location of lesions in stroke patients with deficits in syntactic - brain

Brain (1996), 119, 933-949

Location of lesions in stroke patients with deficitsin syntactic processing in sentence comprehensionDavid Caplan, Nancy Hildebrandt and Nikos Makris

Neuropsychology Laboratory, Massachusetts GeneralHospital, Boston, USA

Correspondence to: David Caplan, MD, NeuropsychologyLaboratory; Vincent Burnham 827, Fruit Street, Boston,MA 02114, USA

SummarySixty patients, 46 with left-hemisphere strokes and 14 withright-hemisphere strokes, and 21 normal control subjectswere tested for the ability to use syntactic structures todetermine the meaning of sentences. Patients enactedthematic roles (the agent, recipient and goal of an action) in12 examples of each of 25 sentence types, which weredesigned to test a wide variety of syntactic operations. Bothright- and left-hemisphere damaged patients performed worsethan control subjects on syntactically complex sentences,and left-hemisphere patients performed worse than right-hemisphere patients. Eighteen patients with left-hemispherestrokes underwent CT scanning to image the perisylvian

association cortex. There was no difference between theperformance of patients with anterior and posterior lesions,and no correlation between the degree of impairment andthe size of lesions in different regions of the perisylviancortex. These results are consistent with the view that syntacticprocessing involves an extensive neural system, whose mostimportant region is the left perisylvian cortex. When theseresults are combined with those of other studies, the picturethat emerges is one in which, within this cortical region, thissystem manifests features of both distributed and localizedprocessing.

Keywords: localization of language functions; syntactic comprehension deficits; localization of syntactic processing;syntactic comprehension in stroke patients

Abbreviations: CVA = cerebrovascular accident (stroke); rCBF = regional cerebral blood flow

IntroductionSentences are the level of the language code at which themeanings of individual words are related to each other toexpress information about events and states in the world(Jackson, 1874). This information indicates who is doing whatto whom (thematic roles), which adjectives are associated withwhich nouns (attribution of modification), what pronounsand other 'referentially dependent' items are related to(coreference), and other similar semantic information. Theability to express this propositional information contributesin an important way to the power that human language hasas a vehicle for thought and communication.

The propositional content of a sentence is determined bythe syntactic structure of that sentence (Chomsky, 1986).Individual words are assigned to different syntactic categories(e.g. noun, verb, preposition). These categories are organizedinto hierarchical structures (e.g. noun phrase, verb phrase) inwhich particular phrases stand in specific relationships toone another (e.g. subject of the verb, object of a preposition).

© Oxford University Press 1996

Propositional meaning is determined by these relationships.For instance, in the sentence 'The dog that scratched the catchased the bird', readers understand that 'the dog' is theagent of 'chased', despite a sequence of words—'the catchased the bird'—that in other circumstances could be takento express a proposition. The sentence is understood this waybecause of the position of the words 'the cat' and 'the dog'in the syntactic structure of the sentence: 'the dog' is thesubject of 'chased' and 'the cat' is the object of 'scratched'and has no syntactic relationship to 'chased' (Fig. 1).

Different aspects of syntactic structure determine differentaspects of meaning. In the sentence discussed above, thesyntactic relationships of subject and object determinethe thematic roles played by noun phrases. In other sentences,different syntactic relationships determine other aspects ofmeaning, such as what a pronoun (e.g. 'him') or a reflexive(e.g. 'himself') refers to, which items are modified by anadjective, etc. (Chomsky, 1986). For instance, in the sentence

Dow

nloaded from https://academ

ic.oup.com/brain/article/119/3/933/396450 by guest on 18 February 2022

934 D. Caplan et al.

NP COMP/ \

Det NII

The dog that 0 scratched the cat chased bird

Fig. 1 Diagram of the syntactic structure of the sentence Thedog that scratched the cat chased the bird', illustrating thehierarchical organization of categories that determines the factthat 'the dog' is the agent of 'chased'.

'Mary's picture of her intrigued Susan', 'her' cannot refer to'Mary', because of the syntactic relationship between 'her'and 'Mary', while in the sentence 'Mary's picture of herselfintrigued Susan', 'herself can only refer to 'Mary', becauseof this syntactic relationship. The syntactic relationshipbetween 'her' or 'herself and 'Mary' that determines whetherthey can be related is known as 'c-command' (Reinhart,1983) and is different from that which relates 'the dog' and'chased' in 'The dog that scratched the cat chased the bird'.

Most researchers believe that determining the meaning ofa sentence requires the assignment of a syntactic structure(parsing) and the use of that syntactic structure in conjunctionwith the meanings of the words in the sentence to determinethe meaning of the sentence (sentence interpretation). Parsingand sentence interpretation are thought to involve a numberof processes and operations that are specific to theconstruction of the particular syntactic relationships thatdetermine different aspects of meaning (Frazier, 1987a, b,1989, 1990; for an alternative view, see MacDonald, 1994).In addition, parsing and sentence interpretation are thoughtto require a processing resource system, whose size affectsthe efficiency and even the feasibility of assigning a syntacticstructure and understanding a sentence (Just and Carpenter,1992). As an illustration of this resource system, considerthe sentence 'The man that the woman that the child huggedkissed laughed'. Most readers cannot assign the thematicroles in this sentence, though they can do so relatively easilyin the two sentences that combine to form it—'The man thatthe woman kissed laughed', and 'The woman that thechild hugged kissed the man'. The trouble subjects haveunderstanding the sentence 'The man that the woman thatthe child hugged kissed laughed' is thought to arise becausethey do not have sufficient working memory capacity tomaintain the intermediate products of computation theygenerate in mind while processing the incoming words inthis complex structure.

Syntactic processing is an important candidate for a

distinctly human cognitive function, whose neural basis istherefore of considerable neurobiological significance(Pinker. 1994). However, the location of the syntacticprocessors that operate during sentence comprehensionremains unclear. It has been suggested that the ability toprocess syntactic structure in sentence comprehension iscarried out in a neural net based in the left-hemisphere,whose most active portion is Broca's area and adjacent partsof the frontal language zone(Mesulam, 1990; Damasio, 1992;Zurif et al., 1993). Evidence supporting this localizationcomes from the fact that a significant number of patientswith Broca's aphasia have difficulty understanding sentencesin which syntactic structure must be used to determinemeaning (Caramazza and Zurif, 1976; Schwartz et al., 1980;Caplan and Futter, 1986). One study of regional cerebralblood flow (rCBF) using I5O-PET has shown a localizedincrease in rCBF in part of Broca's area (the pars opercularis)when subjects made acceptability judgements for syntacticallymore complex compared with syntactically less complexsentences (Stromswold et al., 1996). Studies of event-relatedpotentials have also identified an early negative wave inthe left frontal region associated with aspects of syntacticprocessing (Neville et al., 1991; Kluender and Kutas, 1993).

However, this evidence does not clearly settle the issue ofhow the brain is organized for syntactic processing for severalreasons. First, other data imply that Broca's area is not theonly brain region in which syntactic processing occurs. Manypatients with lesions outside Broca's area have been describedwith syntactic comprehension disorders (Seines et al., 1983;Caplan et al., 1985; Caplan and Hildebrandt, 1988; Tramoet al., 1988). Secondly, there is evidence that Broca's areais not needed for syntactic processing. Many agrammaticpatients with Broca's aphasia demonstrate sensitivity togrammatical structure in grammaticality judgement and othertasks (Linebarger et al., 1983; Tyler, 1985), suggesting thatthey can assign the syntactic structure of a sentence even ifthey cannot use it to determine sentence meaning. Otheragrammatic patients with Broca's aphasia have shown intactsyntactic processing in sentence-picture matching andenactment tasks (Miceli et al., 1983; Caplan et al., 1985;Nespoulous et al., 1988; R. Berndt, C. Mitchum and AHaendiges, unpublished data). Thirdly, there are limitationsto the database supporting localization of syntactic processingin Broca's area. Most of the investigators who havedocumented syntactic processing impairments in agrammaticpatients have not reported specific aspects of lesions, and itis known that lesions in patients with Broca's aphasia oftenextend well beyond Broca's area (Mohr et al., 1978; Vanierand Caplan, 1990). The results of another, less well-controlledPET study implicated regions other than Broca's area insyntactic processing (Mazoyer et al., 1993). Results of event-related potential studies have also suggested that a moreposterior wave (the P600 or SPS) is related to aspects ofsyntactic processing (Neville et al., 1991; Hagoort et al.,1993). Some studies that report this wave have found it tobe maximal in amplitude over the right hemisphere (Osterhout

Dow



Lesions in syntactic comprehension deficits 935

and Holcomb, 1992, 1993), raising the question of whetherevent-related potentials have the necessary spatial resolutionto be definitive in determining the neural sites of languageprocessing. For these reasons, the functional neuroanatomyof syntactic processing and the role that Broca's area playsin this function remain unsettled areas.

