t be

Le

ng H

n, T

Dece 23

This study examines the comprehension patterns of various sentence types by Mandarin-speaking aphasic patients and evaluates thevalidity of the predictions from the Trace-Deletion Hypothesis (TDH) and the Double Dependency Hypothesis (DDH). Like English,

ing aphasic patients comprehension on various types of

revealed that the comprehension of these patients may

1980, among many others; for German and Italian,

Garrett, & Zurif, 1980) and the complete loss of syntac-tic competence approach (e.g. Caramazza & Zurif, 1976;Caplan & Futter, 1986). However, these two approachesrun into diculties in the face of evidence that (1) com-prehension decits and telegraphic production may notnecessarily co-occur (e.g. Miceli, Mazzucchi, Menn, &

* Corresponding author. Fax: +886 3 5725994.E-mail address: ycsu@mx.nthu.edu.tw (Y. Su).

Brain and Language 101sentences in order to investigate the validity of the pre-dictions from two linguistically based hypothesestheTrace-Deletion Hypothesis (TDH) and the DoubleDependency Hypothesis (DDH). In neurolinguistic litera-ture, the term agrammatism was initially used to referto a selective disorder of speech production (i.e. sparseverbal output, disuency, and omission of functionalmorphemes) but intact comprehension resulting fromfocal brain damage. Since Caramazza and Zurifs(1976) seminal work, this traditional view has been chal-lenged, and more and more cross-linguistic studies have

Bates, Friederici, & Wulfeck, 1987; for Japanese, Hagiw-ara, 1993; Hagiwara & Caplan, 1990; for Mandarin Chi-nese, Su & Law, 1993; for Serbo-Croation, Lukatela,Shankweiler, & Crain, 1995; for Spanish, Beretta et al.,2001; Miera & Cuetos, 1998; for Cantonese, Law & Leu-ng, 1998, 2000; for Hebrew, Friedmann, 2000; for Kore-an, Beretta et al., 2001; OGrady & Lee, 2001, 2005). Inthe past three decades, various approaches have beenproposed to account for agrammatic comprehension dif-culties, including the phonological/morphological com-ponent decit approach (e.g. Kean, 1977; Bradley,the canonical word order in Mandarin is SVO, but the two languages dier in that the head noun precedes the relative clause in English,but it follows the relative clause in Chinese. According to the Default Principle as stated in the TDH, the word order discrepancy willmake subject relative clauses more dicult to comprehend for Mandarin agrammatics than object relative clauses, but the DDH predictsthat agrammatic patients from the two languages have the same pattern of selective decits. The results of this study support the pre-diction of the TDH. 2006 Elsevier Inc. All rights reserved.

Keywords: Aphasia; Agrammatism; Comprehension; Mandarin Chinese; Trace-Deletion Hypothesis; Double Dependency Hypothesis

1. Introduction

This study aims to examine Mandarin Chinese-speak-

also be impaired, especially when the crucial cues tointerpret the sentences are syntactic (e.g. for English,Heilman & Scholes, 1976; Schwartz, Saran, & Marin,Asyntactic thematic role assignmenTrace-Deletion Hypothesis and th

Yi-ching Su a,*, Shu-era Institute of Linguistics, National Tsi

b Department of Physical Medicine and Rehabilitatio

Accepted 1Available onlin

Abstract0093-934X/$ - see front matter 2006 Elsevier Inc. All rights reserved.doi:10.1016/j.bandl.2006.12.001y Mandarin aphasics: A test of theDouble Dependency Hypothesis

e b, Yuh-mei Chung b

ua University, Hsin-chu, Taiwan 300

aipei Veterans General Hospital, Taipei, Taiwan 112

ember 2006January 2007

www.elsevier.com/locate/b&l

(2007) 118

ho inmatism

o n

non-moved NPs receive the theme role through the verb,

a

it

d

Language 101 (2007) 118(3) The S-structure representation underlying agrammaticcomprehension lacks traces. In interpretation, aDefault Principle is invoked that is dened as follows:If a lexical NP has no theta-role (that is, it is in a non-thematic position), assign it the theta-role that is canon-ically associated with the position it occupies, unlessthis assignment is blocked. In this case assign it a rolefrom the next lower level in the Thematic Hierarchy.

Based on the TDH, none of the moved elements (i.e. thecat in (1ac) and the dog in (2b and c)) in passives, rel-ative clauses, and cleft sentences in (1) and (2) canreceive thematic roles because there are no traces totransmit the roles to the NPs. Since these moved NPsare in a position that precedes another NP in the clause,they are assigned the agent role based on the DefaultPrinciple. The assignment results in conicting represen-tations for the sentences in (1) as the other NP receivesthe agent role via either the preposition by or the verb.The chance-level performance is the consequence ofguessing between the two NPs, both of which now bear

the athegenDef(Gr dzi

are intact except in the following two respectssky, 1990, p. 97).Acc rd g to the TDH, syntactic representations in agram-c. Subject cleft: It was the dog that [t] chased the cat.

t

ubject-extracted relative: The dog that [t] chasede cat was big.(2) a. Ab. Sctive: The dog chased the cat.Goodglass, 1983), (2) these patients are aware of themeaning and the presence/absence of many functionalmorphemes in well-formedness judgment tasks (e.g. Line-barger, Schwartz, & Saran, 1983; Lukatela, Crain, &Shankweiler, 1988), and (3) they show a word positioneect like normal adults do in an on-line processing task(Shankweiler, Crain, Gorrell, & Tuller, 1989). In the cur-rent study, we will focus our discussion on accounts thatposit partial loss of certain aspects of syntactic process-ing as the locus of agrammatic comprehension dicul-ties. This was the position rst taken by Grodzinsky(1986, 1990, 1995).

Based on the syntactic theory (i.e. Government andBinding theory) of Chomsky (1981), Grodzinsky (1986,1990) proposed the Trace-Deletion Hypothesis to accountfor agrammatic patients chance-level performance in com-prehending sentences with non-canonical ThemeAgentorder as in (1), in contrast to their (near) normal perfor-mance on sentences with canonical AgentTheme orderas in (2).

(1) a. Passive: The cat was chased [t] by the dog.b. Object-extracted relative: The cat that the dogchased [t] was small.

c. Object cleft: It was the cat that the dog chased [t].

2 Y. Su et al. / Brain andt role. For the sentences in (2), the assignment ofault Principle causes no problems because theHowever, there still exist some conditions which elicitedrelatively higher error rates1 from patients grammaticalityjudgments, such as the agreement between the subjects orthe auxiliary verbs of a tag question and its host sentence,as shown in (7).

(7) a. *The little boy fell down, didnt it?b. *John is very tall, doesnt he?

Other judgment tasks which caused diculty for agram-matic patients also involve some kinds of agreement aswell, e.g. Wh-head agreement (as in (8)), headhead agree-ment (i.e. misselection of auxiliaries as in (7b) and (9)), vio-lations of gender or number in pronouns and reexives (asin (10)).1 Thes in te agrahese co. Who __ thought __ was going to get the job?

job*.

. Th ts who Frank thought __ was going to get the

. F ank thought __ was going to get the job.

a. Frank thought he was going to get the job.b * r(5) Wh-moved subcategorization (83.1% correct)a. *The principal frowned the boy.b. *Who did the principal frown?c. Why did the principal frown?

(6) Empty elements (83.7% correct)

c astra ed that chain formation was intact in agrammatics.

(as n (5)) and empty elements (as in (6)), which demon-

correctly judge constructions involving Wh-movement

In ddition, agrammatics also preserved the ability tod. The boy was following the girl.e. John has nally kissed Louise.c. *The boy was followed the girl.(4) a. John was nally kissed Louise.b. The boy was followed by the girl.Shankweiler et al., 1989). Take the active and passive sen-tences in (4) as examples. Linebarger (1989) found thatagrammatic patients were able to discriminate the gram-matical sentences from the ill-formed ones (4a and c) witha passive participle followed by a direct object.

*et a ., 1 88; Schwartz, Linebarger, Saran, & Pate, 1987;

barger, 1989, 1990; Linebarger et al., 1983; Lukatelaand hence the representation matches the correct inter-pretation of the sentences.

Although agrammatic aphasic patients have been shownto demonstrate selective comprehension decits on someconstructions but not others, several studies also exhibitthat they nevertheless can accurately judge the grammati-cality of the constructions they fail to comprehend (Line-

l 9mmatic patients tended to over-accept the ill-formed sentenc-nditions.

is strabilitcultigramture csuggeparse

otitio

CoIfsam

mintur

(13) a. [The boy] was chase + en t by [the girl].

de e4 nn et j

Language 101 (2007) 118 3Assustruc

tive clarespectivg the VP-Internal Subject Hypothesis, thees of the passive and the object-extracted rela-e R-index.

a is R-dependent on b, then they must bear the(12) indexation ConditionCond n.

do n have, or cannot make use of, the Coindexationtion between NPs and their traces as evidenced inwell-formedness judgments), they are not able to makethe mapping from grammatical functions to thematicroles. Further elaborating Linebarger and her colleaguesMapping Hypothesis, Mauner, Fromkin, and Cornell(1993) proposed the Double Dependency Hypothesis(DDH) to account for agrammatic patients problemswith assigning thematic roles and judging agreementviolations.

In order to account for the comprehension dicultiesagrammatic patients display on thematic role assignment,the DDH posits a selective decit as stated in the following(Mauner et al., 1993, p. 349).

(11) The Double Dependency Hypothesis(i) The decit underlying asyntactic comprehension

aects the processing of syntactic referential de-pendencies, and

(ii) When there is only one such dependency the re-sulting syntactic representation, although abnor-mal, is not ambiguous, but when there are twosuch dependencies the resulting representation issemantically ambiguous.

