introduce, bfore the verb is introduced, whatever will beobject of the eating. This contingency, betwverbs of eating and the contexts in which sverbs tend to occur, would then form the bafor contextually driven role assignments of
kind observed in Experiments 1 and 2—if the
t

e-

-a

e

?-se-

-e-inesees

c-atr-

dl

t-

,

s

processor predictively activates representatas a function of these contingencies (cf.man’s network), it would predictively activaat the verb, some representation of whichcourse entities could be referred to next.

Within this framework, the encoding of tcontingencies between one thing and anothings, and the contexts in which they canoccur, encodes also therelationship betweenthe things—the contingencies define the cotions in which those things can find themseland the conditions which then result. The avation of representations reflecting the encodof these contingencies thus corresponds to“recognition” of the appropriate relationshand, as such, to the assignment of a particrole associated with the verb to its objectfollows, then, that the predictive activation averb of alternative representations for whatoccur in subsequent object position correspoto the representation of alternative role assments. This situation is little different, in fafrom the one during lexical access whenacoustic input is compatible with more than olexical entry. There, we see evidence thatresentations corresponding to the alternativetries are activated (reflecting, in effect, a raof predictions regarding how the input will susequently unfold in time) until such time asinput rules out all but a single alternative (eMarslen-Wilson, 1987; Zwitserlood, 1989Thus, although the predictive activation of mtiple alternatives must, according the viewveloped here, correspond to multiple (andmultaneous) role assignments, only a sinalternative—a single assignment—will remonce the referring expression in object posihas been reached.

In principle, this last claim could be cawithin the structure-projecting account; wheforthcoming noun phrase is projected atverb, the referential processes that estabanaphoric dependencies between that projeand the context may establish dependenciemore than one thematically appropriate altetive. At issue, however, is themechanismbywhich such dependencies are computed.predictive contingency account provides suc
mechanism.

antivheatichur-. Fe tctlly“th-n-forbeou

iblybleessesib

rgeiva

suborhowtheennsgicncinsre,outheo-

ax-ureenre:ped

bther

taskthe

ly,ncestextatals inthee-oneaskhav-95)ory, its, thex-

d forda-al.

t thein

tivetaskses.

ter-in-

thenyelf.

onsisob-

tothethe

aticallystedandat if” innlyartsner


Recently, Liversedge, Pickering, Branigand van Gompel (1998) proposed an alternamechanism for contextual influences on tmatic role assignments. They proposed thcontext could introduce constituents whwere sufficiently underspecified as to invite fther specification in a subsequent sentenceinstance, in “the gardener wondered wherplant the shrubs,”whereis assigned an adjunLocative role, but is otherwise semanticavacuous. Subsequently, on encounteringshrubs were planted by . . .,” the earlier Locative role is still available given its earlier uderspecification, and the by-phrase is thereinterpreted as a locative (ordinarily it wouldinterpreted as an agentive). The present accgoes further in that, with the exception possof the filled-role constructions, the plausiantecedents were not assigned thematic rolthe context which then recurred in the subquent targets. These antecedents were plaurecipients of roles associated with the taverb, but they had not already received equlent roles in the context.

The account developed here rests quitestantially on the claim that the data reflect nmal comprehension processes. There are,ever, two broad reasons for interpretingpresent data with caution. First, and as mtioned at the outset, the stop-making-sejudgment task could well tap late or strateprocesses which do not reflect normal sentecomprehension. In this case the data woulddicate that peoplecan assign thematic rolepredictively, but only in circumstances whefor example, it was strategically advantageto do so. It is unclear, nonetheless, whymore naturalistic reading replications of the Bland et al. (1995) data (cf. Pickering and Trler’s 1998 study and other reading measreported in Boland et al. 1995) should not geralize to the contextual cases described hethe judgment task in Boland et al. (1995) tapearly, nonstrategic, processes as evidencethe alternative measures, why should itnot tapsuch processes when the sentences are emded in context? One reason might be thattarget sentences in Experiments 1 and 2 w
distinguished from the context sentences tha
,e-a