In this study, we report on 60 stroke patients in whomsyntactic comprehension deficits were well-characterizedbehaviourally. In 18 patients, left-hemisphere lesions werevisualized by CT scanning. The results provide data relevantto the functional neuroanatomy of syntactic processing duringsentence comprehension.

MethodPatientsForty-six patients (29 English subjects and 17 French) withleft-hemisphere vascular lesions and 14 patients (nine Englishand five French) with right-hemisphere vascular lesionsparticipated in this study. Patients were recruited from hospitaland rehabilitation facilities in the Montreal area.

The 29 English left-hemisphere patients consisted of 15males and 14 females, aged 20-88 (mean 63) years. The 17French left-hemisphere patients included 10 males and sevenfemales aged 38-70 (mean 57) years. The level of educationfor both the English and French subjects ranged from gradeschool through college. Subjects were classified as beingright-handed or having anomalous dominance (Geschwindand Galaburda, 1985) based on questions drawn from theEdinburgh Handedness Scale that were answered by thepatient, the patient's spouse, or another close informant. Allexcept three English subjects and one French subject wereright-handed.

CT scans were obtained for nine English and nine Frenchpatients. The nine English patients for whom CT scans wereobtained included three male and six female patients, eightright-handed and one ambidextrous, with a mean age of 60years. The nine French patients for whom CT scans wereobtained included seven male and two female patients, all ofthem right-handed, with a mean age of 52 years.

The nine English right-hemisphere patients consisted offour males and five females, aged 48-86 (mean 65) years.The five French right-hemisphere patients included two malesand three females aged 27-79 (mean 54) years. Level ofeducation for the both English and French subjects rangedfrom grade school through college. All except one Englishand two French subjects were right-handed.

Normal subjectsTwenty-one normal subjects (11 English and 10 French) werealso tested on a subset of 22 sentences types. The 11 Englishcontrol subjects were aged 16-76 (mean 59) years. The 10French control subjects were aged 52-73 (mean 64) years.Handedness was not recorded in the control subjects, who

were included as a benchmark for the patients' performanceon the comprehension task.

All patients or their spouses and the normal subjects gavetheir informed consent to participate in the study, which hadthe approval of the local ethical committee.

MaterialsTwelve sentences of each of twenty-five sentence types werepresented (see Appendix). Subsets of these sentences havepreviously been used to test comprehension in stroke patients(Caplan et al., 1985; Caplan and Hildebrandt, 1988), patientswith closed head injury (Butler-Hinz et al., 1990), andpatients with dementia of the Alzheimer's type (Rochonet al., 1994). Six sentence types contained only 'full' nounphrases, which are noun phrases like 'the dog' or 'the cat'that refer directly to items in the real world, and six sentencetypes contained pronouns or reflexives ('himself or 'him'),which have to be related to another noun phrase in order tomake reference to an item in the world. The remaining 13sentence types contained what are known as 'empty nounphrases' (Chomsky, 1986)—items such as the understoodsubject of 'to jump' in the sentence 'John promised Bill tojump' or the object of 'scratched' in the sentence 'The dogthat the cat scratched chased the mouse'. To require thatsubjects structure these sentences syntactically and not simplyrely on real-world knowledge to determine the correctmeaning of these sentences, all sentences were constructedsuch that any noun could have accomplished or been therecipient of the action of any verb and could have beenreferred to by any pronoun, reflexive or empty noun phrasein the sentence. Thus, the sentences were structured so as toassess a subject's ability to process a wide range of syntacticstructures in sentence comprehension.

ProcedureSentences were divided into three batteries. The first batterycontained active, passive, dative and relative-clause sentencetypes. The second battery contained sentence types with oneproposition and a reflexive or a pronoun, and matchedsentences with full noun phrases. The third battery containedsentence types with two propositions and either full, reflexive,pronoun or empty noun phrases. Nouns in the first batterywere animal names, and nouns in the second and thirdbatteries were either a definite concrete noun phrase ('theold man', 'the boy') or a relational noun phrase ('the father','his friend'). The three batteries were given in the samesequential order, with the first battery first and the third last,with a training period before each.

Subjects were tested individually in testing rooms at thehospitals or rehabilitation facilities or in their homes. At theonset of each session, the experimenter indicated the namesof the objects (animals or dolls) to each patient and thentested the patient's ability to identify these objects one at a

Dow




AC-PC

Fig. 2 Lateral view of the left-cerebral hemisphere in the human showing the cortical regions ofinterest considered in this study, and cerebral sulci. The shadowed area with the asterisk in it is notincluded in region of interest Tl. F3t = inferior frontal gyrus/pars triangularis; F3o = inferior frontalgyrus/pars opercularis; SG = supramarginal gyrus; AG = angular gyrus; Tl = superior temporal gyrus;ce = central sulcus; prc = precentral sulcus; sf = superior frontal sulcus; if = inferior frontal sulcus;aar = anterior ascending ramus of the sylvian fissure; ahr = anterior horizontal ramus of the sylvianfissure; phr = posterior horizontal ramus of the sylvian fissure; par = posterior ascending ramus of thesylvian fissure; st = superior temporal sulcus; it = inferior temporal sulcus; poc = postcentral sulcus;ip = intraparietal sulcus; im = intermediate sulcus of Jensen; ag = angular sulcus; ao = anterioroccipital sulcus; lo = lateral occipital sulcus. (Modified from Rademacher et al., 1992.)

time and in series. Patients who could not reliably point toall objects in the set were excluded from further testing.

Subjects were told that the purpose of the experiment wasto test their abilities to understand 'who did what to whom'in the sentences. Subjects were instructed to indicate 'whodid what to whom' by acting out the sentence using theitems provided. The experimenter emphasized that subjectsdid not need to show details of the action of the verb, buthad to clearly demonstrate which item was accomplishingthe action and which receiving it. Practice sessions weregiven for each battery, during which some easy and somedifficult sentence types were presented. During these practicesessions, the experimenter did not correct errors that a patientmade, but did ask for repetitions and revisions of responsesin which it was not clear which item initiated and whichitem received an action. Practice continued until the patient'sactions could be clearly interpreted. The experimenter thenread each experimental sentence with a normal, neutralintonational contour and recorded the subject's response (fordetails of the task, see Caplan et al., 1985; Caplan andHildebrandt, 1988).

NeuroimagingEighteen patients with left-hemisphere strokes underwent CTscanning. Scans were performed from 7 days to 7 years afterthe onset of the stroke (one on day seven, two on day eight,and the remainder from 3 months to 7 years after the onsetof the stroke). In 17 subjects, a special protocol was usedto obtain CT images. Scans were supervised by aneuroradiologist. The subject was carefully positioned so thatthe imaging plane was parallel to the canthomeatal (CM)line, which runs almost parallel to the bicomissural (AC-PC) line (Tokunaga et al., 1977; Fox et al., 1986). A Scoutfilm was obtained with a radio-opaque marker on the skullperpendicular to the canthomeatal line at the point of the

external auditory meatus, which corresponds to the positionof the posterior commissure. This marker was visible as awhite dot on the left side of the head in all CT images, andserved to help verify the angle of the scan (Vanier et al.,1985). The brain in its entire width was imaged in series ofsingle slices of 5 mm width in 14 patients and of 10 mmwidth in four patients.

These 17 scans were mapped onto the Talairach andTournoux (1988) atlas, whose templates are parallel to thebicomissural line. The anatomical regions defined in theTalairach and Tournoux atlas correspond roughly tocytoarchitectonic fields (Sanides, 1964; Rademacher et al.,1993), and the atlas has been the basis for localization ofchanges in rCBF and regional cerebral blood volume inactivation studies (Fox et al., 1985; Belliveau et al., 1991;Fox and Lancaster, 1993). The remaining scan, which wasobtained at a different angle, was matched to the templatescorresponding to its angle of imaging in the Damasio andDamasio (1989) atlas. The templates of the Damasio andDamasio atlas were normalized to the Talairach dimensionsso that volumes of lesions and regions of interest would becomparable across the 18 scans.

Film images were traced on transparencies and magnifiedto match templates in the appropriate atlas. One magnificationfactor for each brain was used, which was the ratio of themaximum longitudinal axis of a selected CT scan slice tothe longitudinal axis of its corresponding template. On aslice per slice basis, guided by key landmarks (mainly thesylvian fissure and the hemispheric margins), the surface ofthe lesion was matched to the surface of the atlas template.This matching required small amounts of spatial stretching,rotation and translation. Volumetric analysis of each lesionwas performed by measuring the surface occupied by thelesion in each normalized CT scan slice, multiplying it bythe thickness of the corresponding template, and summing

Dow




Table 1 Mean percentage correct for each sentence type and each subject group

LCVA (n = 46) RCVA (n = 14) CTRL (n = 23)

Baseline sentences without referentially dependent noun phrasesTwo-place activeThree-place activeConjoinedActive conjoined themeThree reflexive expressionsSimple active reflexive expressions

Sentences with overt referentially dependent noun phrasesReflexives, simple noun phrasePronouns, simple noun phrasesSimple active reflexSimple active reflex (friend of X subj)Simple active pronoun (friend of X subj)Simple active pronoun

Sentences with empty referentially dependent noun phrasesTwo-place passiveTruncated passiveTwo-place cleft objectThree-place passiveThree-place cleft objectSubject-object relativeObject-subject relativeObject-object relativeSubject-subject relativePassive conjoined agentObject controlSubject controlNoun phrase-raising/pass object control

See Appendix for descriptions of sentence types. LCVA = left CVA patient; RCVA = right CVA patient; CTRL = control subject;* = not tested on this sentence type.