According to this hypothesis, asyntactic comprehensionwith spared grammaticality judgment arises when subjectsthat the sensitivity to grammatical deformationsong evidence for agrammatic patients preservedy to recover syntactic structure, and that their dif-es on some conditions of comprehension andmaticality judgment all involve noncategorial fea-lashes between anaphorically linked elements. Theyst that although the agrammatics may correctlyan input sentence (including intact chain forma-*George was angry and so did Tom.

(10) a. *The famous man itself attended the ceremony.b. *I helped themselves to the birthday cake.

Based on these ndings, Linebarger and her colleaguesargue(8) Wh-head agreementa. *The pencil who he brought was nice.b. The pencil which he brought was nice.

(9) Headhead agreement

Y. Su et al. / Brain anduse are as depicted in (13a) and (14a),ely.(15c), the correct indexing in (15b) is assured by therequirement that syntactic clauses be interpretable evenin the agrammatics syntax. As with active declaratives,although the indexing in (16c) for subject-extractedative clauses do notmove, and hence are assigned theta-rolesdirectly, unaected by the loss of the Coindexation Condi-tion. The examples in (15a) and (16a) represent the structuresfor an active and a subject-extracted relative clause respec-tively under the VP-Internal Subject Hypothesis.

(15) a. [The boy] is t chasing [the girl].b. [the boy]i, ti, [the girl]jc. [the boy]i, tj, [the girl]j

(16) a. [The boy] [Op that t1 is t2 chasing [the girl]] is tall.b. Opi, t2i, [the girl]jc. Opi, t2j, [the girl]j

Both the active and the subject-extracted relative clauseinclude a single thematic R-dependencythe chain [theboy], t in the active, and Op, t1, t2 in the subject-ex-tracted relative clause. Suppose the R-index of the boyis i and that of the girl is j. Under a coindexation def-icit, the traces are not necessarily coindexed with theboy in the active or the Wh-operator Op in the sub-ject-extracted relative clause. For the active, the index-ing in (15b) is correct, whereas the indexing in (15c)results in both theta-roles being associated with thegirl, while the boy has no role at all. Since no interpre-tation can be assigned to the orphaned NP the boy in2 Oprelative repprondirec ob ects in active declaratives and subject-extracted rel-

U lik passives and object-extracted relative clauses, thein (1 b a d c).

to th th ta-positions occupied by the two traces as shown

Coin exation Condition, indices can be arbitrarily assignedb. [the boy]i, ti, [the girl]j, -enjc. [the boy]i, tj, [the girl]j, -eni

(14) a. [The boy] [Op that [the girl] is t1 chasing t2] is tall.2

b. Opi, t2i, [the girl]j, t1jc. Opi, t2j, [the girl]j, t1i

In each of the structures there are two thematic R-depen-denciesin the passive, [the boy], t and [the girl], -en,and in the object-extracted relative clause, Op, t2 and[the girl], t1. For the passive, suppose the R-expressionthe boy is marked with R-index i, and the girl with R-indexj. There will be two possible assignments of indices to thedependent elements as shown in (13b and c). The Coindex-ation Condition will rule out (13c) for normal subjects, butsince the agrammatic patients cannot make use of this con-dition, both assignments will be permitted. Object-extract-ed relative clauses are treated similarly, i.e. without theresents a null Wh-operator that takes the place of the overtoun.

relative clauses is allowed by the agrammatics gram-mar, it leads to a deviant interpretation and thus isrejected.

In addition to the accounts provided by the TDH andthe DDH to explain agrammatic patients selective compre-hension decits among the sentence types, a further predic-tion made by the TDH is that agrammatic aphasics shouldbe able to perform well on comprehension of the matrix

ernal

chance in cases where there is only one NP in the sen-tence that is available for interpretation as a givenunsatised theta-role (e.g. active, subject cleft, and sub-ject-extracted relative as in (18)). For the three types ofsentences in (18), the theme role is grammatically spec-ied by the direct object, but the agent role is unsatis-ed as the trace is not present to transmit the role tothe subject NP.

(18) a. The dog [VP * chased the cat].5

chase (* (cat))b. It was the dog that [VP * chased the cat].

chase (* (cat))c. The dog [that [VP * chased the cat]] is big.chase (* (cat))

hen there is more than one NP that is available for inter-retation as a given unspecied role, comprehension will beoor, e.g. passive, object cleft, and object-extracted relative

( )

a

A cr

perfo

(2

4 Y. Su et al. / Brain and Language 101 (2007) 118Subject Hypothesis (Koopman & Sportiche, 1991).The RTDH predicts that comprehension is above

3 As pointed out in Footnote 12 of Hickok, Zurif, and Canseco-Gonzalez (1993), the RTDH (and TDH) is intended as a descriptiveaccount of the data and not a theory of the actual decit of the aphasicpatients, and the term trace deletion is used as a convenient way ofsaying that antecedent-trace relations are not being established.4 The VP-Internal Subject Hypothesis states that grammatical subjects

are originally generated inside the verb phrase (VP) and then raised tota-roles via a trace according to the VP-Int4their surfacposition ofe positiextracture) ss s that grammatical subjects receive(RTDH), which, retaining the dening feature oTDH (i.e. traces are deleted at the level of S-sclauses for the relative construction in (1b) and (2b),repeated below as (17a and b), respectively.

(17) a. The cati [that the dog chased ti] was small.b. The dogi [that ti chased the cat] was big.

That is to say, although agrammatic aphasics may havediculty interpreting who is chasing whom in an ob-ject-extracted relative clause such as (17a), they shouldhave above-chance performance on comprehension of asubject-extracted relative clause as in (17b) and the ma-trix clauses of both (17a and b). However, the resultsin Caramazza and Zurif (1976) showed that Brocasand Conduction aphasics performed poorly in a sen-tencepicture matching task when presented withsemantically reversible and improbable center-embedded(i.e. object-extracted) relative sentences paired with asyntactic distractor that reversed the thematic roles ofthe sentence and the adjective assignment. In addition,Sherman and Schweickert (1989) used active, passive,and center-embedded object relative sentences to pairwith either a thematic role reversal distractor, a predi-cate-adjective relation reversal distractor that changedthe adjective assignment, or a thematic role plus pred-icateadjective relation reversal distractor. The resultsprovided further evidence that agrammatic aphasicswere as poor at determining adjective assignment(54.2% correct) as they were at determining thematicroles (41.7% correct) for center-embedded relative sen-tences, but they correctly interpreted most active andpassive sentences. Based on these ndings, Hickok(1992) proposed a Revised Trace-Deletion Hypothesis

f thetruc-

,3 a ume the-on inside the inection phrase (IP), leaving a trace in thtion.erepresentation, Hickok uses a thematic assignment representation (TAR)with the form verb (x (y)), where x denotes the theta-role assigned to thesubject, and y the theta-role assigned to the direct object.6 Following Grimshaw (1990), Hickok assumes that the external

argument of the passive form of a verb is suppressed, as represented bythe symbol . Besides, it is also assumed that the by-phrases are not theta-marked and do not satisfy argument structure positions, and hence thereare two NPs available for interpretation as a given unspecied theta-rolefor a passive sentence. However, it is noted that a fair amount of variationin performance on passives might be expected due to the availability of theby-phrase as a potential heuristic cue for interpretation (see Footnote 15 inHickok et al., 1993).7 According to the RTDH, the adjectives of the matrix clauses in (20) donot recare tree asterisk denotes a deleted trace. In addition to the syntactic5 ThVP * VP *is big (*)

b. The cat [that the dog [VP * chased *]] [VP * is small].is small (*)0) a. The dog [that [ chased the cat]] [ is big].rmance on predicate adjective sentences as in (20).7

resides in the prediction of the comprehensionucial dierence between the TDH and the RTDHchase ( (*))6

b. It was the cat [that the dog [VP * chased *]]chase (* (*))

c. The cat [that the dog [VP * chased *]] is small.chase (* (*))ea

iveted*(19) . The cat was chased by the dog.as in 19 .ppWa theta-role in the predicate-adjective constructions and henceas part of the verb (Footnote 13 in Hickok et al., 1993).

relative construction will cause diculties for thesepatients. The TDH diers from the DDH in that it

tive relation of relative constructions as predicted bythe RTDH.

The test consists of seven types of sentences, as shownin (21) and (22).8 It includes four types with canonicalAgentTheme order such as Active, BA object preposingconstruction, Object Relative, and Object Pseudo-Cleft,9

and three types with non-canonical ThemeAgent ordersuch as Passive, Subject Relative, and Subject Pseudo-Cleft. According to the TDH, the four sentence typeswith canonical AgentTheme order should elicit above-chance performance, because after movement, thethematic roles assigned based on the Default Principleconforms to the correct representation (as in (21)). Onthe contrary, the three types with non-canonicalThemeAgent order should have chance-level perfor-mance, as the assignment of the Default Principle con-icts with the correct representations of these sentences(as in (22)).10

8 Like previous studies (e.g. Caramazza & Zurif, 1976; Grodzinsky,1989; Hickok et al., 1993; Sherman & Schweickert, 1989, among manyothers), we used action verbs in relative clauses and other types ofsentences but predicate adjectives in the matrix clauses of the relativeconstruction in order to keep the structures comparable to other studies,and to avoid the presupposition that the event encoded in a relative clausehappens before the event depicted in the matrix clause, which may makethe processing more complicated in a sentencepicture matching task.9 To our knowledge, only cleft sentences, but not pseudo-cleft sentences,

have been used in testing agrammatic comprehension in English. We donot include cleft sentences because, as the example in (i) shows, a subjectcleft sentence is identical to an active sentence with a copula shi to be inthe front, and the corresponding object cleft sentence in (ii) is not anacceptable form in Mandarin.