oro

e

e

nt

in-let-

--

-

-e

e-

s

s-if

dby

ed-ee

preceded them in terms of presentation andrequirements and that this compromisedgeneralizability of the task. This would imphowever, that the same task applied to sentepresented in the absence of any explicit con(with all the referential infelicities which thimplies) is more representative of normprocessing than when applied to sentencecontext (however distinct). Nonetheless,judgment dataare open to alternative interprtations, even if, as stated at the outset,reason for the adoption of the judgment there was to demonstrate that the same beioral phenomena found by Boland et al. (19could be found in the absence of obligatsyntactic dependencies between the verbarguments, and their antecedents—indeedproportions of “no” responses obtained in Eperiment 2 are comparable to those observethe equivalent constructions (transitives,tives, filled-gap) in the original Boland etstudies.

The second caveat concerns the fact thadata for the different constructions employedExperiments 1 and 2 are open to alternainterpretations, even supposing that thedoes reflect normal comprehension procesFor example, the filled-role data could be inpreted as reflecting anaphoric processesvolved in the identifying the antecedent ofpostverbal pronoun, without postulating apredictive role assignments at the verb itsThe case for predictive assignments restsconvergenceof the data on a single hypothecapable of explaining the range of patternsserved in Experiments 1 and 2.

While further research is clearly neededboth verify and extend the present results,account that has been developed to explainmechanism by which context can drive themrole assignments makes a variety of empirictestable predictions other than the ones tehere. For example, Allopenna, Magnuson,Tanenhaus (1998) have demonstrated thparticipants are asked to “pick up the candya real-world context in which candy is the oitem in front of them whose spoken name stwith /k/, their eyes move to the candy soo
tthan if a competitor item whose name also starts

asghin-of

nernon

ghtc--seemonncanTho

rie

tenencntenon

ase pat h

e aforved

eplargt h

bypuuipeme

antlaneatqu

een

cted/old.

wayde-ogs.

very

t anderenes/rner.

e gotomeen on

ardsty. Itthick

herd ainto

us-om,

very

theveral. She

bytwooked/

d toor ashed/y.

ideolate.

ess-nd at had

outrow

ame

ept


with /k/ (e.g.,candle) is present in the contextwell. The present account makes the straiforward prediction that if participants werestructed to “eat the candy,” the probabilityfixating on the candy would rise at the verbeat,with no equivalent rise in fixation probability othe competitor candle (and the opposite pattwith an increase at the verb in fixationscandleshould arise in response to, e.g., “lithe candle”). If it could be shown that seletional restrictions couldnot influence eye movements in the predicted fashion, constraint-batheories of sentence processing would thselves require considerable theoretical cstraint. The challenge for any theory of senteprocessing is to understand the nature of,constraints on, thematic role assignment.data presented here constitute just one mconstraint on the development of these theo

APPENDIX 1: THE EXPERIMENTALITEMS USED IN EXPERIMENT 1

Each item consists of three sentences. The first senis identical across the four conditions. The second sentintroduces a plausible antecedent or an implausible acedent, and the final target sentence contains either aselecting or a selecting verb.

A young toddler was running around his playroom. It wempty except for some chairs in one corner and somcats/flowers in the other. He bumped/chased a chair thhad run into before.

A pop star carrying only a microphone walked on stagthe start of his concert. The stage was bare exceptcouple of guitars/loudspeakers and two stools. He moplayed a stool with a fair amount of effort.

A faith healer was at work in his room. It was bare excfor a couple of patients/incense burners and somecandles in the corner. He touched/healed a candle thaa crack in it.

A soldier fired some live ammunition into a roommistake during a training exercise. Several officers/comers were inside with a lot of sophisticated monitoring eqment. He checked/wounded some equipment but it seto work OK.

A surgeon entered an operating theatre to transplkidney from one patient to another. Except for the transppatients/operating tables and a row of scalpels, the thwas empty. He examined/reassured a scalpel that wasblunt and useless.