927769816862

746388636657

70677046543749426065694952

958391978887

888696889186

89908873675570698391776861

10010099*9778

999878777878

100*99969688959794*989396

these values across the templates in which the lesionappeared.

Five anatomical regions of interest that correspond to brainstructures within the left-hemisphere language zone weredefined on the Talairach and Tournoux atlas. The regionswere defined following the criteria described by Rademacheret al. (1992), which rely primarily on the morphology of thecerebral sulci. Lesion size in each region of interest wascalculated as described above, and expressed as a percentageof the volume of each region of interest. The five regions ofinterest in which we performed volumetric analyses areshown in Fig. 2, and were defined as follows.

Region of interest 1. This is a region in the superiortemporal gyms, which corresponds to Brodmann's cyto-architectonic area 22. It is defined posteriorly by a coronalplane that passes through the dorsal end of the posteriorascending ramus of the sylvian fissure, ventrally by thesuperior temporal sulcus, anteriorly by a coronal plane passingthrough the posterior end of the temporal pole, and dorsallyby the posterior horizontal ramus of the sylvian fissure. Thecortex corresponding to Heschl's gyrus and the central portionof the planum temporale (areas 41 and 42) are excluded fromthis region of interest.

Region of interest 2. This region corresponds to pars

triangularis and the portion of frontal operculum that underliesthe pars triangularis. It is part of the inferior frontal gyrusand represents Brodmann's cytoarchitectonic area 45. It isdefined posteriorly by the anterior ascending ramus of thesylvian fissure, ventrally by the anterior horizontal ramus ofthe sylvian fissure, anteriorly by the coronal plane that passesthrough the rostral end of the anterior horizontal ramus, anddorsally by the inferior frontal sulcus.

Region of interest 3. This includes the pars opercularisand the part of frontal operculum that underlies the parsopercularis. It is is also part of the inferior frontal gyrus. Itcorresponds to Brodmann's cytoarchitectonic area 44. It isdefined posteriorly by the precentral sulcus, ventrally by theposterior horizontal ramus of the sylvian fissure, anteriorlyby the anterior ascending ramus of the sylvian fissure, anddorsally by the inferior frontal sulcus.

Region of interest 4. This refers to the angular gyrusand corresponds to Brodmann's cytoarchitectonic area 39. Acoronal plane that passes through the caudal end of theanterior occipital sulcus is its posterior border, and theanterior occipital sulcus together with the superior temporalsulcus form its ventral border. A coronal plane that passesthrough the inferior tip of the intermediate sulcus of Jensen

Dow




constitutes its anterior limit, and the intraparietal sulcusdefines the dorsal border of the angular gyrus.

Region of interest 5. This corresponds to the supra-marginal gyrus and the parietal operculum, and representsBrodmann's cytoarchitectonic area 40. Its caudal border is acoronal plane passing through the inferior end of theintermediate sulcus of Jensen, its ventral limit is defined bythe posterior horizontal and posterior ascending rami of thesylvian fissure, and the superior temporal sulcus. Anteriorly,the postcentral sulcus is its border, and dorsally it is delimitedby the intraparietal sulcus.

Because of concerns related to the accuracy of theparcellation of the Talairach and Tournoux atlas accordingto the Rademacher et al. (1992) criteria, a grosser parcellationsystem was also used. This parcellation grouped togetherregions anterior to the pre-central sulcus (regions of interest2 and 3) into a single 'anterior' region of interest. This regioncorresponded to the traditional Broca's area (Brodmann'sareas 44 and 45). A second region of interest was formed bycombining regions of interest 1, 4 and 5 into a single'posterior' region that included perisylvian association cortexposterior to the post-central gyrus. Finally, all five regionsof interest were combined into a single region of interest thatreflected the entire perisylvian association cortex.

ResultsPerformance on sentence comprehension taskPerformance of patients and control subjects on the sentencecomprehension task is shown in Table I.

The mean overall accuracy on the 25 sentences of the 21normal subjects was compared with that of the group of 46patients with cerebrovascular accident (CVA) to the lefthemisphere (L) and the group of 14 patients with CVA tothe right hemishere (R) in a 3X2 between-subjects ANOVAwith subject type [normal controls, (L)CVA, (R)CVA] andlanguage (English, French) as orthogonal factors. There wasa main effect of group \F{2.11) = 20.56, P < 0.001]. Therewas no main effect of language. There was. however, aninteraction between group and language type 1/(2,77) =3.48, P = 0.036]. Simple effects showed that patients with

both left and right CVAs were less accurate overall thannormal controls. For the French patients, the (R)CVA group(overall accuracy = 83%) was significantly better than the(L)CVA group (overall accuracy = 48%). For Englishpatients, (R)CVA patients (overall accuracy = 82%) did notdiffer from the (L)CVA group (overall accuracy = 73%).The English (L)CVA patients performed significantly betterthan the French (L)CVA patients. The difference betweenthe (R)CVA patients and the English (L)CVA patients wasnot statistically significant. This pattern of results indicatesthat damage to both hemispheres affects sentence comprehen-sion, with greater effects following left-hemisphere damagethan right-hemisphere damage.

A point to note is that the English patients with (L)CVAsperformed better than their French counterparts, and thedifference between them and the English (R)CVA patientsdid not reach statistical significance. However, in other studiesusing the same task, English and French aphasic patientshave performed at the same levels (Caplan et al., 1985), andthe English (L)CVA patients' level of performance was lowerthan that of the (R)CVA patients in the present study. Furtheranalyses {see below) indicate that the English (L)CVA patientsshowed the same impairment in syntactic processing as theFrench (L)CVA patients. The somewhat better than expectedoverall performance of the English (L)CVA patients whowere tested in this study is therefore probably an atypicalfeature of the (L)CVA patients in this sample. We cautionagainst concluding from this feature of this group's perform-ance that (L)CVA patients in general perform at the samelevel as (R)CVA patients on this task.

There are many factors that enter into the performanceof this task, and that could have contributed to loweredperformance in the patient groups. To investigate whetherspecifically syntactic aspects of sentence processing wereaffected by lesions in either hemisphere, we comparedperformance on sentences that were more syntacticallycomplex with that on matched sentence types that weresyntactically less complex.

Syntactic complexity was determined in the followingmanner. In most English sentences, the subject noun phraseprecedes the verb, and the verb is followed by an object andthen by one or more prepositional phrases. The noun phrasesin subject and object position are usually assigned thethematic roles of agent (the perpetrator of an action) andtheme (the person or item upon whom the action is enacted),respectively. The noun phrases in the prepositional phrasesplay other thematic roles, such as goal, beneficiary, etc.,depending upon the preposition that is present. This order ofthematic roles—agent-theme-goal (or other)—is known asthe canonical thematic role order for English. It has beenshown that sentences with empty noun phrases are moredifficult than sentences with either full noun phrases orpronouns or reflexives when the order of thematic roles inthe sentence deviates from the canonical agent-theme-goalorder (Caplan et al., 1985; Schwartz et al., 1987; Caplanand Hildebrandt, 1988). This complexity is attributable toprocessing the syntactic structure of these sentences, asopposed to their length or other factors. Thus, to testspecifically syntactic aspects of sentence processing, sevensentence types with empty noun phrases in which thematicroles occurred in non-canonical order were matched fornumber of nouns, thematic roles and propositions with foursentences in which thematic roles appeared in a canonicalorder (see Appendix). For example, the sentence 'The monkeythat the cow hit pushed the goat' was compared with thesentence 'The cow hit the monkey and pushed the goat'.Comparison of performance on these sentences with non-canonical thematic role order to performance on these

Dow




1.0-1

0.9-

CD

6CJ

CCD

CDQ.

0.8-

0.7-

0.5-

0.4

Controls(R)CVA(L)CVA

Simple ComplexSentence type

Fig. 3 Performance of control subjects and patients with (L)CVAand (R)CVA on matched sets of syntactically simple andsyntactically complex sentences.

sentences with canonical thematic role order tests the integrityof syntactic processing.

The results are shown in Fig. 3. The data were analysedin a 3X2X2 ANOVA with subject type [normal controls,(L)CVA, (R)CVA] and language (English, French) asbetween-subject factors and syntactic type (complex, simple)as a within-subject factor. There were main effects of group[F(2,77) = 29.9, P < 0.001] and syntactic type [F(l,77) =66.9, P< 0.001]. There was a groupXlanguage typeinteraction [F(2,77) = 3.3, P = 0.04] and a groupXsyntactictype interaction [F(2,77) = 18.7, P < 0.001]. No other effectswere significant. The effect of group and the interaction ofgroup and language showed the same patterns as the resultsreported above for performance on all sentence types.