Language 101 (2007) 118 5postulates trace-deletion in the agrammatic syntacticrepresentation, and it does not adopt the assumptionof the VP-Internal Subject Hypothesis, and hence atmost one trace is involved in those structures whichresult in comprehension problems. In the next section,we illustrate the similarities and dierences withrespect to the predictions of the hypotheses for thecomprehension patterns of Mandarin-speaking aphasicpatients.

2. Predictions of the hypotheses on mandarin agrammatism

The goal of this study is to test the two linguistical-ly based hypotheses (i.e. the TDH and the DDH) onthe comprehension of Mandarin Chinese-speaking apha-sic patients. Like English, the canonical word order inMandarin is SVO, but the two languages dier in thatthe head noun precedes the relative clause in English,but it follows the relative clause in Chinese. SinceMandarin is a language without rich inectionalmorphology or case marking, the similarity anddierence on word order between English andMandarin provide a good contrast to evaluate thevalidity of the two hypotheses in two respects. The rstissue we would like to examine is whether the compre-hension decits of Mandarin-speaking aphasics can bet-ter be accounted for by postulating one trace (as theTDH) or two thematic R-dependencies (as the DDH)in the representations, and whether patients assignthematic roles based on the Default Principle as statedin the TDH. Secondly, we are concerned with whetherIn the agrammatic representation shown in (20), the ma-trix predicate has one unsatised role but there are twoNPs in the sentence that are available for interpretation.Since there is more than one possible NP to ll in theunsatised role, the RTDH predicts poor comprehensionperformance on these sentences. Using a sentencepicturematching task and a truth-value judgment task, Hickoket al. (1993) provides evidence to support the RTDHfrom a case study of an agrammatic aphasic who dis-played poor performance on the passive, the object cleftand the predicate adjective of subject relative construc-tions but nearly perfect performance on the subject cleftsentences.

To recapitulate, the afore-mentioned hypotheses (i.e.the TDH, the RTDH, and the DDH) all correctly pre-dict that in English, agrammatic aphasics would displayabove-chance performance on active, subject-extractedrelative clauses, and subject cleft sentences, butchance-level performance on passive, object-extractedrelative clauses, and object cleft sentences. In addition,the RTDH also predicts that the matrix clauses of the

Y. Su et al. / Brain andMandarin-speaking aphasics will also show poor com-prehension performance on the matrix predicate-adjec-Since pseudo-cleft sentences in English also extract either the subject or theobject NP of a verb to the end of the sentence as in (21d) and (22c), theorder of the two NPs is the same as in Mandarin. Therefore, even withinEnglish, pseudo-cleft sentences can be a good test for the predictions of thetwo hypotheses in comparison with the relative and cleft sentences; that is,the TDH will predict Object Pseudo-Cleft to elicit above-chance perfor-mance and Subject Pseudo-Cleft chance-level performance, whereas theDDH will predict the reverse.10 Like English, the moved NP in the subject position of a passivesentence is assigned the Agent role according to the Default Principle,which conicts with the Agent role of the NP in the by-phrase. As forSubject Relative and Subject Pseudo-Cleft sentences, since the moved NPappears in a position following another NP in the clause, i.e. the secondNP position, it is assigned a Theme role based on the Default Principle asthat is the thematic role canonically associated with the position itoccupies. Therefore, unlike English object-extracted relative clauses, inwhich two NPs bear the Agent role, in Mandarin Subject Relative andSubject Pseudo-Cleft sentences, two NPs receive the Theme role, andhence chance performance is predicted. As pointed out in Footnote 2 ofHickok et al. (1993), given that one theta-role is assigned normally, theaphasic patients should be able to infer that the remaining role goes with

the other NP. However, the results from previous and the current studyshow that patients with comprehension decits are unable to do this.

Lan6 Y. Su et al. / Brain andFor relative construction, since the head noun pre-cedes the relative clause in English but follows theclause in Mandarin, the crucial dierence between thepredictions of the TDH on English and Mandarin isthat subject-extracted relatives will be predicted tocause no diculty for English-speaking agrammaticsbut to elicit chance-level performance for Mandarin-guage 101 (2007) 118speaking patients as in (22b). For BEI passive con-struction, (22a) illustrates the conicting representationsunder the assumption that Mandarin passives arederived in a similar way to English (i.e. the Movementapproach, e.g. Li, 1990). However, if we consider Chi-nese passives as biclausal sentences in which bei is ana-lyzed as a verb taking an embedded clause as its

since no trace is present in the syntactic representation.The second possibility assumes that the VP-InternalSubject Hypothesis also applies in Mandarin (e.g.Huang, 1993), and the Default Principle will assignthe Agent role to the matrix subject NP, which con-icts with the correct representation, and hencechance-level performance will be predicted.

the subject is directly generated in [Spec, IP] (e.g.Aoun & Li, 1989), the two types of relative and pseu-do-cleft sentences will be predicted to elicit above-chance performance in Mandarin, as only one thematicR-dependency is present in the representation. For Pas-sive, if the Non-Movement approach is taken, and theVP-Internal Subject Hypothesis is assumed, comprehen-

Table 1Predictions of the TDH on the seven types of sentences in Mandarin

Sentence types Prediction

Active Above-chancePassive

Y. Su et al. / Brain and LanTable 1 summarizes the predictions of the TDH on theseven types of sentences in Mandarin.

For the DDH, since it relies on whether the sentenc-es involve one or two moved NPs to account for theselective comprehension decits, it has basically thesame predictions on the pattern of performance forEnglish and Mandarin. For sentences in which theobject NP is displaced, given that the subject movescomplement and the Theme NP does not move fromthe post-verb position11 (as shown in (23), i.e. theNon-Movement approach, e.g. Huang, 1999; Tang,2001; Ting, 1998; among others), there can be twopossible predictions. The rst possibility assumes thatsubject is directly generated as [Spec, IP] in MandarinChinese, as proposed in Aoun and Li (1989), andhence the TDH will predict above-chance performance

Movement approach ChanceNon-Movement approachSubject in Spec of IP Above-chanceVP-Internal subject Chancefrom [Spec, VP], comprehension will break down, forthere are two possibilities to relate the two NPs tothe two thematic roles. That is to say, assuming thatMandarin passives are derived in a similar way to Eng-

11 Note that the Theme NP is considered as not moved from the post-verb position only for long passives, i.e. with the Agent NP following bei.According to this analysis, the Theme NP in the embedded clause is a nulloperator undergoing A 0-movement and will be bound by the matrixsubject NP. For short passives, i.e. without the Agent NP, they are arguedto involve A-movement of PRO.sion may break down, for there are two antecedent-trace relationsone for the matrix subject NP andthe other for the embedded subject NP. However, ifthe subject is directly generated in [Spec, IP], theDDH will predict above-chance performance. As forBA construction, if ba is analyzed as a verb and theTheme NP base-generated in the post-ba position (i.e.the Non-Movement approach, e.g. Bender, 2000), thelish (i.e. the Movement approach, e.g. Li, 1990) andthat the VP-Internal Subject Hypothesis also appliesin Mandarin (e.g. Huang, 1993), the DDH will predictthat like English, Mandarin agrammatics will nd Pas-sive, Object Relative, and Object Pseudo-Cleft hard tocomprehend, whereas Active, Subject Relative, andSubject Pseudo-Cleft will be easier. However, if theVP-Internal Subject Hypothesis is not assumed, i.e. if

Sentence types Prediction

BA Construction Above-chanceSubject Relative ChanceObject Relative Above-chanceSubject Pseudo-Cleft ChanceObject Pseudo-Cleft Above-chance

guage 101 (2007) 118 7DDH predicts that above-chance performance will beelicited, since at most one NP (i.e. the Agent subject)is moved when the VP-Internal Subject Hypothesis isassumed, as shown in (24). On the other hand, if theTheme NP is analyzed as moving from the post-verbto the post-ba position (i.e. the Movement approach,e.g. Zou, 1993), the prediction will be chance-level per-formance if the VP-Internal Subject Hypothesis isassumed, but above-chance performance if the subjectis considered to be directly generated in [Spec, IP], asillustrated in (25).

nouns, nine inanimate nouns, and six pairs of adjectives.12

In addition, each type of sentence construction consisted of

lexical plus thematic role reversal distractor will sometimes be wrongly

8 Y. Su et al. / Brain and Language 101 (2007) 118three semantic conditionsreversible (in which both argu-ment nouns were animate, as shown in (21) and (22)), plau-sible (in which the agent was animate and the themeinanimate, as in (26a)), and implausible (in which the agentwas inanimate but the theme animate, as in (26b)). Thesemantic conditions were manipulated to examine the pos-sibilities that Brocas aphasics rely on semantic constraintsinstead of syntactic information for sentence comprehen-sion (as found in Caramazza & Zurif, 1976), or that thesepatients fail to assign thematic roles and adjectives onlyThe detailed predictions of the DDH on the perfor-mance of Mandarin-speaking agrammatic patients withrespect to the seven types of sentences are summarizedin Table 2.

In addition to the above predictions on the comprehen-sion of various sentence types, wewill also test the predictionof the RTDH, i.e. whetherMandarin-speaking aphasics willshow poor comprehension performance on the matrix pred-icateadjective relation in relative constructions if theVP-In-ternal Subject Hypothesis is assumed.

3. Methods

3.1. Materials

Seven types of sentences were employed for this study,as illustrated in (21) and (22). Each type of sentences wascomposed of the same set of 10 action verbs, 12 animatein complex structures (e.g. center-embedded relative sen-tences, as found in Sherman & Schweickert, 1989). Ten tri-al sentences were constructed for each condition, hencegenerating 30 trials for each of the seven types of sentences.

