Brain & Language 152 (2016) 14–27

Contents lists available at ScienceDirect

Brain & Language

journal homepage: www.elsevier .com/locate /b&l

Deficit-lesion correlations in syntactic comprehension in aphasia

http://dx.doi.org/10.1016/j.bandl.2015.10.0050093-934X/� 2015 Elsevier Inc. All rights reserved.

⇑ Corresponding author at: Neuropsychology Laboratory, 175 Cambridge Street,Suite 340, Fruit Street, Boston, MA 02114, United States.

E-mail address: dcaplan@partners.org (D. Caplan).

David Caplan ⇑, Jennifer Michaud, Rebecca Hufford, Nikos MakrisNeuropsychology Laboratory, Department of Neurology, Massachusetts General Hospital, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 7 December 2014Revised 9 September 2015Accepted 24 October 2015Available online 10 December 2015

Keywords:AphasiaSyntactic comprehensionLocalization

The effects of lesions on syntactic comprehension were studied in thirty-one people with aphasia (PWA).Participants were tested for the ability to parse and interpret four types of syntactic structures and ele-ments – passives, object extracted relative clauses, reflexives and pronouns – in three tasks – objectmanipulation, sentence picture matching with full sentence presentation and sentence picture matchingwith self-paced listening presentation. Accuracy, end-of-sentence RT and self-paced listening times foreach word were measured. MR scans were obtained and analyzed for total lesion volume and for lesionsize in 48 cortical areas. Lesion size in several areas of the left hemisphere was related to accuracy in par-ticular sentence types in particular tasks and to self-paced listening times for critical words in particularsentence types. The results support a model of brain organization that includes areas that are specializedfor the combination of particular syntactic and interpretive operations and the use of the meanings pro-duced by those operations to accomplish task-related operations.

� 2015 Elsevier Inc. All rights reserved.

1. Introduction

This paper presents new data regarding deficit-lesion correla-tions in the area of syntactic comprehension in PWA. We beginwith a brief introduction to syntactic comprehension, then reviewwork relating lesions to disorders of syntactic comprehension, andthen present our study.

The term ‘‘syntactic comprehension” refers to the processes ofassigning syntactic structure to linguistic input (often called ‘‘pars-ing”) and using that structure to determine propositional meanings(sometimes called ‘‘interpretation”). Syntactic comprehension is animportant human cognitive function because propositional mean-ings express relations between concepts that are not inherent inword meanings themselves, such as who is accomplishing andreceiving an action (thematic roles of agent, theme, etc.), howmen-tal states are related to one another (what a person believes,desires, intends, etc.), and others, which are critical to the powerof language to represent the world and to aid in thinking and com-municating. The propositional meaning of a sentence is deter-mined by its syntactic structure, not simply by associating wordsto one another, allowing sentences to express unlikely or evenimpossible relations between items. For example, sentences suchas ‘‘The man bit a dog” or ‘‘A dog was bitten by the man” mean thata particular man bit a dog, not the more likely event that a dog bit

the man, because the syntactic structure of these sentences forcesthis interpretation. The ability to represent unlikely events allowshumans to express ideas about what might happen under variouspossible circumstances; that is, to express counterfactual state-ments. This ability is critical for inter-individual communicationthat is used in planning of actions, scientific work, instruction,social organization, and other human activities that involve morethan one person.

Although there is considerable disagreement about manydetails of syntactic structures and how they are constructed fromauditory input, there is also widespread agreement about basic fea-tures of these representations and their processing. Virtually allcontemporary linguistic theories maintain that syntactic represen-tations are complex sets of syntactic categories (noun, verb, verbphrase, etc.) that are hierarchically organized, and that differentstructures – or different relations among categories in these hierar-chical structures – determine different aspects of propositionalmeaning (thematic roles, the antecedents of pronouns and reflex-ives, etc.) (Chomsky, 1995; Culicover & Jackendoff, 2005). Virtuallyall models also agree that, although some aspects of syntactic rep-resentations – such as their hierarchical structure – are found inother domains such as mathematics, music and even action organi-zation, and in some animal functions, the specific combination ofnodes, organization, and semantic interpretation found in syntaxis a unique biological entity (Caplan & Gould, 2008). Virtually allmodels of speech comprehension maintain that syntactic struc-tures are built and interpreted incrementally (as each word isencountered) (Hale, 2001; Levy, 2008; Lewis & Vasishth, 2005).