The effect of syntactic type and the interaction of groupand syntactic type are relevant to the question of whetherpatients showed deficits in syntactic processing. Simpleeffects showed that performance of all groups was better onthe simple sentences than on the complex sentences. Forsimple sentences, control subjects performed better than bothpatient groups, and the two patient groups did not differ intheir performance. For complex sentences, control subjectsperformed better than both patient groups, and patients withright-hemisphere lesions performed better than those withleft-hemisphere lesions. These results indicate that both left-and right-hemisphere damaged patients have more difficultycomprehending sentences that require more complex syntacticoperations to be understood than sentences that do not. Thisdeficit is greater for patients with left-hemisphere lesionsthan for patients with right-hemisphere lesions, but it ariseswith damage to either hemisphere. The English (L)CVApatients showed the same disturbances of syntactic processingas were seen in the French (L)CVA patients (the three-wayinteraction of languageXsentence typeXhemisphere was notsignificant).

</>CDCOCO

oCD.a

6 -

2 -

-0.2 0.0 0.2 0.4

Syntactic complexity score0.6

Fig. 4 Graph of the number of subjects with left and right CVAsshowing different magnitudes of syntactic complexity effects.

Effects of right-hemisphere lesionsThe presence of a syntactic complexity effect is expectedfollowing damage to the perisylvian region of the lefthemisphere. However, it is somewhat surprising that aneffect of syntactic complexity would arise following right-hemisphere stroke. Its presence would be readily explained,however, if it were due to the performance of one ortwo patients, who might be right-hemisphere dominant forlanguage. To determine whether this was the case, wecalculated a syntactic complexity score for each patient,consisting of the average of his or her performance(expressed as percent correct) on the seven syntacticallycomplex sentences subtracted from his or her performanceon the matched syntactically simple baseline sentences. Forthe (R)CVA patients, these scores ranged from close to zero(-0.044), indicating the absence of a syntactic complexityeffect, to 0.514, indicating a considerable effect. Thirteen ofthe 14 scores of the (R)CVA patients were positive, indicatingthat 13 of the 14 patients contributed to the syntacticcomplexity effect. As shown in Fig. 4. only one right-hemisphere patient had a complexity score that wassubstantially above the mean score of the left-hemispheregroup and that indicated a major syntactic complexity effect.With the exception of this one patient, the scores of the right-hemisphere patients approximate a normal distribution. Thispattern suggests that the syntactic complexity effect in theright-hemisphere patient group is not due to the performanceof one or two patients, but is due to small complexity effectsappearing in most patients.

Effects of left-hemisphere lesionsWe analysed the performance of patients with left-hemispherelesions to gain clues as to the determinants of their sentence

Dow



940 D. Caplan et al

Table 2 Factor loadingseach factor

Sentencetype

ST19ST4ST6ST14ST22ST5ST2ST15ST8ST11ST7ST10ST9ST13ST3ST12ST24ST23ST18ST21ST16ST20ST25ST17ST1

Variance (%)

Factor 1

0.859550.859160.858970.855160.853810.850070.842460.84040.837430.831550.831280.824290.818780.797440.797370.795390.794850.784220.756630.747450.723150.700030.69730.670320.54815

63.1

and variance accounted for by

Factor 2

0.31956-0.28987-0.36908-0.27535

0.010360.06886

-0.32420.23936

-0.03096-0.27093-0.16539-0.07572-0.33878

0.23839-0.09189-0.01246

0.195940.27470.07758

-0.192580.351820.52692

-0.051510.119180.36464

6.2

Factor 3

0.145890.019890.052290.207010.16013

-0.181020.048050.0569

-0.28533-0.2347

0.02504-0.15093-0.28031-0.37192

0.12815-0.42088

0.18158-0.1178

0.117460.24770.17586

-0.038020.312780.50633

-0.2248

5.0

Factor 4

-0.06984-0.10883

0.15141-0.04501-0.03516

0.0709-0.24518-0.31975-0.21175

0.130090.231090.30870.07827

-0.03572-0.04886-0.04885

0.147210.35834

-0.44259-0.26493-0.18863

0.116730.26760.28947

-0.03516

4.2

comprehension impairments. A factor analysis with varimaxrotation was carried out on the 25 sentence types in order toexplore the nature of the relationship between performanceon these sentence types. Four factors were extracted. Loadingsof sentence types on factors and proportionate variancecontribution are shown in Table 2, which shows that the firstfactor accounted for almost two-thirds of the variance. Witha cut of 0.5 for inclusion of a variable in interpretation of afactor, all variables loaded on the first factor, with only onesentence type (simple active reflexive) loading on the secondfactor, and one sentence type (subject control) loading onthe third. This analysis, which replicates the results of Caplanet al. (1985) and extends those results to a much larger setof sentences, shows that patient performance is largelyaffected by a single factor. This factor has been thought ofas the availability of the processing resource discussed in theintroduction to this paper that constrains overall sentenceprocessing ability (Caplan et al., 1985). This analysis thussuggests that a useful approach to the study of the localizationof syntactic processing would be to see whether overallperformance and the syntactic complexity score describedabove, which would be principally determined by theavailability of this resource, differ as a function of lesionsite and/or correlate with lesion size in a particular site.

As indicated above, CT scans were obtained on 18 (L)CVApatients. To determine whether these patients were similar toother aphasic patients with (L)CVAs, the performance of

these 18 (L)CVA patients was compared with that of theremaining (L)CVA patients in a 2X2 [group (patients withand without scans) Xsentence type (syntactically complexversus baseline)] ANOVA. There were main effects of group[F( 1,44) = 5.9, /><0.02] and sentence type [F(l,44) =101.5, P < 0.001), and a significant interaction between thetwo factors [F(l,44) = 6.4, P < 0.02]. Analysis of simpleeffects showed that subjects with and without scans performedat the same level on the baseline sentences, and that subjectswith scans performed worse than those without scans on thecomplex sentences. The syntactic complexity index describedabove was also computed for each patient, and the magnitudeof this index compared in the patients with and without scansusing Student's t test. This analysis confirmed the results ofthe ANOVA in showing that the syntactic complexity indexwas greater in the patients with scans than the patientswithout (t = 2.52, P < 0.02). These results indicate that thesample of patients who were scanned had more difficultywith syntactic processing than the remaining patients. Giventhat the English (L)CVA population tested here performedsomewhat better than other (L)CVA groups (see above), thefact that the 18 patients whose scans were examined wereamong the more affected patients provides reassurance thatthey are typical of aphasic (L)CVA patients. In addition, thefact that they had clear problems in syntactic processingmakes the patients who were scanned good subjects fora study of the lesion sites associated with disorders ofthis function.

The cortical extent of the lesion sites in the 18 (L)CVApatients who had undergone CT scans is shown in Fig. 5.Table 3 lists the percent of each region of interest occupiedby the lesion in each patient, along with each patient'sperformance on the sentence comprehension task. Toinvestigate the relationship between the site and size of theselesions and performance on the sentence comprehension task,t tests, correlational analyses and regression analyses wereperformed.

First, we tested the hypothesis that patients with damageto different regions of the perisylvian association cortexdiffer in syntactic processing in sentence comprehension byusing t tests. It was impossible to compare patients withpurely anterior lesions with those with purely posteriorlesions, because only one patient had a purely anterior lesion.We therefore compared the performance of six patients whosevisible lesions were confined to the 'posterior' region ofinterest with that of the remaining 12 patients, whose lesionsinvolved the 'anterior' region of interest (Broca's area).Independent / tests showed that neither overall accuracy onthe 25 sentence types nor the syntactic complexity scorediffered between the two groups.

Although the factor analysis suggested that a single factoraccounted for most of the variance in patients' performances,previous research has shown that individual patients can haveselective impairments affecting particular syntactic operations(Caplan and Hildebrandt, 1988). Moreover, as noted inthe Introduction, psycholinguistic models postulate different

Dow



NC EG


AD ^~—T~^ SD_—-T-^ CD,

Fig. 5 Diagrams of the left lateral hemisphere of the human brain depicting the approximate extent of the lesion in each of the 18subjects with (L)CVAs whose CT images were analysed. Using a stylized hemisphere as a template, each case was reconstructed fromaxial views of the CT scans. Shaded areas include both cortical and white matter lesions. Anglophone patients are presented in the leftpanel and francophone patients in the right.

Table 3 Individual lesion volumes as a percentage of normalized volume of regions of interest in perisylvian cortex, andperformance on sentence comprehension task

Patient

EnglishW.B.N.C.E.G.A.L.L.M.E.M.D.S.G.T.L.Z.

FrenchA.D.S.D.CD.H.D.R.M.F.R.F.S.G.S.J.V.