A keen amateur gardener was clearing out his gr
house. Soon, all that was left were a few small plants an
t-

,

d--ederes.

cee--

ete

ta/

te

ad

t--d

atreite

-

some old knives/stone slabs on the floor. He inspesharpened a plant that looked particularly stunted and

A policeman was sent to some cellars where two runagirls were reportedly hiding out. The cellars seemedserted except for the girls/some rats and a few stray dHe ignored/handcuffed a dog that was sniffing aroundnoisily.

A repairman was called out to a house late one nighwas directed to the utility room. All he could see there wa few plants and two dilapidated looking washing machicats. He moved/repaired a plant that was lying in the co

Sally was having a comfortable evening at home. Shup to get some wine/magazines and to look through svideo cassettes. She chose/sipped a video that had bespecial offer.

A dog was sniffing around inside an open cupbolooking for food. There wasn’t much inside; just some rupipes and an open tin of dog food/a bag of old nailsnudged/devoured some pipes that were covered inmould.

Cathy was preparing breakfast. There wasn’t much infridge except for a bag of mushrooms/jug of milk anbottle of water. She got/fried some water and poured ita glass.

A policeman walked into a room and looked round spiciously. There were just a couple of men/bags in the rolying on some rugs. He noticed/arrested a rug with alarge blood stain.

A nurse carrying various syringes walked into one ofrooms off the main ward. She was surprised to see sewomen/dolls and some garden gnomes lying on the bedlifted/injected a gnome and carried it to a chair.

A mother carrying a bottle of milk went into the baroom. It was empty except for a playpen in which laytiny babies/teddy bears and some toy animals. She strfed a toy that she was given as a child.

A restaurant chef opened up his fridge as he startemake lunch. He was surprised to find it empty except fpile of spoons and a few potatoes/empty dishes. He wabaked some spoons that were particularly old and dirt

Tommy was thirsty. He opened the fridge, but insthere was just some cake and some apple juice/chocHe tried/drank some cake but thought it had gone off.

A rock musician was getting ready backstage. His dring room was empty except for some guitars/towels afew photographs. He dropped/tuned a photograph thabeen handed to him.

A car was driving downhill when it suddenly veeredof control. In its path were some pigeons/dustbins and aof bollards. It missed/injured several bollards that cclose to being destroyed.

Mike was looking through his larder. It was empty exc
dfor a variety of old broken cups and some stale biscuits/

int

le o/ate

gunld ticke

ingagetice

ngsgeso

reat/c

edreed

ranoye

bp th

eesd th

ularnotom

oinhis

bee

onts. Hble

illagtwht/

lled

. Herkedome

. Shewasorem

lderinhis

the

tedasthe

dis-rrietthedo.

utedt thearty

nderThehim.ften.

Hectorered

Heonfi-als.

er.

withe to

ro-ed an

ec-il./ASheques-

post.ver-


mouldy boxes. He took/ate some cups and went backthe kitchen.

John opened up his tool bag. There were just a coupspanners in it and several biscuits/screws. He removedspanner that was covered in thick grease.

A game hunter entered a small clearing, with hisraised and ready. There were just a few rats/some ocans in the center, and some bones lying nearby. He kikilled a bone that he thought was a snake.

A music student was practicing a piece for an upcomaudition. She was practicing in front of two large stlights beside her teachers/some loudspeakers. She noimpressed a light that was flickering on and off.

A guest at a dinner party was thirsty. The only thiwithin reach were a couple of jugs of water/flower arranments and some ornate candles. She toppled/pouredcandles that then made a terrible mess.

Mrs. Green was preparing lunch. She looked at the bplates and the jars that were in her cupboard. She gosome jars that were right at the top.

A motorcyclist suddenly lost control of his bike. Hheaded toward several street lights and a group of chilsome railings. He missed/scared a light that he managswerve around.

Giles had just finished dining in an expensive restauHe ordered some brandy/chocolates and cigars. He enjdrank a cigar every so often after his meals.

A student was studying hard. He was surroundedbooks/ashtrays and coffee cups. He grabbed/read a cuhad dropped under his chair.

Paul was preparing his dinner. He’d bought some chand some eggs/biscuits. He ate/boiled some cheese andrank a cold beer.