D. Caplan et al. / Brain & Language 152 (2016) 14–27 15

On-line behavioral measures of syntactic comprehension reflectthe operation of parser/interpreter more directly than end-of-sentence measures, which are affected by memory for the contentof a sentence, response selection, and other cognitive operations.

The areas of the brain that are involved in syntactic comprehen-sion are of interest for many reasons. Clinically, knowing whatbrain areas support this function would be expected to help predictthe effects of lesions. Scientifically, understanding the neural basisfor syntactic comprehension would provide information about theway the human brain is organized to support a unique, anduniquely human, function. This could be a model for other humancognitive functions, or provide evidence that different human cog-nitive functions are supported in different ways by the brain.

The effects of lesions on syntactic comprehension provide infor-mation about the areas of the brain that are necessary for this func-tion. The ‘‘deficit-lesion correlation” approach requires both ananalysis of the deficits in normal functions that, along with com-pensatory mechanisms, produce the observed abnormal behaviorsand an analysis of the brain areas that are lesioned. Lesions can bedescribed in many ways (e.g., as areas of infarction, areas of hypop-erfusion, areas of hypometabolism, patterns of disconnection, etc.);the focus of most work has been on areas of infraction and theimplications of their associated deficits for functional specializa-tion of areas of the brain (‘‘localization of function”). We brieflyreview the criteria for demonstrating that a person with aphasiahas a deficit in syntactic comprehension and methods for estab-lishing the location and size of a lesion.

Criteria for diagnosing a syntactic comprehension emergedfrom the first paper on this subject, by Caramazza and Zurif(1976). These authors described PWAwho could not match syntac-tically complex, semantically reversible (‘‘experimental”) sen-tences (1) to a picture but could match syntactically simple,semantically reversible (‘‘baseline”) sentences (2) and semanticallyconstrained sentences (3) to pictures (the term ‘‘semanticallyreversible” indicates that any person or item in the sentence couldeither perform or receive the action depicted by the verb in thesentence).

1.

Syntactically complex, semantically reversible sentence The boy who the girl chased was tall

2.

Syntactically simple, semantically reversible sentence The boy chased the tall girl

3.

Semantically irreversible sentence The apple the boy was eating was red

Caramazza and Zurif (1976) explained the selectivity of the abnor-mal comprehension performance in the following way. The goodperformance on semantically constrained (or ‘‘irreversible”) sen-tences (3) indicated that their PWA were able to understand wordsand to combine the concepts that words evoked into propositions.The good performance on syntactically simple, semantically reversi-ble sentences (2) further indicated that they could apply simple‘‘heuristics,” such as assigning the nouns in a sentence the thematicroles of agent and theme on the basis of their order of occurrence.The poor performance on syntactically complex, semantically rever-sible sentences (1) indicated that they could not assign the thematicroles in a sentence by applying syntactic rules to sequences ofwords.

Since 1976, the criteria for diagnosing a syntactic comprehen-sion deficit have been refined, although the essentials of the crite-ria have remained the same. The intent and effect of therefinements have been to increase the likelihood that a personwith aphasia who has an observed pattern of behavior has a deficitin syntactic comprehension and not in a related functional ability.

One widely adopted practice is to match the baseline and exper-imental sentences more closely. Thus, for instance, rather than use(2) as the baseline for (1), a baseline such as (4) might be used:

4.

Syntactically simple, matched, semantically reversiblesentence The boy who chased the girl was tall

(4) is semantically reversible and can be understood by using aheuristic based on the order of the nouns in the sentence (thesentence-initial NP is the agent of every verb), and so qualifies asa baseline sentence. The fact that (4) is matched to (1) in terms ofwords, length, and number of thematic roles allows for the conclu-sion that selectively poor performance on (1) is due to an inabilityto apply the parsing and interpretive operations found in (1) andnot in (4) more clearly than a difference in performance between(1) and (2) does.