Normalized

total lesionvolume (cm3)

1487927

1013

1021

1415

1302332

11518

1037

2394

Percentage

ROI1(STG)

887826251900

857

740

29168

49000

of normalized

ROI2(F3t)

8200

180

210

1000

10043

1000

611800

ROI occupied

ROI3(F3o)

6600

560

210

1000

892318

1000

93611217

by lesion

ROM(AG)

33952424130000

00

1741

00001

ROI5(SG)

4043217400

<125

1

71571675

8404

11

All ROIs(perisylvian)

56513548

64

<148

2

64271861

529

836

Performance

Totalpercentcorrect

648138629575657660

193564372828365856

Syntacticcomplexityindex

0.290.240.360.180.170.310.250.130.14

0.130.440.570.610.300.370.510.260.43

ROI = region of interest; STG = superior temporal gyms; F3t = pars triangularis; F3o = inferior frontal gyrus/pars opercularis;AG = angular gyrus; SG = supramarginal gyrus.

types of parsing operations that construct different aspects ofsyntactic form (Frazier, 1990). Therefore, using independentt tests, we analysed the performance of patients with andwithout lesions in the anterior region of interest on 19separate measures that correspond to particular syntacticoperations. Each of these measures consisted of the difference

between a baseline sentence and a matched sentence, thatrequired the same processing as the baseline sentence, plusan additional syntactic operation. For instance, one measureof the ability to construct and interpret the passive formconsisted of performance on active sentences (baseline) minuspassive sentences that each had two noun phrases. The 19

Dow




Table 4 Measures of specific syntactic operations

Baseline Matched sentence Operation

1 Two-place active (ST1)2 Two-place active (ST1)3 Three-place active (ST5)4 Active conjoined theme (ST13)5 Object control (ST16)6 Object control (ST16)

7 Two-place active (ST1)8 Three-place active (ST5)9 Conjoined (ST8)

10 Object-subject (ST10)11 Subject-subject (ST 12)12 Object-object (ST11)13 Object-subject (ST 10)14 Three reflexive expressions (ST15)15 Two-place active (ST1)16 Simple active relexive-expression,

complex noun phrase (ST22)17 Three reflexive expressions (ST15)18 Two-place active (ST1)19 Simple active relexive-expression,

complex noun phrase (ST22)

P2 (ST2)PI (ST3)P3 (ST6)Passive conjoined agent (ST14)Subject control (ST17)Noun phrase-raising (English ST25)Passive object control (French ST25)Two-place cleft object (ST4)Three-place cleft object (ST7)Object-subject (STIO)Object-object (ST11)Subject-object (ST9)Subject-object (ST9)Subject-subject (ST12)Reflexives, simple (ST18)Reflexive, simple (ST20)Reflexive, complex NP (ST21)

Pronouns, simple (ST19)Pronoun, simple (ST24)Pronoun, complex noun phrase (ST23)

PassivePassivePassivePassiveAntecedent of pronounAntecedent of nounphrase-traceObject relativizationObject relativizationObject relativizationObject relativizationObject relativizationCentre embeddingCentre embeddingAntecedent of reflexiveAntecedent of reflexiveAntecedent of reflexive

Antecedent of pronounAntecedent of pronounAntecedent of pronoun

measures are listed in Table 4. None of these comparisonswas significant.

We also explored the effect of lesion size within a regionon performance through correlational analyses. We separatelycorrelated (i) overall accuracy on the entire set of 25 sentencetypes, (ii) the overall syntactic complexity score describedabove, and (iii) the 19 separate measures that correspond toparticular syntactic operations, with (a) normalized lesionvolume in the language zone, (b) normalized lesion volumein each of the five regions of interest, and (c) normalizedlesion volume in the anterior and posterior regions of interest.None of these 168 correlations were significant. To look forany non-linear relationships that might have obscured acorrelation, we created separate plots for each of these 168correlations. We found no evidence of non-linear relationsin any of these plots.

Finally, correlational analyses were performed to identifywhether lesion size in a particular region of interest affectedperformance, once the effect of overall lesion size had beentaken into account. The percentages of each region of interestoccupied by a lesion were correlated against the residuals oftwo regression analyses—one in which the normalized lesionvolume in the language zone was regressed against the overallaccuracy scores and one in which this value was regressedagainst the overall syntactic complexity scores. (NB In thisanalysis, the five regions of interest were reduced to threeby combining the two frontal and the two parietal regions ofinterest, resulting in regions of interest that represent thefrontal, parietal and temporal portions of the language zone.This reduction was undertaken to reduce the ratio ofindependent variables to cases, thereby allowing for theanalysis of these data by regression analyses.) There were

no significant correlations of lesion extent in any of theregions of interest with the residuals of either of theseregressions. This indicates that, when the effect of totalperisylvian lesion volume is removed, there is still noparticular area within this region in which lesion sizecorrelates with the degree of syntactic processing impairment.

These analyses indicate that there was no difference betweenthe patients with and without anterior perisylvian lesions withrespect to their overall level of performance, the magnitude ofa syntactic processing deficit and the presence of impair-ments of specific parsing operations. They also indicate thatthere was no relationship between lesion size in either theanterior or posterior language area and the overall magnitudeof a sentence comprehension deficit, the overall magnitude ofa syntactic processing deficit, or the magnitude of deficits inspecific syntactic operations. This suggests that lesions that areconfined to the posterior perisylvian cortex have essentiallythe same effect on syntactic processing as lesions that affectboth the posterior and anterior perisylvian regions.

However, these analyses could be misleading because thepatients were studied and scanned at different times inrelation to their lesions. We undertook four analyses todetermine whether this was likely to be the case. First, wedetermined that the mean interval from stroke to testing wasthe same in the patients with and without anterior lesions.For the 12 patients with anterior lesions, time since strokeranged from 4 to 84 months with a mean of 25 months(SD = 25); for the six patients without anterior lesions, timesince stroke ranged from 8 to 65 months with a mean of 28months (SD = 8). Secondly, we correlated the duration ofillness, measured in months between stroke and time oftesting, with the overall syntactic complexity score. Because

Dow




Table S Performance (number correct of 12 trials) of five patients on different sentence types

Sentence types

Sentences with full noun phraseTwo-place activeThree-place activeConjoinedActive conjoined themeThree reflexive-expressionsSimple active reflexive-expression

Sentences with pronouns or reflexivesReflexives, simple noun phrasePronouns, simple noun phraseSimple active reflexiveSimple active reflexive ('friend of X' subj)Simple active pronoun ('friend of X' subj)Simple active pronoun

Sentences with empty noun phrasesTwo-place passiveTruncated passiveTwo-place cleft objectThree-place passiveThree-place cleft objectSubject-object relativeObject-subject relativeObject-object relativeSubject-subject relativePassive conjoined agentObject controlSubject controlNoun phrase-raising/passivized object control

Total correct

Patients

E.M.

12119

108

10

89

12111112

12129925779

101165

227

F.S.

1257941

75

12069

2830104041522

109

G.S.

10127

1125

82

10568

1011116769567900

173

L.M.

121212121212

121212121212

12121212127

109

121212126

284

L.Z.

10739

1112

81012967

978782233

12952

181

See Appendix for descriptions of sentence types.

the values for illness duration were not normally distributed,we also correlated the log of illness duration with theoverall syntactic complexity score. Neither correlation wassignificant. Thirdly, we repeated all correlational analyses ina subset of 10 patients, who were tested between 3 and 24months after stroke. These patients were also selected so asto exclude the patient whose scan was not taken along thecanthomeatal plane, and to exclude the three patients whosescans were obtained within 2 weeks of stroke. None of thecorrelations were significant.

Fourthly, though there were insufficient numbers of patientsin this smaller selected set to compare those with lesionsthat only affected Broca's area with those with lesions thatspared this region, we were able to select five patients withroughly equal-size small lesions, three of which primarilyaffected Broca's area, and two of which spared it. Weanalysed their patterns of performance, which are shownin Table 5.

Three patients (E.M, F.S. and G.S.) had lesions that affectedBroca's area. Overall performance did not correspond to thepercentage of Broca's area that was affected in these patients.G.S., a 48-year-old French patient, had a lesion that occupied

<10% of Broca's area and extended through the pre- andpost-central gyri to minimally affect the supramarginal gyrus.Her overall comprehension score was 58%. E.M., a 57-year-old English patient, had a lesion that was almost entirelyrestricted to Broca's area, of which it occupied ~20%. Hisoverall comprehension score was 75%. F.S., a 60-year-oldFrench patient, had a lesion that was also largely restrictedto Broca's area and minimally affected the adjacent areas ofthe second frontal and pre-central gyri; the lesion occupied alittle less than half of Broca's area. His overall comprehensionscore was 36%. The two patients with lesions that sparedBroca's area also showed no correspondence between lesionsize and degree of impairment. L.Z., a 62-year-old Englishpatient, had a lesion primarily located in the superior temporalgyrus, of which it occupied 7%. Her overall comprehensionscore was 60%. L.M., a 21-year-old English patient, had alesion that occupied 19% of the superior temporal gyrus and13% of the angular gyrus. Her overall comprehension scorewas 95%.