12 The ten action verbs were la to pull, da to beat, zhui to chase, tui topush, bao to hold in the arms, tuo to drag, bei to carry on the back,yao to bite, pai to pat, tian to lick. The six pairs of adjectives includedhei/bai black/white, pang/shou fat/thin, da/xiao big/small, nianqing/laoyoung/old, fang/yuan square/round, gao/ai tall/short. The nineinanimate nouns were chezi car, shadai sandbag, qiu ball, yizi chair,shu book, pixiang suitcase, beibao backpack, xie shoe, zhuozi table.The twelve animate nouns included nanhai boy, nuhai girl, nanren man,nuren woman, gou dog, mao cat, yisheng doctor, bingren patient,xiaotou thief, jingcha policeman, shizi lion, xiong bear.Appendix A lists all the 30 active sentences and the othersix types of sentences were derived from these active sen-tences by applying the relevant transformations.

Two versions of the comprehension test were developedbased on the same set of sentences, and the only dierencebetween the two versions was in the number of picturesaccompanying the target sentence.13 Version One consistedof 210 trial sentences (10 trials 3 conditions 7 sentencetypes), and the picture that depicted the meaning of the tar-get sentence was presented with three distractor picturesa lexical reversal distractor that changed the adjectiveassignment, a thematic role reversal distractor, and a lexi-cal plus thematic role reversal distractor. In Version Two,each target sentence appeared twice, once with a thematic

13 Version One was rst developed under the assumption that if theagrammatic patients decit is solely syntactic, the lexical reversaldistractor will not be their possible choice, and only the correct pictureand the thematic role reversal distractor will be relevant. Besides, thechosen only for relative construction. However, among the four Brocasaphasics tested using Version One, only one patient exhibited selectivecomprehension decits, we hence conjectured that a four-choice picturetask might be too dicult to process for some patients and thus developedVersion Two. One anonymous reviewer suggested that generally at chanceperformance be found only in Version One but not in Version Two. Wechecked previous studies using a sentencepicture matching task, andfound that Grodzinsky (1989), Beretta and Munn (1998) and Miera andCuetos (1998) used a three-choice task, but Caramazza and Zurif (1976),Sherman and Schweickert (1989), Hickok et al. (1993), Lukatela et al.(1995), and Beretta et al. (2001) adopted a two-choice task. Regarding thecriteria for chance level, we considered both 50% and 75% for VersionOne, as discussed in Section 4 on results. What is noteworthy is that LDC,who participated in both versions of tests, exhibited selective comprehen-sion decits on thematic role assignment but not on adjective assignment,regardless of the number of distractor pictures.

role reversal distractor, and once with a lexical reversal dis-tractor, thus generating 420 trial sentences. All the pictureswere simple, black and white line drawings. Figs. 1 and 2

illustrate the pictures used for a sentence in reversibleand plausible conditions, respectively. In both versions,the trial sentences were divided into 7 (for Version One)

Table 2Predictions of the DDH on the seven types of sentences in Mandarin

Sentence types Prediction Sentence types Prediction

Active Above-chancePassive SubjectrelativeMovement approach Chance Subject in Spec of IP Above-chanceNon-Movement approach VP-Internal Subject Above-chanceSubject in Spec of IP Above-chance Object RelativeVP-Internal Subject Chance Subject in Spec of IP Above-chance

BA Construction VP-Internal Subject ChanceMovement approach Subject Pseudo-CleftSubject in Spec of IP Above-chance Subject in Spec of IP Above-chanceVP-Internal Subject Chance VP-Internal Subject Above-chance

Non-Movement approach Object Pseudo-CleftSubject in Spec of IP Above-chance Subject in Spec of IP Above-chanceVP-Internal Subject Above-chance VP-Internal Subject Chance

Y. Su et al. / Brain and Language 101 (2007) 118 9Fig. 1. Pictures used for pang nanhai tui lao popo A fat boy is pushing an oldreversal distractor. (d) Lexical plus role reversal distractor.lady. (a) Target picture. (b) Thematic role reversal distractor. (c) Lexical

Language 101 (2007) 11810 Y. Su et al. / Brain andor 14 (for Version Two) blocks, with reversibility condi-tions and sentence types counter-balanced in each block.

3.2. Procedure

A sentencepicture matching task was administered. Atthe beginning of the test, each subject was familiarizedwith the task with twelve practice trials, including one tri-al for active and passive sentences, and two trials for theother ve types of sentences. Before each sentence waspresented, the picture array was exposed. The subjectwas asked to listen to each sentence as many times ashe/she needed and look carefully at all the pictures inthe array, and then point to the picture that best depictedthe meaning of the sentence. Feedback was given to thesubject for the practice trials but not for the experimentaltrials. Subjects were tested individually in a quiet room ineither one 1-h session or two half-hour sessions. All the

Fig. 2. Pictures used for lao popo tui fang yizi An old lady is pushing a squaredistractor. (d) Lexical plus role reversal distractor.responses of each subject were recorded on an answersheet by the investigator.

3.3. Subjects

Nine aphasic subjects participated in this study, includ-ing six non-uent type aphasics (four males and twofemales) and three uent Wernicke-type aphasics (allmales). Four of the Broca-type aphasics (LDC, ZZH,LGC, and HYY) were tested using Version One, and theother ve subjects were tested using Version Two exceptfor LDC, who participated in the study with both versionsof the test.14 All were outpatients of the Speech PathologyDivision of the Department of Physical Medicine and

stool. (a) Target. (b) Thematic role reversal distractor. (c) Lexical reversal

14 Since Version One was administered before Version Two wasdeveloped, the other three Broca-type aphasics were not available whenVersion Two was carried out except for LDC.

Rehabilitation at Taipei Veterans General Hospital in Tai-wan. All subjects were right-handed native speakers ofMandarin Chinese, and had at least 6 years of education.The age range was 3171 for the non-uent type aphasicsand 6063 for the uent Wernicke-type aphasics. The clas-sication of patients was based on a Mandarin version ofthe Boston Diagnostic Aphasia Examination (BDAE,Goodglass & Kaplan, 1972), and all the patients were atleast 3 months post-onset at the beginning of testing. Neu-rologically intact control subjects, matched for age andeducational background, were also tested. Since the controlsubjects performed awlessly on both Version One andVersion Two, they will not be discussed further. Detailedinformation about each patient is given in Table 3.

4. Results

4.1. Version One

For LDC, a post hoc Tukey test revealed that SubjectRelative elicited signicantly worse performance thaneither Active, Object Pseudo-Cleft (q(7,12) = 4, p < .01),or Passive, BA construction, Object Relative (q(7,12) =3.33, p < .05), and that Subject Pseudo-Cleft elicited worse

Subjects BDAE Sex Age at the time of Time pos

months 12 Head injury (subdural eusion in L frontalregion)

monthsnths 6 CVA (L frontal)nths 6 CVA (L basal ganglia & corona radiation)

6 L MCA (middle cerebral artery) territoryinfarction

1 month 16 Head injury (L intra-cerebral hemorrhage)month 18 L intra-cerebral hemorrhage

nth 12 Lacunar infarction (L corona radiation)nth 9 Lacunar infarction (L lentiform nucleus)nth 14 Cerebral infarction (L temporal-parietal)

erally associated with anomia. However, for CQM, his picture naming for therespectively), but his phrase repetition (scored 6) and sentence reading (scoredination, CQM scored 52, which was comparable to LDC (64), LCS (51), andspeech production problem was relatively milder than other Brocas patients.

Table 4Percentage of errors on each sentence type by four Broca-type aphasics(Version One)

Subject A P B SR OR SPC OPC

LDC 3 10 10 43a 10 57a,b 3ZZH 17 23 13 30 27 30 20LGC 47a,b 53a,b 43a 53a,b 43a 57a,b 47a,b

HYY 73b 70b 67a,b 67a,b 50a,b 57a,b 63a,b

a 50% as chance-level error rate (N = 30, df = 1). Since each sentencewas paired with four pictures in Version One, we consider both 50% errorrate (i.e. assuming the patients did not have problems with lexical adjectiveassignment) and 75% error rate (i.e. assuming all the four pictures wererandomly selected by the patients) as chance-level.b 75% as chance-level error rate (N = 30, df = 3).

Y. Su et al. / Brain and Language 101 (2007) 118 11classication testing onset

LDC Broca (apraxia) M 31 1 yr 4

1 yr 6ZZH Broca M 71 8 moLGC Broca F 55 7 moHYY Broca (apraxia) F 59 3 yr

CQM Broca (anomia)a M 40 14 yrLCS Trans. Motor M 31 1 yr 8XHY Wernicke M 60 3 moKHC Wernicke M 62 4 moXCJ Wernicke M 63 8 mo

a As one of the anonymous reviewers pointed out, Brocas aphasia is genMandarin version of BDAE scored lower than LDC (49% vs. 82% correct,9) were higher than LDC (scored 2 and 4, respectively). As for word discrimLGC (51). In general, CQMs naming problem was more profound, but hisFour of the Broca-type aphasics were tested using Ver-sion One. Table 4 presents the percentage of errors bythe four subjects on the seven types of sentences.

The four Brocas aphasics did not display a homoge-neous pattern of responses to the seven types of sentenc-es in Version One. LDC had more diculties on SubjectRelative and Subject Pseudo-Cleft. ZZH showed above-chance performance for all sentence types, whereasLGC and HYY evidenced problems for virtually alltypes of sentences. The eect of sentence type was exam-ined in a one-way analysis of variance with number oferrors in the three semantic conditions as the randomvariable for each subject. Only LDC showed a signicanteect on this variable (F(6,12) = 10.29, p = .0002), andthe error rates from the other three subjects did not dif-fer signicantly among the seven sentence types (p > .05).