A second change in approach has been to study syntactic com-prehension in PWA using on-line measures rather than end-of-sentence accuracy. As noted, end-of-sentence performanceinvolves memory for sentence meaning (and possibly form) andis distant from the incremental processing of syntactic structure.Studies using word monitoring (Tyler, 1985; Tyler, Ostrin, Cooke,& Moss, 1995), on-line anomaly detection (Shankweiler, Crain,Gorell, & Tuller, 1989), cross modal priming (Balogh, Zurif,Prather, Swinney, & Finkel, 1998; Burkhardt, Piñango, & Wong,2003; Love, Swinney, Walenski, & Zurif, 2008; Love, Swinney, &Zurif, 2001), self-paced listening (Caplan & Waters, 2003; Caplan,Waters, DeDe, Michaud, & Reddy, 2007) and eye tracking in sen-tence picture matching (Hanne, Sekerina, Vasishth, Burchert, &De Bleser, 2011; Meyer, Mack, & Thompson, 2012) and in the visualworld paradigm (Dickey, Choy, & Thompson, 2007; Dickey &Thompson, 2009; Thompson & Choy, 2009) have provided empiri-cal data relevant to mechanisms that might underlie these disor-ders. Hypotheses regarding the mechanisms that producesyntactic comprehension disorders that have emerged from on-line studies include the ideas that the deficit consists of slowed lex-ical processing (Balogh et al., 1998; Love et al., 2001, 2008), slowedprocessing of syntactic structure (Burkhardt et al., 2003), slowedintegration of lexical and syntactic information (Meyer et al.,2012), and excessive sensitivity to meanings derived from sourcesother than parsing and interpretation (Caplan, 2015).

A third change, not widely adopted, is to gather information onthe performance of PWA in more than one task. Caplan, DeDe, andMichaud (2006), Caplan, Waters, DeDe, Michaud, et al. (2007) andCaplan, Michaud, and Hufford (2013a) showed that performance inindividual PWA differed for the same sentence type in differenttasks, indicating that a performance in one task is not a reliablemeasure of parsing and interpretation. Caplan et al. (2006,2013a) and Caplan, Waters, DeDe, Michaud, et al. (2007) arguedthat abnormalities in syntactic comprehension that are seen onlyin one task are deficits in the ability to combine parsing and inter-pretation with the operations needed to perform a task, not deficitsin parsing and interpretation themselves. Deficits in parsing andinterpretation themselves should be task-independent.

On the neurological side, delineation of lesions on CT and MRscans is now standard in lesion-deficit correlation studies of syn-tactic comprehension deficits. The methods used to determinethe location and size of lesions vary across studies and all face chal-lenges in identifying some lesion boundaries (e.g., boundaries oflesions that abut the subarachnoid space or the ventricle). A fewstudies have included imaging of perfusion and metabolism, whichusually identify larger areas of lesion than seen on CT or ‘‘struc-tural” MR scans, and of white matter connectivity.

16 D. Caplan et al. / Brain & Language 152 (2016) 14–27

We know of know of six papers in the literature that that reportdeficit-lesion correlations for syntactic comprehension deficits inchronic stroke PWA that meet the criteria that deficits have beenclosely related to syntactic comprehension and that use modern,reliable imaging techniques—Caplan, Hildebrandt, and Makris(1996), Caplan, Waters, and DeDe (2007), Warren, Crinion,Lambon Ralph, and Wise (2009), Thothathiri, Kimberg, andSchwartz (2012), Tyler, Wright, Randall, Marslen-Wilson, andStamatakis (2010) and Tyler et al. (2011).1

Caplan, Hildebrandt, and Makris (1996) obtained CT scans in 18PWA with left hemisphere strokes. The cortex and lesions weresegmented and parcellated using computer-assisted algorithms.Five perisylvian regions of interest (ROI)s were defined followingthe Rademacher, Galaburda, Kennedy, Filipek, and Caviness(1992) criteria: the pars triangularis and the pars opercularis ofthe inferior frontal gyrus, the supramarginal gyrus, the angulargyrus, and the first temporal gyrus excluding the temporal tipand Heschl’s gyrus. Normalized lesion volume was calculatedwithin each ROI. Syntactic comprehension was assessed using anobject manipulation task presenting 12 examples of each of 25sentence types. The ‘‘experimental” and ‘‘baseline” sentence typeswere constructed to assess the ability to understand sentencescontaining overt referentially dependent noun phrases (pronounsor reflexives) and sentences containing what are referred to asphonologically empty noun phrases (PRO, NP-trace and wh-trace)in Chomsky’s mode, of syntax (Chomsky, 1986, 1995). The compar-ison of ‘‘experimental” and ‘‘baseline” sentence types resulted innineteen separate measures of particular syntactic operations.