A professor was waiting to be interviewed on a popscience program. He took a packet of cigarettes/someand a bag of sweets out of his pocket. He inspected/lit ssweets that smelled of mint and liquorice.

Jack was waiting for the school bus. He took some cand some chewing gum/a box of matches out ofpocket. He examined/chewed some coins that hadgiven to him.

A veterinary surgeon was at work in his surgery. In frof him were some cats/notes and some reference bookinspected/vaccinated a book that was lying near his ta

Fred carelessly bumped against a table at the local vfete. There were some delicate vases/bin liners andplastic bottles filled with water on the table. He bougdamaged some water that looked like it was fizzy.

APPENDIX 2: THE EXPERIMENTALITEMS USED IN EXPERIMENT 2

There were three sets of constructions; datives, fi
role, and object control.
o

fa

nd/

d/

-me

d/ut

n/to

t.d/

yat

een

ese

s

n

e.

eo

-

Datives. The first context–target pairconstitutes the plausible antecedentcondition and the second the implausibleantecedent condition.

Hank parked his van outside the local military basedelivered some machine guns to them and left./Hank pahis van outside the preschool nursery. He delivered smachine guns to the military base next door.

Carol was sitting beside an undergraduate studentexplained an intricate theorem to him in detail./Carolsitting beside a toddler. She explained an intricate theto his father.

Churchill had always been very impressed by the estatesman. Churchill attributed the privacy law to him1933./Churchill had always been very impressed byarmy cook. Churchill attributed the privacy law tobrother of the cook.

The city hospital was doing very well. People donamuch more blood to it this year./The political party wdoing very well. People donated much more blood tohospital following its campaign.

Harriet stood up in front of the uneasy pupils. Shetributed the science exams to them straight away./Hastood up in front of the car salesmen. She distributedscience exams to their children to see how they would

John arrived late at the campus party. He contribsome cheap liquor to it that evening./John arrived late apublic library. He contributed some cheap liquor to the phe went to later that night.

The chief executive had a junior assistant working uhim. He delegated the routine typing to him quite often./chief executive had a senior manager working underHe delegated the routine typing to his secretary quite o

A doctor was in his surgery checking a patient.administered some pain killers to her very gently./A dowas in his surgery checking his goldfish. He administsome pain killers to a patient when she came in.

A barrister was in discussion with some lawyers.entrusted some important documents to them in cdence./A barrister was in discussion with some criminHe entrusted some important documents to their lawy

A manufacturer of hi-tech microscopes was chattingtwo biologists. He demonstrated an electron microscopthem very effectively./A manufacturer of hi-tech micscopes was chatting with two cleaners. He demonstratelectron microscope to some people later on.

A witness to a violent robbery was sitting with a dettive. She described the armed thieves to him in detawitness to a violent robbery was sitting with her baby.described the armed thieves to the detective that wastioning her.

The President had to select someone for an honoraryHe appointed a young student to it somewhat contro
sially./The President had to select someone for the vice

atio

intrhaba

HeTedhis

ith

icely./A. H

mein

ttoi

Heaysnew

d owd

geo

SheancSh

sly.

ce../Thted

ritehrisuriteale

iliesonthethe

somhalbo

. He/Jefe h

ned

cher

mewas

ond

thehistheir

heourts to

ttlethewas

urseforetheher

sta-fore

andfore

r icewasto her

ves

hisout

ut aroom

e in-

to/heir

ith/other

the


presidency. He appointed a young student to an educcommittee and then made his decision.

Suzanne has a very temperamental husband. Sheduced the new ambassador to him at dinner./Suzannevery temperamental parrot. She introduced the new amsador to the original owner of the parrot.

Ted was walking along with his college advisor.expressed his political ambitions to him rather shyly./was walking along with his two-year-old. He expressedpolitical ambitions to his mentor who was walking wthem.

A doctor was looking up the address of the local hospHe referred a dying patient to them somewhat reluctantdoctor was looking up the address of the local plumbersreferred a dying patient to that plumber’s homeopath.