Table 3Background Information on Aphasic SubjectsAlthough Brocas aphasia may sometimes be associated with anomia, the twoanomic patients in Miera and Cuetos (1998) do not demonstrate the kind of cperformance than either Active, Object Pseudo-Cleft(q(7,12) = 5.34, p < .01), or Passive, BA construction, ObjectRelative (q(7,12) = 4.67, p < .01), but the comparisonbetween Subject Relative and Subject Pseudo-Cleft wasnot signicant (p > .05).

Since only LDC demonstrated a selective pattern ofcomprehension, and his performance for active sentenceswas nearly perfect, we further examined the types of errorshe committed, i.e. whether he tended to select a role rever-sal distractor, a lexical reversal distractor, or a lexical plusrole reversal distractor. As shown in Table 5, 33 out of the41 total errors made by LDC were thematic role reversalerrors. This is particularly obvious for the two sentencetypes on which he had chance-level performance, i.e. Sub-ject Relative and Subject Pseudo-Cleft, in which the errorswere predominantly role reversal. This demonstrated thatthe comprehension diculty he had mainly involved the

t- Education(years)

Etiologytypes of symptoms do not necessarily go hand in hand. For example, theomprehension diculties found in their agrammatic patients.

sentences were more dicult than non-reversible plausible

Table 5Number of errors for each distractor type by LDC

Rolereversal

Lexicalreversal

Role and lexicalreversal

Active 0 0 1Passive 0 2 1BA Construction 3 0 0

Table 7Percentage of errors on each sentence type by each subject for VersionTwo

Subject A P B SR OR SPC OPC

12 Y. Su et al. / Brain and Language 101 (2007) 118sentences. As for the comparison of reversible and non-re-versible implausible sentences, LDC and ZZH showed moreerrors on the former, but LGC and HYY displayed moreproblems on the latter. To examine the dierences amongthe three semantic conditions, the data were submitted to aone-way analysis of variance with number of errors in eachsentence type as the random variable. No signicant dier-encewas found forLDCandHYY(p > .05), but a signicanteect was obtained for ZZH (F(2,12) = 12.83, p = .0003)and LGC (F(2,12) = 7.58, p = .004). A post hoc Tukey testshowed that for ZZH, plausible non-reversible sentenceswere signicantly easier than reversible (q(3,12) = 3.14,p < .01) and implausible non-reversible sentences(q(3,12) = 2, p < .05), but the dierence between reversibleand implausible non-reversible sentences was not signicant(p > .05). For LGC, only the dierence between plausibleand implausible non-reversible sentences reached signi-cance (q(3,12) = 3.14, p < .05).

4.2. Version Two

Three of the non-uent aphasics (two Broca-type andone transcortical motor) and three uent Wernicke-typeaphasics were tested using Version Two. Table 7 showsassignment of thematic roles rather than the predicate-ad-jective relation as predicted by the RTDH.

Table 6 presents the individual subjects error rates onthe three semantic conditions.

In general, reversible and non-reversible implausible

Subject-relative 10 2 1Object-relative 2 0 1Subject Pseudo-

Cleft17 0 0

Object Pseudo-Cleft

1 0 0the percentage of errors made by each patient for the sevensentence types.

For the three non-uent type aphasics, CQM demon-strated above-chance performance on all the sentencetypes, whereas LDC and LCS had more diculties on Sub-

Table 6Percentage of errors on each semantic condition by four Broca-typeaphasics (Version One)

Subject Reversible Plausible Implausible

LDC 26 13 20ZZH 37 6 26LGC 50 33 64HYY 63 57 71ject Relative and Subject Pseudo-Cleft than on the otherve types of sentences.15 As for the three Wernickes apha-sics, XHY and XCJ displayed chance-level performance onthe two types of relative sentences (XCJ also hadchance-level performance on Passive), but KHC showedchance-level performance on Active, Passive, and BA con-structions. The results were submitted to a one-way analy-sis of variance to examine the eect of sentence type. Asignicant eect was found for the three non-uent typeaphasics, i.e. LDC (F(6,12) = 15.085, p = .000022), LCS(F(6,12) = 7.94, p = .0007), and CQM (F(6, 12) = 4.16,p = .013), but not for the three uent type patients (i.e.XHY, XCJ, and KHC, p > .05). Post hoc Tukey analysesrevealed that for LDC, Subject Relative elicited signicant-ly worse performance than Active (q(7,12) = 7.67, p < .01),Passive (q(7,12) = 4.67, p < .05), BA construction(q(7,12) = 7, p < .01), Object Relative (q(7,12) = 6.34,p < .01), and Object Pseudo-Cleft (q(7,12) = 6.67,p < .01). LDCs Subject Pseudo-Cleft was also signicantlyworse than Active (q(7,12) = 7.33, p < .01), Passive(q(7,12) = 4.33, p < .05), BA construction (q(7,12) = 6.66,p < .01), Object Relative (q(7,12) = 6, p < .01), and ObjectPseudo-Cleft (q(7,12) = 6.33, p < .01). For LCS, SubjectRelative was signicantly worse than Active, Object Pseu-do-Cleft (q(7,12) = 6.33, p < .01), BA construction (q(7,12) = 5.33, p < .01), and Object Relative (q(7,12) = 5,p < .01). His Subject Pseudo-Cleft also elicited signicantlyworse performance than Active and Object Pseudo-Cleft(q(7,12) = 4.66, p < .05). For CQM, none of the sentencetypes diered signicantly from each other.

The results in Table 7 include both trials with thematic

LDC 0 15 3 38a 7 37 5LCS 3 18 8 35 10 27 3CQM 3 12 5 12 15 15 5XHY 27 37 35 42a 47a 22 33XCJ 32 42a 35 38a 42a 35 32KHC 40a 40a 40a 30 35 32 37

a 50% as chance-level performance (N = 60, df = 1).role distractor and trials with lexical distractor. In orderto investigate the predictions of the hypotheses in moredetail, we list the number of errors for the two types of dis-tractors in each sentence type in Table 8.

The three non-uent patients (i.e. LDC, LCS, andCQM) exhibited a distinctive pattern from the three uentpatients regarding distractor types; that is, they hardlymade any errors (at most 3) when the target picture was

15 Although LDC and LCS made more errors on Subject Relative andSubject Pseudo-Cleft than on the other ve types of sentences, only LDCserror rate on Subject Relative was in chance-level. The error rates ofLDCs Subject Pseudo-Cleft and LCSs Subject Relative were slightlyabove chance.

p < .01), and Object Pseudo-Cleft (q(7,12) = 4.66,

Lanpaired with a lexical distractor, whereas the uent aphasicsmade at least ve errors out of 30 trials with a lexical dis-tractor across all the sentence types. This amounts to say-ing that the errors committed by the non-uent aphasicswere mainly thematic role assignment, whereas the errorsmade by the uent aphasics could be either thematic roleassignment or adjective assignment. Like the results ofLDC in Version One, the prediction of the RTDH thatagrammatics would also have diculty on the predicateadjective assignment of the matrix clause in relative con-struction was not attested by our results from the non-u-ent patients in Version Two, either.

Since the hypotheses we are concerned with (i.e. theTDH and the DDH) mainly consider thematic role assign-ment, we take the error rates of the seven sentence typeswith thematic role distractor for further analysis. For thethree non-uent aphasics, CQM demonstrated abovechance performance for all the sentence types, LCS showedchance-level performance for Passive, Subject Relative, andSubject Pseudo-Cleft, and LDC had below chance perfor-mance on Subject Relative and Subject Pseudo-Cleft, butabove chance performance for the other ve types of sen-tences. For the three uent aphasics, XCJ and KHC dis-played chance-level performance on all the sentencetypes, and XHY had chance-level performance on Passive,

Table 8Number of errors for Thematic Role Distractor vs. Lexical Distractor oneach sentence type by each subject for Version Twoc

Subject A P B SR OR SPC OPC

LDC 0/0 6/3 2/0 23b/0 3/1 21b/1 3/0LCS 1/1 11a/0 5/0 20a/1 5/1 16a/0 2/0CQM 1/1 5/2 2/1 7/0 7/2 9/0 2/1XHY 7/9 13a/9 9/12a 16a/9 17a/11a 8/5 10a/10a

XCJ 12a/7 16a/9 12a/9 17a/6 16a/9 15a/6 10a/9KHC 15a/9 17a/7 14a/10a 12a/6 10a/11a 13a/6 17a/5

a 50% as chance level performance.b Below chance performance (N = 30, df = 1).c The number in front of the slash is the errors from thematic role trials,

and that after the slash from lexical trials.

Y. Su et al. / Brain andSubject Relative, Object Relative, and Object Pseudo-Cleft.The results were submitted to a one-way analysis of vari-ance to examine the eect of Sentence Type. A signicanteect was found for the three non-uent aphasics, i.e.LDC (F(6,12) = 24.56, p = .000001), LCS (F(6,12) =8.06, p = .00067), and CQM (F(6,12) = 4.47, p = .0099),but not for the three uent aphasics (p > .05). Post hocTukey analyses revealed that for LDC, Subject Relativeelicited signicantly more errors than Active (q(7,12) =7.67, p < .01), Passive (q(7,12) = 5.67, p < .01), BA con-struction (q(7,12) = 7, p < .01), Object Relative andObject Pseudo-Cleft (q(7,12) = 6.67, p < .01). His SubjectPseudo-Cleft also had signicantly more errors than Active(q(7,12) = 7, p < .01), Passive (q(7, 12) = 5, p < .01), BAconstruction (q(7,12) = 6.33, p < .01), Object Relativeand Object Pseudo-Cleft (q(7, 12) = 6, p < .01). For LCS,Subject Relative had signicantly more errors than Activep < .01). For CQM, Subject Pseudo-Cleft elicited signi-cantly more errors than Active (q(7,12) = 2.67, p < .05),BA construction and Object Pseudo-Cleft (q(7,12) = 2.33,p < .05).