Similarly, a comparison of the performance of the threesubjects with primarily Broca's area lesions and the two withlesions that spared this region showed no specific type of

Dow




sentence that was affected by lesions in either location. Allfive of the patients had problems comprehending at leastsome relative clauses. L.M., with the second largest lesionin the group of five patients, but the best overall performance,only had problems with sentences of this type and one othersentence type with an empty noun phrase. L.Z. (with aprimarily temporal lesion) had difficulty with a set ofsentences with pronouns and reflexives and full noun phrases,and with all the sentences with empty noun phrases exceptone type of passive sentence. The patients with primarilyanterior lesions had a variety of impairments other thanthose affecting relative clauses. E.M. showed impairments insentences in which the subject of the main clause was relatedto the subject of an embedded infinitive (subject control andnoun phrase-raising sentences), in sentences with three fullnoun phrases, and in some sentences with reflexives orpronouns. F.S. showed impairments in all but the simplestsentence types. G.S. showed impairments in longer sentenceswith full noun phrases (conjoined sentences and sentenceswith three full noun phrases), all sentences with reflexivesor pronouns except the most simple type, and several passivesentence types. Most of these patterns are interpretable inpsycholinguistic terms, but present no particular patternacross the patients with lesions in particular locations.

Overall, this more detailed analysis of single cases withsmall lesions of roughly comparable size, who were testedat about the same time after their strokes, illustrates that thedegree of variability found in quantitative and qualitativeaspects of patients' performances are not easily related tolesion location or the size of lesions in the anterior orposterior portion of the perisylvian association cortex.

DiscussionThe results of this study provide information about the neuralstructures that are involved in sentence comprehension. Theyshow that sentence comprehension is affected by lesions inboth the left and the right hemisphere, more so by the former.This finding is consistent with other reports in the literature(De Renzi and Fagiolini, 1978). The more specific resultsof this study pertain to syntactic processing in sentencecomprehension. They add to the evidence that lesionsthroughout the left perisylvian association cortex areassociated with disorders affecting this process. They alsoraise the question of a possible contribution of the righthemisphere to this aspect of sentence processing. We shalldiscuss the results for the right and the left hemisphereseparately.

The performance of the right-hemisphere population onsyntactically complex sentences was significantly lower thanthat of the normal control subjects. The analysis of theperformance of individual patients makes it unlikely that thepoorer performance of right-hemisphere patients than controlsubjects was due to a few patients in this group whowere right-hemisphere dominant for this aspect of languageprocessing. The finding that there were significant effects of

syntactic complexity—independent of sentence length—onthe performance of the right-hemisphere patients, providessupport for the view that the right hemisphere plays somerole related to assigning sentence structure and/or using it todetermine sentence meaning.

It is not yet clear what the role of the right hemisphere insyntactic processing is. The magnitude of syntactic processingimpairments was roughly normally distributed in both theright- and left-hemisphere lesioned patients, with a greaterdegree of impairment in the left-hemisphere group. Thissuggests that there might be a reduction in resources availablefor syntactic processing that varies in its extent in bothpopulations. The greater effect of left-hemisphere lesionscould reflect a specialization of this resource capacity forsyntactic processing within the left-perisylvian cortex; therole of the right hemisphere might be to provide a lessspecialized working memory capacity that makes a lessercontribution to syntactic processing (Caplan and Hildebrandt,1988; Waters et al., 1995). More specific protocols will haveto be used to determine exactly what aspects of the totalsentence comprehension process are accomplished by theright hemisphere. In addition, to determine the specificity ofany deficit in sentence comprehension for lesions in theterritory of the middle cerebral artery of the right hemi-sphere, it will be necessary to study patients with frontallesions and other lesions outside the perisylvian cortex andto compare their performances with those of patients withright-hemisphere perisylvian lesions.

The left-hemisphere damaged patients performed morepoorly and showed greater effects of syntactic processingthan those with right-hemisphere lesions. It is possible thatthe poorer performance of the left-hemisphere patients andthe presence of greater syntactic complexity effects in thispopulation compared with the right-hemisphere patients isdue to larger lesions in the left- than in the right-hemispheregroups. However, though this possibility cannot be ruled outwithout additional data, it seems unlikely that the left-hemisphere lesions were on average twice to three times aslarge as those in the right hemisphere. The most likelyexplanation for the poorer performance and greater syntacticcomplexity effects in the left-hemisphere patients is that theirlesions affected neural structures that are more cruciallyinvolved in syntactic processing.

The data from the 18 patients in whom CT scanning wasavailable indicate that deficits in syntactic processingfollow lesions in all parts of the perisylvian association cortexof the left hemisphere. Before considering the implicationsof this pattern for the functional neuroanatomy of language,we must ask whether the observed pattern of deficit-lesionrelationships might reflect deficiencies in the methodologyof this study.

The most obvious limitation of this study is that only 18subjects were scanned. However, this limitation should beseen within the context of previous research on this topic:the present study is the largest study of patients in whomradiological data have been obtained and who have been

Dow




tested for syntactic processing in sentence comprehension.In addition, the comprehension data available on each patientconsists of performance on 12 examples of each of 25sentence types, while most previous studies provide resultsfor at most three or four sentence types. Thus, though thelimited number of patients requires that caution be exercisedin accepting these results, the data constitute the largestdataset presently available on this topic and can provide atentative basis for theory construction.

A second concern is that differences in the time fromstroke to testing could have affected the results. Severalanalyses speak against this possibility. The magnitude of thefunctional impairment was not correlated with time sincestroke. Analyses of a subset of subjects whose strokes werebetween 3 months and 2 years of testing were identical tothe analyses of the larger group. Though neither of thesefindings rules out the possibility that patients with longerperiods of recovery could have lesser deficits, or that therecould be a non-linear relationship between time since lesionand degree of recovery, they combine to make thesepossibilities less likely to have obscured the relationship oflesion location and size to impairments on this task. Moreover,it should be born in mind that cortical re-organization post-stroke cannot explain the presence of deficits followinglesions to brain regions not premorbidly involved in theexercise of the function. Therefore, the finding that patientswith lesions in many parts of the perisylvian cortex havesyntactic processing deficits has implications for thefunctional neuroanatomy of this aspect of languageprocessing.

A third concern is that in this study we tested sentencecomprehension through the use of a single task, objectmanipulation and that the results thus reflect the demands ofthis task as well as those of sentence comprehension. Toaddress the concern that the results of an object-manipulationtask might not generalize to other tasks, in a separate study,we correlated the performance of 17 aphasic patients on 10sentence types on object manipulation and sentence-picturematching tasks (Caplan et al., 1995). The Spearman rankorder coefficient (p) for performance on the sentence typesacross the two tasks was 0.66 (P = 0.04). Thus, at leastfor the overall measure of sentence processing, patients'performance on the object manipulation task correlates witha very different task, and can be taken as an externally validmeasure of their sentence processing capacities.

A fourth issue is that this study tested sentence comprehen-sion in an 'off-line' task—one that reflects the end-point ofthe comprehension process, rather than examine 'on-line'syntactic processing—i.e. the time course of constructingsyntactic representations. It has been claimed that on-linemeasures reveal deficits in aspects of syntactic processing inBroca's aphasia, and this has been taken to implicate Broca'sarea as the locus of certain syntactic operations in sentencecomprehension (Zurif et al., 1993). If this claim is correct,we must identify the source of the qualitatively similar off-line impairments in patients with and without Broca's area

lesions in this study. One possibility is that certain on-lineoperations are affected by lesions in Broca's area, whileothers that interfere with the same final product of thecomprehension process are affected by lesions elsewhere.More detailed on-line studies of syntactic processing inaphasia and of the consequences of different disturbances ofon-line processing for final comprehension are needed beforethese possibilities can be settled.

Fifthly, in the present study we have not attempted todefine the white matter tracts that are lesioned in thesepatients. Damage to these tracts can cause de-efferentationand de-afferentation of cortical areas, with functional con-sequences that may be similar to those caused by lesions tothe regions themselves (Klippel, 1908; Geschwind, 1965;Kosslyn et al., 1993). The possible importance of whitematter lesions is highlighted by case D.S., who had a verysmall cortical lesion, but who only responded correctly to65% of the sentences and who had a syntactic complexityindex of 0.25. It is possible that analyses that took whitematter tracts into account might reveal a higher degree oflocalization of syntactic processing.