Jack knew the local farm very well. He delivered solive sheep to them last week./Jack knew the local swimmpool very well. He delivered some live sheep to the abanext door.

Bill was always very generous with his school friends.lent his new pencils to them this morning./Bill was alwvery generous with the local business. He lent hispencils to the manager’s son this morning.

David enjoyed the swans he encountered. He tossebread crumbs to them if possible./David enjoyed the crohe encountered. He tossed old bread crumbs to the pias well.

Nancy spent the afternoon with her supervisor.showed her doctoral thesis to him rather nervously./Nspent the afternoon with her three-year-old nephew.showed her doctoral thesis to his father rather nervou

The Chief Inspector had two policewomen in his offiHe granted generous maternity leave to them that dayChief Inspector had two policemen in his office. He grangenerous maternity leave to their female colleague.

Harold was wondering what to do about his favouclerks. He awarded a large bonus to each of them at Cmas./Harold was wondering what to do about his favoexhibits. He awarded a large bonus to each of the speople responsible for them.

The local priest was always very kind toward the famhe worked with. He gave a warm smile to all the familiesSunday./The local priest was always very kind towardcharities he worked with. He gave a warm smile to allcharity workers on Sunday.

Janet was watching an elephant closely. She tossedfresh hay to it rather nervously./Janet was watching a wclosely. She tossed some fresh hay to the cattle on the

Jeff managed to visit the last companies on his listgave his latest drawings to them for consideration.managed to visit the last monuments on his list. He gavlatest drawings to the architects for consideration.

Henry spent a lot of time with his neighbour. He loa
his sharp axe to him this morning./Henry spent a lot of time
n

o-s as-

.

e

gr

ldsns

ye

e

t-

s-

eeat.

fis

with his child. He loaned his sharp axe to his child’s teathis morning.

Tom was negotiating with a local client. He sold sodiamond earrings to her and also a silver necklace./Tomnegotiating with a traffic warden. He sold some diamearrings to a client whose car she was ticketing.

Ken fulfilled his promise to the terrorists. He gavelarge ransom to them on the following day./Ken fulfilledpromise to his daughters. He gave the large ransom tokidnappers at midnight.

Tim was in court with his two ex-wives. He offered talimony payments to them as advised./Tim was in cwith his two brothers. He offered the alimony paymenthis ex-wife as advised.

Andrea wrote a card for the infant. She sent a little rato him when she was in Ohio./Andrea wrote a card forpolitician. She sent a little rattle to his baby when shein Ohio.

A doctor was rummaging inside his case while the nlooked on. He handed the syringe needle to her beleaving./A doctor was rummaging inside his case whilebaby girl looked on. He handed the syringe needle tomum before leaving.

Amelia had included her niece in her last will and tement. She willed her wedding gown to the adolescent bedying./Amelia had included her nephew in her last willtestament. She willed her wedding gown to his wife bedying.

Alice was surrounded by playmates. She handed hecream to them and put her hands in the air./Alicesurrounded by gangsters. She handed her ice creamlittle child and put her hands in the air.

Filled-Role Constructions. The first alternatilisted in the context sentence correspondto the plausible antecedent condition andthe second to the implausible antecedentcondition.

Mike’s grandparents heard him talking to/aboutdaughter. He reminded them to take their grandchildtonight.

The team saw the Olympic coach arguing with/abofemale athlete. He persuaded them to leave thestraight away.

We overheard Sue talking to/about her brother. Shvited us to drive her to the pub.

The young woman listened to the dean being rudeabout the boys. He instructed her to avoid following tbad example.

The film producers heard the director in discussion wdiscussing an actress. He forced them to provide anone pretty quickly.

You watched the nurse querying the drugs with/for
patient. She asked you to help count all the tablets.

boufen

outes

lice

tod b

to/e th

bound

. Sh

eir

f th

ingler.

thern.

s o./H

her.