Table 9 presents the individual subjects error rates onthe three semantic conditions.

Although the four non-uent aphasics in Version Oneall showed clearly lower error rates for plausible non-re-versible condition, the pattern was not found for eitherthe non-uent or the uent patients in Version Two exceptfor KHC. The eect of Semantic Conditions was examinedin a one-way analysis of variance, and a signicant eectwas obtained only for KHC (F(2,12) = 7.92, p = .0034).A post hoc Tukey test showed that KHCs implausiblenon-reversible condition elicited signicantly more errorsthan plausible non-reversible condition (q(3,12) = 3.86,p < .05), but the comparison between reversible and plausi-ble non-reversible conditions did not reach signicance.

5. Discussion

The results from our Mandarin-speaking aphasics dem-onstrated a distinction in comprehension performance(q(7,12) = 6.34, p < .01), BA construction and Object Rel-ative (q(7,12) = 5, p < .01), and Object Pseudo-Cleft(q(7,12) = 6, p < .01). His Subject Pseudo-Cleft also hadsignicantly more errors than Active (q(7,12) = 5,

Table 9Percentage of errors on each semantic condition by each subject forVersion Twoa

Subject Reversible Plausible Implausible

LDC 17 14 14LCS 14 19 11CQM 9 9 11XHY 38 35 31XCJ 36 36 37KHC 39 25 44

a Here both sentences with thematic role distractor and lexical distractorwere included for analysis. Sentences with thematic role distractor showedthe same pattern; i.e. only KHC had a signicant eect.

guage 101 (2007) 118 13between clinical aphasic syndromes; that is, none of the u-ent Wernickes aphasics showed selective pattern of com-prehension diculty, but three out of the six non-uentaphasics did. What the uent Wernickes aphasics in thisstudy displayed was an across-the-board problem with allof the sentence types tested. This diers from Lukatelaet al. (1995), who found that Serbo-Croatian-speaking Bro-cas and Wernickes aphasics as well as normal subjectsshowed the same pattern of errors, and they diered fromeach other only on the quantity of errors made. One possi-bility for this cross-linguistic discrepancy is that Mandarindoes not have rich case morphology to provide cues forthematic role assignment as Serbo-Croatian does, andhence has to rely more on word order for sentence interpre-tation. This conjecture is supported by Caramazza and

Zurifs (1976) study in English, another language withoutrich case morphology. The results reported in Caramazzaand Zurif (1976) showed that, unlike their normal subjectsand Brocas or Conduction aphasics, the English-speakingWernickes aphasics had chance-level performance even foractive sentences and semantically-constrained center-em-bedded sentences. In contrast, although not all of the Bro-cas patients showed selective decits as an eect ofsentence complexity, a subgroup of these patients (i.e.LDC, LCS, and CQM) did evince diculties on some typesbut not others. The nding that only a subgroup of agram-matic aphasics manifests selective comprehension decits isin line with the studies by Berndt, Mitchum, and Haendiges(1996) for English, Burchert, De Bleser, and Sonntag(2003) for German, and Caramazza, Capasso, Capitani,and Miceli (2005) for Italian, among others.

With respect to the manipulation of semantic condi-tions, it is noteworthy that unlike Sherman and Schweick-

(two Brocas, i.e. ZZH and LGC, and one Wernickes,i.e. KHC) exhibited signicant eects on semantic condi-tions; i.e. plausible non-reversible sentences were easierthan reversible and implausible non-reversible ones. Thisnding suggests that, at least for the aphasic patients in

14 Y. Su et al. / Brain and Language 101 (2007) 118ert (1989), who found both semantic and syntacticinformation contributed to sentence comprehension inthe agrammatic patients they tested, the three non-uentaphasics who showed sentence type eect in our studydid not display any signicant eect on semantic con-straints. Table 10 presents the detailed results from thesethree patients (LDC, LCS, and CQM).

As can be seen in Table 10, although for Subject Pseudo-Cleft, LDC made slightly less errors in plausible non-re-versible condition than the other two conditions in bothVersion One and Version Two, and for Passive, he mademore errors in reversible condition than the other two con-ditions in Version Two, the tendency was not consistent asit did not carry over to relative construction. Besides, nei-ther LCS nor CQM demonstrated any tincture of adoptingsemantically based heuristic procedures. However, amongthe six patients (three Brocas from Version One and threeWernickes from Version Two) who showed across-the-board diculties for all sentence types, three of them

Table 10Number of errors in each condition/sentence type from LDC, LCS, andCQM

Subject A P B SR OR SPC OPC

LDC (Version One)Reversible 1 2 1 5 2 7 0Plausible 0 1 0 4 0 3 1Implausible 0 0 2 4 1 7 0

LDC (Version Two)Reversible 0 6 0 8 1 9 0Plausible 0 2 2 7 1 5 2Implausible 0 1 0 8 2 8 1

LCSReversible 1 2 2 6 3 6 0Plausible 0 7 2 8 2 7 1Implausible 1 2 1 7 1 3 1

CQMReversible 2 1 1 2 4 2 1

Plausible 0 3 1 2 2 3 1Implausible 0 3 1 3 3 4 1the current study, heuristic procedures are employed onlywhen the syntactic processing breaks down, and the adop-tion of heuristics is not limited to agrammatic patients.Therefore, syntactic and semantic processing can indeedbe dissociable as postulated in Caramazza and Zurif(1976).

One of our major research interests for the current studyis in which of the linguistically based hypotheses (i.e. theTDH or the DDH) can better account for our Mandarinresults. Recall the predictions from the two hypotheses assummarized in Tables 1 and 2. The crucial distinctionbetween the TDH versus the DDH resides in the relativeconstruction and the pseudo-cleft sentences. According tothe TDH, whether the order of the NPs in a sentence con-forms to the canonical Agent-Theme order of the languageplays an important role in determining how agrammaticpatients interpret the sentence, and hence it predicts thatfor Mandarin, subject relative and subject pseudo-cleft sen-tences will pose more diculty than object relative orobject pseudo-cleft sentences. On the other hand, theDDH relies on the number of chains in the sentence asan index of comprehension diculty, and thus predict thatMandarin, like English, will nd that object relative andobject pseudo-cleft sentences elicit more errors than subjectrelative and subject pseudo-cleft constructions if the VP-In-ternal Subject Hypothesis is adopted. However, under theassumption that in Mandarin Chinese, subject is directlygenerated as [Spec, IP], the DDH will predict the two typesof relative and pseudo-cleft sentences to elicit above-chanceperformance. Our results from LDC and LCS, who dis-played chance or below chance performance on subject rel-ative and subject pseudo-cleft sentences but near normalperformance on object relative and object pseudo-cleft,show that the prediction of the TDH but not the DDH issupported. A similar pattern was also found from a Man-darin-speaking Brocas aphasic CZ as reported in Su andLaw (1993), and three Cantonese-speaking aphasicsLCC (transcortical motor aphasia) and WCK (conductionaphasia) in Law and Leung (1998), and CKC (anomicaphasia) in Law and Leung (2000).16 Among the casesreported in either Mandarin or Cantonese who exhibitedselective comprehension impairment, none of them showedthe pattern as predicted by the DDH, i.e. above chance per-formance on subject relative but chance or below chanceperformance on object relative sentences.

16 The error rates on subject relative sentences by these patients wereCZ (42.5%, chance), LCC (35%, chance), CKC (35%, chance), and WCK(70%, below chance), and the error rates for object relative sentences wereall above chanceCZ (17.5%), LCC (20%), CKC (0), and WCK (0). The

results of WCK were from a visual modality task and those of the otherthree patients were from an auditory task.

LanWith respect to the active, passive, and BA construc-tions, although LDC, LCS and CQM all made more errorson passive than on active or BA constructions, only LCSshowed chance-level performance on passive when themat-ic role reversal distractors were considered (as in Table 8).Similar variation was also found in Cantonese. The Can-tonese-speaking patients LCC, THM, and WCK in Lawand Leung (1998) showed above chance performance onactive sentences but chance-level performance on full pas-sive sentences in both visual and auditory modality tasks,whereas WPC displayed above chance performance in avisual modality task, but chance-level performance in anauditory modality task on both active and passive sentenc-es. Another two Cantonese-speaking patients CKC andLKH reported in Law and Leung (2000) showed abovechance accuracy on active, the corresponding BA sentenc-es, and passive sentences in visual comprehension, but inauditory comprehension CKC exhibited chance-level per-formance on full passive but above chance performanceon the other two types of sentences, whereas LKH demon-strated chance-level performance on all the three sentencetypes. Since Mandarin- and Cantonese-speaking patientsdid not all demonstrate chance-level performance on pas-sives, one may conjecture that this can be attributed tothe equivocal status of Mandarin and Cantonese passivesas being derived through movement or not. However, pre-vious studies on English (Berndt et al., 1996) and Italian(Caramazza et al., 2005) also revealed extensive variabilityin agrammatic aphasic patients comprehension of activeand passive sentences. Both studies found three distinctpatterns of performances: both active and passive compre-hended better than chance; both sentence types compre-hended no better than chance; active sentencescomprehended better than chance, but passive sentencescomprehended no better than chance. Since similar vari-ability exists cross-linguistically, a better explanationshould be that the by-phrase provides a potential heuristiccue for interpretation in the passive sentences (e.g. Foot-note 15 in Hickok et al., 1993), but such a cue is not avail-able for either relative construction or pseudo-cleftsentences. As for the pattern of results from the BA con-struction, it can be easily explained by the TDH, as thetwo NPs follow the canonical AgentTheme order, regard-less of whether the movement or non-movement analysis isadopted. To account for the overall above chance perfor-mance on the BA construction, the DDH will have toassume a non-movement analysis, or to stipulate that thesubject is directly generated in [Spec, IP]. The latter possi-bility will then lead to the prediction that the relative andthe pseudo-cleft sentences all elicit above-chance perfor-mance in Mandarin, which is not veried by our results.