A final concern is that CT scanning primarily identifiesareas of necrosis, and is not very sensitive to the presenceof hypoperfusion or hypometabolism in cerebral tissue.Several studies in which investigators used [l8F]fluoro-deoxyglucose PET, have demonstrated larger areas of hypo-perfusion than those shown to be necrotic by CT scanningin aphasic patients (Metter et al., 1983, 1984, 1986, 1987,1989, 1990). It has been suggested that cortical areas thatare hypometabolic may not sustain normal cognitive functions(Kosslyn et al., 1993). Therefore, it is possible that a greaterdegree of localization might be observed if measurements ofmetabolism, blood flow and/or oxygen extraction were usedto assess CNS damage. One can hope that future work willbring together different measures of CNS function withextensive and detailed cognitive analyses of patients' deficits,to address these issues. Until such time, we are limited tothe currently available data.

Accepting these data, provisionally, we can ask what theirimplications are for the functional neuroanatomy of the leftperisylvian association cortex for syntactic processing. Thefact that lesions in all parts of the perisylvian cortex affectedsyntactic processing is consistent with one of two modelsthat have been proposed for the functional neuroanatomy ofcognitive processes. One is a model according to which thereare significant individual differences in the localization ofsyntactic processing across the population (Caplan, 1987a,b). The second is a neural net model, in which representationsare distributed throughout a region of the brain (McClellandand Rumelhart, 1986; McClelland and Kawamoto, 1986;McClelland et al., 1989).

Evidence against individual differences in localization ofsyntactic processing comes from the recent PET study,referred to above (Stromswold et al., 1996), that showed anincrease in rCBF during syntactic processing in the parsopercularis of Broca's area in all subjects studied. However,

Dow




only strongly right-handed college-educated males, betweenthe ages of 20 and 30 years, with no left-handed familymembers were studied in that experiment. If there areindividual differences in the localization of parts of the neuralsystem that is responsible for syntactic processing, thesedifferences may be related to sex, handedness, familialhandedness, age, educational level or other factors. Thepresent study did not include enough patients to determinewhether correlations between the degree of impairment insyntactic processing and the size of lesions in particularlocations would be greater if the correlations were confinedto subjects of a certain age, sex, educational level orhandedness profile. Larger studies, both involving deficit-lesion correlational analyses and functional activation inneurologically normal subjects, with more subjects in eachof these groups are needed to explore this issue.

The distributed neural net model maintains that the neuralsystem that is responsible for syntactic processing includesa cortical region that extends along the sylvian fissure. Thismodel would predict impairments in syntactic processingafter lesions throughout this region, and thus is compatiblewith the results of the present study. Distributed modelscould possibly also be compatible with the evidence forlocalization found in the PET study of Stromswold et al.(1996). It has been shown that neural net models can achievesome degree of internal structure; i.e. neural nets that aretrained to accomplish a function frequently develop in sucha way that a particular stimulus maximally activates aparticular subset of the units in the net (Plaut and Shallice,1993). This could correspond to a distributed system in whichthere is a local increase in activity, observable as an increasein rCBF, when a particular stimulus is processed. There isone aspect of the data that poses a challenge to the distri-buted neural net model; namely, the finding that therewas no correlation between total lesion size and severityof impairment. Most neural net models obey the principleof mass action (Lashley, 1950), such that the larger the lossof computational elements, the greater the overall decrementin performance (McClelland and Rumelhart, 1986; Pattersonet al., 1989). Possible areas of research thus include an effortto see if a neural net model that develops an internal structurecan be lesioned in such a way that there are similar effectsof lesions throughout the net but no effect of overall lesionvolume on performance.

In summary, the data presented here are consistent withthe conclusion that several regions of the left perisylviancortex form critical parts of a neural system responsible forsyntactic processing. Other data suggest some degree oflocalization of this function within the pars opercularis. Thecomplete picture is consistent with the models proposed byMesulam (1990) and Damasio (1992). which involve bothdistributed and localized features. Many aspects of thesemodels remain to be developed to account for the entirepattern of results seen in both deficit-lesion correlationalstudies and functional neuroimaging studies with normalsubjects.

AcknowledgementsWe wish to thank Pierre Delplas, David Kennedy, RandallBenson, Jeremy Schmahmann and David Gow for assistanceobtaining, registering and interpreting the CT scans andperforming the regression analyses. The work reported herewas partially supported by a grant from the National Instituteof Deafness and other Communication Disorders (DC00942).

ReferencesBelliveau JW, Kennedy DN Jr, McKinstry RC, Buchbinder BR,Weisskoff RM, Cohen MS, et al. Functional mapping of the humanvisual cortex by magnetic resonance imaging. Science 1991; 254:716-19.

Butler-Hinz S, Caplan D, Waters G. Characteristics of syntacticcomprehension deficits following closed head injury versus leftcerebrovascular accident. J Speech Hear Res 1990; 33: 269-80.

Caplan D. Discrimination of normal and aphasic subjects on a testof syntactic comprehension. Neuropsychologia 1987a; 25: 173-84.

Caplan D. Neurolinguistics and linguistic aphasiology. Cambridge(UK): Cambridge University Press, 1987b.

Caplan D, Futter C. Assignment of thematic roles to nouns insentence comprehension by an agrammatic patient. Brain Lang1986: 27: 117-134.

Caplan D, Hildebrandt N. Disorders of syntactic comprehension.Cambridge (MA): MIT Press, 1988.

Caplan D, Baker C, Dehaut F. Syntactic determinants of sentencecomprehension in aphasia. Cognition 1985; 21: 117-75.

Caplan D, Waters GS, Hildebrandt N. Sentence comprehension inaphasic patients in sentence-picture matching tests. Brain Lang1995; 51: 147-9.

Caramazza A, Zurif EB. Dissociation of algorithmic and heuristicprocesses in language comprehension: evidence from aphasia. BrainLang 1976; 3: 572-82.

Chomsky N. Knowledge of language: its nature, origin, and use.New York: Praegen 1986.

Damasio AR. Aphasia [see comments]. [Review]. N Eng J Med1992; 326: 531-9.

Damasio H. Damasio AR. Lesion analysis in neuropsychology. NewYork: Oxford University Press, 1989.

De Renzi E, Fagiolini P. Normative data and screening power of ashortened version of the Token Test. Cortex 1978; 14: 41-9.

Fox PT, Lancaster JL. Proceedings: BrainMap: Workshop I. SanAntonio (TX): Research Imaging Center/UTHSCSA, 1993.

Fox PT, Perlmutter JS, Raichle ME. A stereotactic method ofanatomical localization for positron emission tomography. J ComputAssist Tomogr 1985; 9: 141-53.

Fox PT, Mintun MA, Raichle ME, Miezin FM, Allman JM, VanEssen DC. Mapping human visual cortex with positron emissiontomography. Nature 1986; 323: 806-19.

Frazier L. Sentence processing: a tutorial review. In: Coltheart M,

Dow




editor. Attention and performance XII: The psychology of reading.Hove: Lawrence Erlbaum, 1987a: 559-86.

Frazier L. Theories of sentence processing. In: Garfield JL, editor.Modularity in knowledge pepresentation and natural-languageunderstanding. Cambridge (MA): MIT Press, 1987b: 291-307.

Frazier L. Against lexical generation of syntax. In: Marslen-WilsonW, editor. Lexical representation and process. Cambridge (MA):MIT Press, 1989: 505-28.

Frazier L. Exploring the architecture of the language-processingsystem. In: Altmann GTM , editor. Cognitive models of speechprocessing: psycholinguistic and computational perspectives.Cambridge (MA): MIT Press, 1990: 409-33.

Geschwind N. Disconnexion syndromes in animals and man. Brain1965; 88: 237-94, 585-644.

Geschwind N, Galaburda AM. Cerebral lateralization: biologicalmechanisms, associations, and pathology: I—III. A hypothesis anda program for research. [Review]. Arch Neurol 1985; 42: 428-59,521-52, 634-54.

Hagoort P, Brown C, Groothusen J. The syntactic positive shift(SPS) as an ERP measure of syntactic processing. Lang CognitProcess 1993; 8: 439-83.

Jackson HH. On the nature of the duality of the brain. Med Press1874; 1: 19-41.

Just MA, Carpenter PA. A capacity theory of comprehension:individual differences in working memory. Psychol Rev 1992; 99:122-49.

Klippel MM. Discussion surl'aphasie. Rev Neurol 1908; 16:611-36.

Kluender R, Kutas M. Subjacency as a processing phenomenon.Language and Cognitive Processes 1993; 8: 573-633.

Kosslyn S, Daly P, McPeek R, Alpert N, Kennedy D, Caviness V.Using locations to store shape: an indirect effect of a lesion. CerebCortex 1993; 3: 567-82.

Lashley KS. In search of the engram. Sympos Soc Exp Biol 1950;4: 454-82.

Linebarger MC, Schwartz MF, Saffran EM. Sensitivity togrammatical structure in so-called agrammatic aphasics. Cognition1983; 13: 361-92.

MacDonald MC. Probabilistic constraints and syntactic ambiguityresolution. Lang Cognit Process 1994; 9: 157-201.

Mazoyer BM, Tzourio N, Frak V, Syrota A, Murayama N, LevrierO, et al. The conical representation of speech. J Cognit Neurosci1993; 5: 467-79.