Heh

thet./H

thaha

ploy

andnot

theive

hend

and

sometry

ogog a

efts.her

thero tell

tionuch-

teders to

. K.rec-inu--

ra,-

90).for-d

arl-sing

C J.ron--ters,

C elpym-:

E

F nce

G M.nceh,


You heard the detectives asking a question to/athe suspect. They invited you to help the suspect dehimself.

You listened to the social worker reporting to/absome parents. He advised you to introduce the changMonday.

Jack stood by as Martha complained to/about the powomen. She asked him to drive her to her home.

James was reading while his mother talked sternlyabout his sisters. She urged him to avoid being temptedrugs.

The children stood around while the doctor talkedabout their mother. He pressured them to let her havoperation.

The squad listened to the coach being supportive to/athe female runner. He encouraged them to follow him rothe track.

You heard the salesclerk shout at/about the vandalstold you to report the vandals right away.

The twins listened to their father talking to/about thmother. He asked them to be especially nice to her.

The boys listened as the wizard warned/warned ogirl. He advised them to leave the city that night.

Mary’s parents listened as Richard spoke so convincto/about her. He persuaded them to keep a diary for h

Object Control/Transitive Constructions

Mike quickly chose a video. He ate his sandwich inTV room./He reminded his brother to get some popco

The football manager was worried about some pieceequipment. He drank his beer and soon calmed downpersuaded his team to avoid them all.

Sue was looking forward to the dance. She ironedskirt and put it on./She invited her friend to go with he

The chief executive was thinking about the report.sacked his secretary for being too sloppy./He askedsecretary to be less sloppy.

The director rewrote the screenplay first thing inmorning. He fried some eggs and had his breakfasforced the actors to work over Christmas.

The nurse was rushed off her feet with all the choresneeded seeing to. She inoculated some children whojust arrived./She implored the administrators to emmore staff.

The lawyer toyed with his mug. He bit his sandwichchewed it noisily./He instructed his client to pleadguilty.

The policeman went up to the door. He handcuffedthief and drove him away./He convinced the thief to ghimself up.

Martha was disappointed with the film. She sippedwine and watched another film./She persuaded her frie
get another one.
td

on

-

/y

e

ut

e

e

y

fe

r

is

e

td

rto

Jane quickly washed a large plate. She bit her lipstarted to cook./She urged her child not to drop it.

The salesman described the new car. He chewedgum all through his talk./He pressured some clients toit out.

The team coach liked the toffee with the team lprinted on it. He sharpened his pencil while chewintoffee./He encouraged the squad to adopt the logo.

The salesclerk was talking about the latest spate of thShe licked her ice cream while she talked./She toldmanager about them straight away.

The children listened to the story. They kissed their faand asked for another./They persuaded their father tthem another.

The sorcerer looked at the crystal. He drank the pohe’d made with it./He advised his assistant to avoid toing it.

Jack looked in the direction of the chair. He vaccinathe dog that was asleep there./He persuaded its ownhave it cleaned.

REFERENCES

Allopenna, P. D., Magnuson, J. S., & Tanenhaus, M(1998). Tracking the time course of spoken wordognition using eye movements: Evidence for contous mapping models.Journal of Memory and Language,38, 419–439.

Altmann, G. T. M., Garnham, A., van Nice, K., & HenstJ. A. (1998). Late closure in context.Journal of Memory and Language,38, 459–484.

Boland, J. E., Tanenhaus, M. K., & Garnsey, S. M. (19Evidence for the immediate use of verb control inmation in sentence processing.Journal of Memory anLanguage,29, 413–432.

Boland, J. D., Tanenhaus, M. K., Garnsey, S. M., & Cson, G. N. (1995). Verb argument structure in parand interpretation: Evidence from wh-questions.Jour-nal of Memory and Language,34, 774–806.

ohen, J. D., MacWhinney, B., Flatt, M., & Provost,(1993). PsyScope: A new graphic interactive enviment for designing psychology experiments.Behavioral Research Methods, Instruments, and Compu25, 257–271.

rain, S., & Fodor, J. D. (1985). How can grammars hparsers? In D. Dowty, L. Karttunen, & A. Zwick(Eds.),Natural language parsing: Psychological, coputational, and theoretical perspectives.CambridgeCambridge Univ. Press.

lman, J. L. (1990). Finding structure in time.CognitiveScience,14, 179–211.

odor, J. D., & Frazier, L. (1980). Is the human senteparsing mechanism an ATN?Cognition,8, 417–459.

arnsey, S. M., Tanenhaus, M. K., & Chapman, R.(1989). Evoked potentials and the study of sentecomprehension.Journal of Psycholinguistic Researc
18, 51–60.

nceatioche.

as

G Mionge o

G singo-

G othd

L vananby--

. Ses-

M in

M 5).