In general, the patterns of selective comprehensiondecits found from the non-uent Mandarin-speakingagrammatic patients in the current study and the Canton-ese-speaking patients in Law and Leung (1998, 2000) can

Y. Su et al. / Brain andbetter be accounted for by the TDH than by the DDH.The results may suggest that the VP-Internal SubjectHypothesis does not apply in Mandarin or in Cantonese,and the Default Principle is adopted to help determiningthe assignment of thematic roles by agrammatic patients.However, when we consider the Non-Movement approachof Chinese BEI passive construction, i.e. if bei is analyzedas a verb taking an embedded clause as its complementand the Theme NP does not move from the post-verb posi-tion, the prediction made by the TDH would be above-chance performance under the assumption that the subjectis directly generated in [Spec, IP] (see Table 1). Neither theresults from the Mandarin-speaking patients in the currentstudy nor the patterns found in Cantonese as reported inLaw and Leung (1998, 2000) substantiate this prediction.Therefore, the VP-Internal Subject Hypothesis may stillapply in Mandarin as well as in Cantonese, and what iscrucial about agrammatic thematic role assignment residesin the Default Principle. Moreover, the relative ease of theobject-extracted relative and pseudo-cleft sentences versusthe diculty of the subject-extracted sentences for Manda-rin- and Cantonese-speaking agrammatic patients alsoshed light on the important role played by the canonicalAgent-Theme order in languages without rich inectionalor case morphology as Mandarin and Cantonese, regard-less of the number of chains in the sentence. The ndingsfrom the current study suggest that when syntactic repre-sentation cannot be consistently utilized for interpretation,the cognitive mechanisms underlying agrammatic sentenceprocessing employ the canonical AgentTheme order asstated in the Default Principle for sentence comprehension.Accordingly, the Default Principle should be indispensablein accounting for agrammatic patients selective compre-hension decits, especially when word order is the exclusivecue for interpretation.

Our second research issue concerns the prediction of theRTDH that agrammatic patients would also have dicultyon the predicate adjective assignment of the matrix clausein relative construction, it was not supported by the resultsfrom any of our non-uent patients as shown in Tables 5and 8, nor by the results from Cantonese in Law and Leung(1998, 2000). This nding suggests that since in English arelative clause intervenes between the subject NP and thematrix predicate, whereas in Mandarin the two are adja-cent to each other (as shown in (27)), a long-distance asso-ciation (Hickok et al., 1993, p. 387) between the subjectNP of a matrix clause and its VP-internal trace is the sourceof comprehension diculty found in English.

guage 101 (2007) 118 15

ther relevant studies may shed some light on the issues.egarding the relation between working memory limitationnd sentence comprehension, Miera and Cuetos (1998)

he two groups of patients both had a lowmemory span com-

ae

arin is SVO, but the two languages dier in that theead noun precedes the relative clause in English, butt follows the relative clause in Chinese. Although bothypotheses can explain the patterns found in Englishur results provide clear evidence for the important rolehe canonical AgentTheme order plays in agrammaticomprehension. In order to have a better understandingbout the nature of decits in agrammatic comprehen-ion, more cross-linguistic studies on languages withord order properties similar to Mandarin and Canton-

omments and suggestions on the earlier draft. We are alsorateful for Yosef Grodzinsky for valuable discussion

LanM

yntactic and semantic processing can indeed be dissocia-le. In addition, despite some problems regarding the

skinny girl hit fat manThe skinny girl hit the fat man.crinit21doRafstpin(nawcnetAicadttccdNtpuainlo

mwctt

sb

1,

,

dacgdhihotcasweattoc

A

acgCCTNt

A

I

1

2,

l

,

,

.ared with normal subjects. In addition, the Hebrew-speak-g agrammatic aphasics tested in Friedmann and Gvion2003) failed in the comprehension of object relatives butot on subject relatives, but the conduction aphasics,lthough showing severe working memory limitation, didell on all types of relative clauses irrespective of the ante-edent-gap distance. Since patients in the current study wereot under any kind of time constraints when they were test-d, it would be unlikely that the selective patterns found fromhe three patients resulted from limited processing capacity.s for whether agrammatic comprehension reects any def-its in chain formation, Zurif, Swinney, Prather, Solomon,ndBushell (1993) report that Brocas aphasics in their studyid not show the normal priming for the head noun phrase ofhe relative clause at the corresponding position in the rela-ive clause where a trace is linked to the head noun via ahain, whereasWernickes aphasics showed the normal pro-essing reection of chain formation. Another piece of evi-ence comes from a recent study by Friedmannovogrodsky, and Gvion (2005), in which they exploredhe nature of decits in Hebrew-speaking agrammaticatients and children with syntactic SLI who had dicultiesnderstanding object relative clauses. The subjects weresked to read aloud and paraphrase nounverb homographscorporated in object relative clauses as shown in the fol-wing example.

The results showed that the majority (87%) of errorsade by agrammatic patients were to read the homographhich should be a verb as a noun after the trace in relativelauses, but children with SLI read the homographs afterhe trace correctly, although both groups failed to interprethe object relative sentences.To sum up, the data obtained in this study fromandarin-speaking aphasics allow us to conclude thatound that Spanish-speaking agrammatic patients displayedelective comprehensiondecits but not anomic patients, butA nal point worth mentioning is whether the selectiveomprehension decits manifested in agrammatic patientseect a deciency in the linguistic structures as proposedthe hypotheses discussed here, or result fromageneral lim-ation in processing capacity (e.g. Crain, Ni, & Shankweiler,001; Haarmann & Kolk, 1991; Miyake, Carpenter, & Just994, 1995). Although in the current study we do not haveirect evidence to address this question, the ndings from

6 Y. Su et al. / Brain anhen-Sheng Luther Liu for providing references onhinese syntax, and Yi-Cheng Su for drawing the pictureshe completion of this work was partially supported by aational Science Council Grant (NSC 91-2411-H-003-052)o the rst author.

ppendix A. Trials for active sentences

. Reversible condition

. nianqing nyuren la shou nanhaiyoung woman pull skinny boyThe young woman pulled the skinny boy.

. shou nyuhai da pang nanrense (e.g. the relative construction in Japanese or Korean)re certainly needed to provide more empirical evidenceo test the validity of the predictions from these linguis-ically based hypotheses and to examine the interactionf the canonical AgentTheme order and morphologicaues on aphasic sentence processing.

cknowledgment

We thank all the subjects for participating in this studynd John Truscott as well as the anonymous reviewers foromprehension decits than the DDH for these two lan-uages. Like English, the canonical word order in Man-seeeby Law and Leung (1998, 2000) all showed the principls stated in the TDH can better account for the selectivfrom Mandarin as well as previous studies in CantonesDefBerult Principle as discussed in Mauner (1995) andtta and Munn (1998), among others, our resultguage 101 (2007) 118

lr

,q

,cernatagrta0tera

dtonog

leroro

heend

an

Lan10. da shizi tian xiao qiu

The skinny boy patted the white table.

skinny boy pat white table9. shou nanhai pai bai zhuozi

The big cat bit the black shoe.

big cat bite black shoe8. da mao yao hei xie

The fat man carried the small backpack.

fat man carry small backpack7. pang nanren bei xiao beibao

The skinny doctor dragged the small suitcase.

skinny doctor drag small suitcase6. shou yisheng tuo xiao pixiang

The young woman held the big book.

young woman hold big book5. nianqing nyuren bao da shu

The old lady pushed the square chair.

old lady push square chair4. lao popo tui fang yizi

The small dog chased the big ball.

small dog chase big ball3. xiao gou zhui da qiu

The fat man hit the black sandbag.

fat man hit black sandbag2. pang nanren da hei shadaifat boy pull white carThe fat boy pulled the white car.1. pang nanhai la bai cheziII. P ausible conditionThe small lion licked the big bear.

small lion lick big bear10. xiao shizi tian da xiong

The short thief patted the tall policeman.

short thief pat tall policeman9. ai xiaotou pai gao jingcha

The small cat bit the big dog.

small cat bite big dog8. xiao mao yao da gou

The fat man carried the skinny girl.

fat man carry skinny girl7. pang nanren bei shou nyuhai

The skinny doctor dragged the fat patient.

skinny doctor drag fat patient6. shou yisheng tuo pang bingren

The fat boy held the old lady.

fat boy hold old lady5. pang nanhai bao lao popo

The old lady pushed the fat boy.

old lady push fat boy4. lao popo tui pang nanhai

The big dog chased the small cat.

big dog chase small cat3. da gou zhui xiao maoY. Su et al. / Brain andbig lion lick small ballThe big lion licked the small ball. Brain and Language, 3, 572582.:

l

.

.