None of the correlations between these nineteen measures ofparticular syntactic operations with normalized lesion volume inthe language zone, normalized lesion volume in each of the five

1 Several papers that we are aware of have been said to provide data that arerelevant to syntactic comprehension but fail to meet the criteria above. In severalpapers, the behavioral criteria for a syntactic comprehension deficit were not met.Karbe et al. (1989) related hypometabolism (based on FDG PET) to comprehension onthe Western Aphasia Battery. The WAB does not distinguish between semanticallyreversible and irreversible sentences, or present pairs of reversible sentences thatdiffer minimally from one another. Kempler, Curtiss, Metter, Jackson, and Hanson(1991) also measured hypometabolism based on FDG PET and related it toperformance on the Token Test. The Token Test requires the examinee to designateor manipulate different numbers of tokens, each of which is described by phraseswith different numbers of adjectives. It therefore creates a strong confound betweenshort term memory and syntactic comprehension. The final section of the Token Testpresents a larger range of sentences but does not contain minimal baseline-experimental pairs; Kempler et al. did not report correlations of PET measures withthe final section of the Token Test. Two papers (Amici et al., 2007; Dronkers, Wilkins,Van Valin, Redfern, & Jaeger, 2004) used the Curtiss–Yamada ComprehensiveLanguage Evaluation (CYCLE) to assess syntactic comprehension. The CYCLE is asentence picture matching test with both syntactic and lexical foils for many items,and both papers that used the CYCLE did not separate errors in which lexical foilswere selected from errors in which syntactic foil were selected, thereby includingerrors that reflect disorders of word comprehension in their measure of syntacticcomprehension. Amici et al. (2007) divided the sentences in the CYVCLE into threesets and reported effects of lesions for each set, but each set contained some sentencetypes with lexical foils that were not removed from the analyses. In addition, theCYCLE does not allow comparison of performance on matched baseline andexperimental reversible sentences. With respect to lesion measurement, in manyearly studies, lesion location was inferred on the basis of aphasic syndrome (e.g.,individuals with Broca’s aphasia were assumed to have lesions in Broca’s area –Caramazza & Zurif, 1976; Grodzinsky, 1990) or on the basis of a combination ofbehavioral and neurological data (Caplan, Baker, & Dehaut, 1985). In others, CT andMR scans were analyzed by applying templates by eye (Dick et al., 2001; Dronkerset al., 2004), using a method that is known to have low inter-observer reliability(correlations of about 80% for analyses that assign the quartile involvement of largeregions of interest by a lesion: Naeser & Hayward, 1978). Dick et al. (2001) found norelation between lesion location in 56 PWA and performance on an actor identifi-cation test with 16 examples of each of four sentence types – active, subject cleft,object cleft, and passive. Tramo, Baynes, and Volpe (1988) showed unanalyzed CTscans in three PWA with Broca’s aphasia. The one PWA who had a posterior lesionwas able to matched reversible passive sentences to pictures two years after herstroke; the 2 PWA who had anterior lesions were at chance.

ROIs, and normalized lesion volume in the anterior and posteriorROIs were significant. These correlations remained insignificantwhen the effect of overall lesion size was partialled out. The resultsall remained unchanged in ten PWA who were studied andscanned at the same time relative to their lesions. Detailed analy-ses of single cases with small lesions of comparable size, who weretested at about the same time after their strokes, indicated that thedegree of variability found in quantitative and qualitative aspectsof PWA’ performances was not related to lesion location or the sizeof lesions in the anterior or posterior portion of the perisylvianassociation cortex. Caplan et al. (1996) concluded that these resultswere inconsistent with both localizationist and distributed modelsof the neural basis for syntactic comprehension, and that they sug-gested that different brain regions were required for syntacticcomprehension in different individuals.

Caplan, Waters, and DeDe (2007) studied 42 PWA secondary toleft hemisphere strokes and 25 control subjects for the ability toassign and interpret three syntactic structures (passives, objectextracted relative clauses, and reflexive pronouns) in enactment,sentence–picture matching and grammaticality judgment tasks.The sentence–picture matching and grammaticality judgmenttasks were presented both with uninterrupted auditory presenta-tion and in a self-paced listening fashion (as in Caplan & Waters,2003; Ferreira, Anes, & Horine, 1996; Ferreira, Henderson, Anes,Weeks, & McFarlane, 1996). Differences in accuracy on experimen-tal and baseline sentences were used as measures of the ability toconstruct and utilize specific syntactic structures to determineaspects of sentence meaning, and a ‘‘combined syntactic complex-ity score” was obtained (accuracy on passive and object relatives vsactives and subject relatives). Self-paced listening times correctedfor word duration and frequency were used as measures of localprocessing load. Differences in these times at words at which pars-ing and interpretive operations applied in experimental sentencesand corresponding words in corresponding baseline sentenceswere used as measures of on-line parsing and interpretation.