M ents

P pro-ok.

P singe

P forrma-for

P rs-

S tialC.

sJ:

S

T s aam-.

sing

Z ten-

(


Garrod, S. C., & Sanford, A. J. (1994). Resolving sentein a discourse context: How discourse representaffects language understanding. In M. A. Gernsba(Eds.),Handbook of Psycholinguistics(pp. 675–698)San Diego: CA: Academic Press.

Gernsbacher, M. A. (1990).Language comprehensionstructure building.Hillsdale, NJ: Erlbaum.

ernsbacher, M. A., Hargreaves, D. J., & Beeman,(1989). Building and accessing clausal representatThe advantage of first mention versus the advantaclause recency.Journal of Memory and Language,28,735–755.

ibson, E., & Hickok, G. (1993). Sentence Proceswith Empty Categories.Language and Cognitive Prcesses,8, 147–161.

orrell, P. (1993). Evaluating the direct association hypesis: A reply to Pickering and Barry.Language anCognitive Processes,8, 129–146.

iversedge, S. P., Pickering, M. J., Branigan, H. P., &Gompel, R. P. G. (1998). Processing argumentsadjuncts in isolation and context: The case ofphrase ambiguities in passives.Journal of Experimental Psychology: Learning, Memory and Cognition,24,461–475.

MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M(1994). The lexical nature of syntactic ambiguity rolution. Psychological Review,101,676–703.

arslen-Wilson, W. D. (1987). Functional parallelismspoken word recognition.Cognition,25, 71–102.

auner, G., Tanenhaus, M. K., & Carlson, G. N. (199Implicit arguments in sentence processing.Journal ofMemory and Language,34, 357–382.
(
snr

.s:f

-

d

.

urphy, G. L. (1984). Establishing and accessing referin discourse.Memory and Cognition,12, 489–497.

ickering, M. (1993). Direct association and sentencecessing: A reply to Gorrell and to Gibson and HickLanguage and Cognitive Processes,8, 163–196.

ickering, M., & Barry, G. (1991). Sentence proceswithout empty categories.Language and CognitivProcesses,6(3), 229–259.

ickering, M. J., & Traxler, M. J. (1998). Strategiesprocessing unbounded dependencies: Lexical infotion and verb-argument assignment. [submittedpublication]

ritchett, B. L. (1992).Grammatical competence and paing performance.Chicago: Univ. of Chicago Press.

pivey-Knowlton, M., & Tanenhaus, M. (1994). Referencontext and syntactic ambiguity resolution. InClifton, L. Frazier, & K. Rayner (Eds.),Perspectiveon sentence processing(pp. 415–439). Hillsdale, NErlbaum.

towe, L. (1986). Evidence for on-line gap location.Lan-guage and Cognitive Processes,1, 227–245.

rueswell, J. C., & Tanenhaus, M. K. (1994). Towardlexicalist framework of constraint-based syntacticbiguity resolution. In C. Clifton, L. Frazier, & KRayner (Eds.),Perspectives on sentence proces(pp. 155–179). Hillsdale, NJ: Erlbaum.

witserlood, P. (1989). The locus of the effects of sential-semantic context in spoken-word processing.Cog-nition, 32, 25–64.

Received June 23, 1998)Revision received January 19, 1999)

Documents

Thematic Role Assignment in Context · 2019. 4. 1. · Thematic Role Assignment in Context Gerry T. M. Altmann University of York, Heslington, York YO10 5DD, United Kingdom Two experiments