.

lagrammatic aphasic patient. Brain and Language, 27, 111134.Caramazza, A., & Zurif, E. B. (1976). Dissociation of algorithmic and

heuristic processes in language comprehension: evidence from aphasia.Capl , D., & Futter, C. (1986). Assignment of thematic roles by an

a Language, 87, 323342.

th dierence? Agrammatic sentence comprehension in German. BrainBurc rt, F., De Bleser, R., & Sonntag, K. (2003). Does morphology make

p cesses. Cambridge, MA: MIT Press.BradBy, D. C., Garrett, M. E., & Zurif, E. B. (1980). Syntactic decits incas aphasia. In D. Caplan (Ed.), Biological studies of mentaC nition, 58, 289308.

si of reversible sentences in agrammatism: a meta-analysisBern , R. S., Mitchum, C. C., & Haendiges, A. N. (1996). Comprehen-

B in and Language, 79, 407425.(2in01). The eects of scrambling on Spanish and Korean agrammaticrpretation: why linear models fail and structural models surviveBeret , A., Schmitt, C., Halliwell, J., Munn, A., Cuetos, F., & Kim, S

a ammatism. Brain and Language, 65, 404421.Beret , A., & Munn, A. (1998). Double-agents and trace-deletion in

a lysis. Journal of East Asian Linguistics, 9, 105145.aBendross-linguistic study. Brain and Language, 32, 1967., E. (2000). The syntax of Mandarin BA: reconsidering the verbaBates E., Friederici, A., & Wulfeck, B. (1987). Comprehension in aphasia

In uiry, 20, 141172.Aoun J., & Li, Y.-H. A. (1989). Scope and constituency. LinguisticRefe encesThe small ball licked the big lion.

small ball lick big lion10. xiao qiu tian da shizi

The black table patted the fat boy.

black table pat fat boy9. hei zhuozi pai pang nanhai

The white shoe bit the small cat.

white shoe bite small catIII. Implausible condition

1. hei chezi la shou nanhaiblack car pull skinny boyThe black car pulled the skinny boy.

2. bai shadai da shou nanrenwhite sandbag hit skinny manThe white sandbag hit the skinny man.

3. xiao qiu zhui da gousmall ball chase big dogThe small ball chased the big dog.

4. yuan yizi tui nianqing nyurenround chair push young womanThe round chair pushed the young woman.

5. xiao shu bao lao poposmall book hold old ladyThe small book held the old lady.

6. da pixiang tuo pang yishengbig suitcase drag fat doctorThe big suitcase dragged the fat doctor.

7. da beibao bei shou nanrenbig backpack carry skinny manThe big backpack carried the skinny man.

8. bai xie yao xiao mao

guage 101 (2007) 118 17

Caramazza, A., Capasso, R., Capitani, E., & Miceli, G. (2005). Patterns ofcomprehension performance in agrammatic Brocas aphasia: a test ofthe Trace Deletion Hypothesis. Brain and Language, 94, 4353.

Chomsky,N. (1981).Lectures onGovernment andBinding.Dordrecht:Foris.Crain, S., Ni, W., & Shankweiler, D. (2001). Grammatism. Brain and

Language, 77, 294304.

Linebarger, M. C. (1989). Neuropsychological evidence for linguisticmodularity. In G. N. Carlson & M. K. Tanenhaus (Eds.), Linguisticstructure in language processing (pp. 197238). Dordrecht: Kluwer.

Linebarger, M. C. (1990). Neuropsychology of sentence parsing. In A.Caramazza (Ed.), Cognitive neuropsychology and neurolinguistics

18 Y. Su et al. / Brain and Language 101 (2007) 118Friedmann, N. (2000). Agrammatic comprehension of OVS and OSVstructures in Hebrew. Behavioral and Brain Sciences, 23, 3334.

Friedmann, N., & Gvion, A. (2003). Sentence comprehension andworking memory limitation in aphasia: a dissociation betweensemantic-syntactic and phonological reactivation. Brain and Lan-guage, 86, 2339.

Friedmann, N., Novogrodsky, R., & Gvion, A. (2005). Syntacticmovement in agrammatism and S-SLI: two dierent impairments.Talk given in Generative Approaches to Language Acquisition(GALA), September 810, Siena, Italy.

Goodglass, H., & Kaplan, E. (1972). The assessment of aphasia and relateddisorders. Philadelphia, PA: Lea & Febiger.

Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press.Grodzinsky, Y. (1986). Language decits and the theory of syntax. Brain

and Language, 27, 135159.Grodzinsky, Y. (1989). Agrammatic comprehension of relative clauses.

Brain and Language, 37, 480499.Grodzinsky, Y. (1990). Theoretical perspectives on language decits.

Cambridge, MA: MIT Press.Grodzinsky, Y. (1995). A restrictive theory of trace deletion in agramma-

tism. Brain and Language, 50, 2751.Hagiwara, H. (1993). The breakdown of Japanese passives and theta-role

assignment principle by Brocas aphasics. Brain and Language, 45,318339.

Hagiwara, H., & Caplan, D. (1990). Syntactic comprehension in Japaneseaphasics: eects of category and thematic role order. Brain andLanguage, 38, 159170.

Haarmann, H. J., & Kolk, H. H. J. (1991). A computer model of thetemporal course of agrammatic sentence understanding: the eects ofvariation in severity and sentence complexity. Cognitive Science, 15,4987.

Heilman, K. M., & Scholes, R. J. (1976). The nature of comprehensionerrors in Brocas, Conduction, and Wernickes aphasias. Cortex, 12,258265.

Hickok, G. (1992). Agrammatic comprehension and the trace-deletionhypothesis. Occasional Paper No. 45, MIT Center for CognitiveScience. MIT, Cambridge, MA.

Hickok, G., Zurif, E., & Canseco-Gonzalez, E. (1993). Structural descrip-tion of agrammatic comprehension. Brain and Language, 45, 371395.

Huang, C.-T. J. (1993). Reconstruction and the structure of VP: sometheoretical consequences. Linguistic Inquiry, 24, 103138.

Huang, C.-T. J. (1999). Chinese passives in comparative perspective. TsingHua Journal of Chinese Studies, 29, 423509.

Kean, M. L. (1977). The linguistic interpretation of aphasic syndromes:agrammatism in Brocas aphasia, an example. Cognition, 5, 946.

Koopman, H., & Sportiche, D. (1991). The position of subjects. Lingua,85, 211258.

Law, S.-P., & Leung, M.-T. (1998). Sentence comprehension in CantoneseChinese aphasic patients. Aphasiology, 12, 4963.

Law, S.-P., & Leung, M.-T. (2000). Sentence processing decits in twoCantonese aphasic patients. Brain and Language, 72, 310342.

Li, Y.-H. A. (1990). Order and constituency in Mandarin Chinese.Dordrecht: Kluwer Academic.(pp. 55122). Hillsdale, NJ: Erlbaum.Linebarger, M. C., Schwartz, M. F., & Saran, E. M. (1983). Sensitivity to

grammatical structure in so-called agrammatic aphasics. Cognition, 13,361392.

Lukatela, K., Crain, S., & Shankweiler, D. (1988). Sensitivity toinectional morphology in agrammatism: investigation of a highlyinected language. Brain and Language, 33, 115.

Lukatela, K., Shankweiler, D., & Crain, S. (1995). Syntactic processing inagrammatic aphasia by speakers of a Slavic language. Brain andLanguage, 49, 5076.

Mauner, G. (1995). Examining the empirical and linguistic bases ofcurrent theories of agrammatism. Brain and Language, 50, 339368.

Mauner, G., Fromkin, V. A., & Cornell, T. L. (1993). Comprehension andacceptability judgments in agrammatism: disruptions in the syntax ofreferential dependency. Brain and Language, 45, 340370.

Miceli, G., Mazzucchi, A., Menn, L., & Goodglass, H. (1983). Contrastingcases of Italian agrammatic aphasia without comprehension disorder.Brain and Language, 19, 6597.

Miera, G., & Cuetos, F. (1998). Understanding disorders in agrammaticpatients: capacity or structural decits? Brain and Language, 64,328338.

Miyake, A., Carpenter, P. A., & Just, M. A. (1994). A capacity approachto syntactic comprehension disorders: making normal adults performlike aphasic patients. Cognitive Neuropsychology, 11, 671717.

Miyake, A., Carpenter, P. A., & Just, M. A. (1995). Reduced resourcesand specic impairments in normal and aphasic sentence comprehen-sion. Cognitive Neuropsychology, 12, 651679.

OGrady, W., & Lee, M. (2001). The isomorphic mapping hypothesis:evidence from Korean. Brain and Cognition, 46, 226230.

OGrady, W., & Lee, M. (2005). A mapping theory of agrammaticcomprehension decits. Brain and Language, 92, 91100.

Schwartz, M. F., Linebarger, M. C., Saran, E. M., & Pate, D. S. (1987).Syntactic transparency and sentence interpretation in aphasia. Lan-guage and Cognitive Processes, 2, 85113.

Schwartz, M. F., Saran, E. M., & Marin, O. S. M. (1980). The wordorder problem in agrammatism: I.Comprehension. Brain and Lan-guage, 10, 249262.

Shankweiler, D., Crain, S., Gorrell, P., & Tuller, B. (1989). Reception oflanguage in Brocas aphasia.Language and Cognitive Processes, 4, 133.

Sherman, J. C., & Schweickert, J. (1989). Syntactic and semanticcontributions to sentence comprehension in agrammatism. Brain andLanguage, 37, 419439.

Su, Y.-C., & Law, S.P. (1993) A study of syntactic comprehension of aChinese agrammatic patient. Paper presented at the InternationalConference on Chinese Linguistics 2, Paris.

Tang, S.-W. (2001). A complementation approach to Chinese passives andits consequences. Linguistics, 39, 257295.

Ting, J. (1998). Deriving the bei-construction in Mandarin Chinese.Journal of East Asian Linguistics, 4, 319354.

Zou, K. (1993). The syntax of the Chinese BA construction. Linguistics,31, 715736.

Zurif, E., Swinney, D., Prather, P., Solomon, J., & Bushell, C. (1993). Anon-line analysis of syntactic processing in Brocas and Wernickesaphasia. Brain and Language, 45, 448464.

Asyntactic thematic role assignment by Mandarin aphasics: A test of the Trace-Deletion Hypothesis and the Double Dependency HypothesisIntroductionPredictions of the hypotheses on mandarin agrammatismMethodsMaterialsProcedureSubjects

ResultsVersion OneVersion Two

DiscussionAcknowledgmentTrials for active sentencesReferences