McClelland JL, Kawamoto AH. Mechanisms of sentence processing:assigning roles to constituents of sentences. In: McClelland JL,Rumelhart DE, editors. Parallel distributed processing, Vol. 2.Cambridge (MA): MIT Press, 1986.

McClelland JL, Rumelhart DE, editors. Parallel distributedprocessing, Vol. 2. Cambridge (MA): MIT Press, 1986.

McClelland JL, St John M, Taraban R. Sentence comprehension: aparallel distributed processing approach. Lang Cognit Process 1989;4: 287-335.

Mesulam M-M. Large-scale neurocognitive networks and distributedprocessing for attention, language, and memory. [Review], AnnNeurol 1990; 28: 597-613.

Metter E, Riege W, Hanson W, Jackson C, Kempler D, VanLanckerD. Comparison of metabolic rates, language and memory, andsubcortical aphasias. Brain Lang 1983; 19: 33^17.

Metter EJ, Riege WH, Hanson WR, Camras LR, Phelps ME, KuhlDE. Correlations of glucose metabolism and structural damage tolanguage function in aphasia. Brain Lang 1984; 21: 187-207.

Metter EJ, Kempler D, Hanson R, Jackson C, Mazziotta JC,Phelps ME. Cerebral glucose metabolism: differences in Wernicke's,Broca's and conduction aphasias. Clin Aphasiol 1986; 16: 97-104.

Metter EJ, Kempler D, Jackson CA, Hanson WR, Riege WH,Camras LR, et al. Cerebellar glucose metabolism and chronicaphasia. Neurology 1987; 37: 1599-606.

Metter EJ, Kempler D, Jackson C, Hanson W, Mazziotta JC, PhelpsME. Cerebral glucose metabolism in Wernicke's, Broca's andconduction aphasia. Arch Neurol 1989; 46: 27-34.

Metter EJ, Hanson WR, Jackson CA, Kempler D, van Lancker D,Mazziotta JC, et al. Temporoparietal cortex in aphasia. Arch Neurol1990; 47: 1235-8.

Miceli G, Mazzucchi A, Menn L, Goodglass H. Contrasting casesof Italian agrammatic aphasia without comprehension disorder.Brain Lang 1983; 19: 65-97.

Mohr JP, Pessin MS, Finkelstein S, Funkenstein HH, Duncan GW,Davis KR. Broca aphasia: pathologic and clinical. Neurology 1978;28: 311-24.

Nespoulous J-L, Dordain M, Perron C, Ska B, Bub D, Caplan D,et al. Agrammatism in sentence production without comprehensiondeficits: reduced availability of syntactic structures and/or ofgrammatical morphemes? A case study. Brain Lang 1988; 33:273-95.

Neville H, Nicol JL, Barss A, Forster Kl, Garrett MF. Syntacticallybased sentence processing classes: evidence from event-related brainpotentials. J Cognit Neurosci 1991; 3: 151-65.

Osterhout L, Holcomb PJ. Event-related brain potentials elicited bysyntactic anomaly. J Mem Lang 1992; 31: 785-806.

Osterhout L, Holcomb PE. Event-related potentials and syntacticanomaly: evidence of anomaly detection during the perception ofcontinuous speech. Lang Cognit Process 1993; 8: 413-37.

Patterson K, Seidenberg MS, McClelland JL. Connections anddisconnections: acquired dyslexia in a computational model ofreading processes. In: Morris RGM, editor. Parallel distributedprocessing: implications for psychology and neurobiology. Oxford:Clarendon Press, 1989: 131-81.

Pinker S. The language instinct: the new science of language andmind. London: Allen Lane, 1994.

Plaut DC, Shallice T. Deep dyslexia: a case study of connectionistneuropsychology. Cognit Neuropsychol 1993; 10: 377-500.

Rademacher J, Galaburda AM, Kennedy DN, Filipek PA, CavinessVS Jr. Human cerebral cortex: localization, parcellation, andmorphometry with magnetic resonance imaging. J Cognit Neurosci1992: 4: 352-74.

Dow




Rademacher J, Caviness VS Jr. Steinmetz H, Galaburda AM.Topographical variation of the human primary cortices: implicationsfor neuroimaging, brain mapping, and neurobiology. Cereb Cortex1993; 3: 313-29.

Reinhart T. Anaphora and semantic interpretation. London: CroomHelm, 1983.

Rochon E, Waters GS, Caplan D. Sentence comprehension inpatients with Alzheimer's disease. Brain Lang 1994; 46: 329^9.

Sanides F. The cyto-myeloarchitecture of the human frontal lobeand its relation to phylogenetic differentiation of the cerebral cortex.J Himforsch 1964; 6: 269-82.

Schwartz M, Saffran E, Marin O. The word order problem inagrammatism I: Comprehension. Brain Lang 1980; 10: 249-62.

Schwartz MF. Linebarger MC, Saffran EM, Pate DS. Syntactictransparency and sentence interpretation in aphasia. Lang CognitProcess 1987: 2: 85-113.

Seines OA, Knopman DS, Niccum N, Rubens AB, Larson D.Computed tomographic scan correlates of auditory comprehensiondeficits in aphasia: a prospective recovery study. Ann Neurol 1983;13: 558-66.

Stromswold K, Caplan D, Alpert N, Rauch S. Localization ofsyntactic comprehension by positron emission tomography. BrainLang 1996. In press.

Talairach J, Tournoux P, Co-planar stereotaxic atlas of the humanbrain. Stuttgart: Thieme, 1988.

Tokunaga A, Takase M, Otani K. The glabella-inion line as abaseline for CT scanning of the brain. Neuroradiology 1977; 14:67-71.

Tramo MJ, Baynes K, Volpe BT. Impaired syntactic comprehensionand production in Broca's aphasia: CT lesion localization andrecovery patterns. Neurology 1988; 38: 95-8.

Tyler LK. Real-time comprehension processes in agrammatism: acase study. Brain Lang 1985; 26: 259-75.

Vanier M, Caplan D, CT-scan correlates of agrammatism. In: MennL, Obler LK, editors. Agrammatic aphasia. Amsterdam: Benjamins,1990: 97-114.

Vanier M, Lecours AR, Ethier R, Habib M, Poncet M, Milette PC,et al. Proportional localization system for anatomical interpretationof cerebral computed tomograms. J Comput Assist Tomogr 1985;9: 715-24.

Waters GS, Caplan D, Rochon E. Processing capacity and sentencecomprehension in patients with Alzheimer's disease. CognitNeuropsychol 1995; 12: 1-30.

Zurif E, Swinney D, Prather P, Solomon J, Bushell C. An on-lineanalysis of syntactic processing in Broca's and Wernicke's aphasia.Brain Lang 1993; 45: 448-64.

Received July 12, 1995. Revised November 13, 1995.Accepted December 29, 1996

Dow




Appendix: Sentence typesBaseline (simple) sentences: no referentially dependentnoun phrases(1)* Two-place active

The frog hit the monkey.(5)* Three-place active

The rabbit passed the cow to the goat.(8)* Conjoined

The monkey scratched the rabbit and patted theelephant.

(13) Active conjoined themeThe frog patted the monkey and the elephant.

(15) Three reflexive-expressionsThe old man knew that his friend scratched theboy.

(22) Simple active reflexive-expression ('friend of X'subject)

The father of the boy pointed to the old man.

Overt referential dependencies ('himself, 'him')(18) Reflexives, simple noun phrase

The old man said that the father hit himself.(19) Pronouns, simple noun phrase

The old man believed that the father tickled him.(20) Simple active reflexive

The old man kicked himself.(21) Simple active reflexive ('friend of X' subject)

The father of the boy scratched himself.(23) Simple active pronoun ('friend of X' subject)

The father of the boy kicked him.(24) Simple active pronoun

The old man tickled him.Empty referential dependencies(2)f Two-place passive

The monkey was hit by the frog.(3) Truncated passive

The rabbit was patted .

(4)^ Two-place cleft objectIt was the cow that the rabbit kissed .

(6^ Three-place passiveThe elephant was given to the monkeyby the frog.

(7)1 Three-place cleft objectIt was the goat that the rabbit gavecow.

to the

(9)f Subject-object relativeThe monkey that the rabbit grabbed shook thegoat.

(10) Object-subject relativeThe goat hit the rabbit that grabbed the cow.

(11) Object-object relativeThe monkey tickled the frog that the goatshook .

(12) Subject-subject relativeThe frog that held the cow caught theelephant.

(14) Passive conjoined agentThe elephant was hitfrog.

(16)* Object control

by the monkey and the

The old man told the father(17) Subject control

to pray.

to sleep.The old man swore to the father _(25)f Noun phrase-raising (English)

The old man seems to the father to bebending over.

Passivized object control (French)Le vieillard a t incit par lep re manger.

*Baseline sentence type used in syntactic complexity analysis;Sentence types with empty referential dependencies andnoncanonical word order.

Dow



Dow



location of lesions in stroke patients with deficits in syntactic - brain

Documents