Magnetic resonance scans and 5-deoxyglucose positron emis-sion tomography data were obtained on 31 PWA and 12 controls.Six cortical regions of interest were determined using the samealgorithms as in Caplan et al. (1996). Four encompassed the peri-sylvian area: the inferior frontal region (consisting of the inferiorfrontal gyrus pars opercularis and triangularis and the adjacentportion of the frontal operculum), the posterior half of the superiortemporal gyrus (corresponding to Wernicke’s area), the inferiorparietal lobe (consisting of the angular and supramarginal gyri),and the insula. Two were outside the perisylvian area – the inferioranterior temporal lobe (consisting of the anterior inferior temporalgyrus and the temporal pole), and the superior parietal lobe (con-sisting of the region posterior to the postcentral gyrus and the pre-cuneus). These six ROIs included all the areas of the brain that hadbeen related to syntactic comprehension in either deficit-lesioncorrelation or functional neuroimaging studies at the time of pub-lication. The percent of these ROIs that was lesioned on MR and themean PET counts per voxel in these ROIs (normalized to contrale-sional brain areas in the distribution of the anterior and posteriorcerebral arteries, not affected by transneuronal degeneration viathe corpus callosum from infarcted areas of the left hemisphere,and cerebellum) were calculated.

Regression analyses were used to relate lesion measures andperformance. Percent lesion volume in Wernicke’s area, the infe-rior parietal lobe and the anterior inferior temporal area were sig-nificant predictors of the combined syntactic complexity score inobject manipulation, and total lesion volume and total subcorticallesion volume were significant predictors of the combined syntac-tic complexity score in sentence picture matching with uninter-rupted presentation. FDG PET activity in the left inferior parietallobe and in the entire perisylvian cortex were significant predictors

2 Tyler et al. (2010) also conducted an fMRI study in which BOLD signal activityassociated with different conditions was reported. This aspect of their study is norelated to deficit-lesion correlations.

D. Caplan et al. / Brain & Language 152 (2016) 14–27 17

of the combined syntactic complexity score in object manipulation,and FDG PET activity in the insula was a significant predictor of thecombined syntactic complexity score in sentence picture matchingwith uninterrupted presentation. No neural measures were signif-icant predictors of on-line processing. Examination of individualparticipants showed that PWA who performed at similar levelsbehaviorally had lesions of very different sizes, and that PWA withequivalent lesion sizes varied greatly in their level of performance.Caplan, Waters, and DeDe (2007) concluded that these results wereconsistent with those of Caplan et al. (1996) in not providing sup-port for either localizationist or distributed models of the neuralbasis for syntactic comprehension, and in suggesting that differentbrain regions were required for syntactic comprehension in differ-ent individuals. A point that was not discussed by Caplan, Waters,and DeDe (2007), but that we introduce here, is that the effects oflesions differed for object manipulation and sentence picturematching, suggesting the lesions in these brain areas support theinteraction of syntactic comprehension and task performance,not syntactic comprehension itself.

Warren et al. (2009) studied the effect of lesion location andsize and the functional connectivity of the left anterolateral supe-rior temporal cortex (LalSTC) in 24 PWA and 11 controls who weretested on spoken and written word and sentence comprehensionusing subtests of the Comprehensive Aphasia Test (Swinburn,Porter, & Howard, 2004) and on syntactic comprehension withthe Test for Reception of Grammar (TROG: Bishop, 1979). In theaphasic group, there was a significantly positive correlationbetween the connectivity strength between left and right alSTCand spoken word and sentence comprehension performances. Nomeasures of functional connectivity were correlated with perfor-mance on the TROG. PWA who showed positive functional connec-tivity between left and right alSTC performed better on spokenword and sentence comprehension than PWA whose functionalconnectivity measures were negative, but did not differ on theTROG. The study therefore did not show effects of the neuralparameters that were measured on syntactic comprehension.

Thothathiri et al. (2012) studied 79 PWA using a sentence–pic-ture matching task with five examples of each of six sentence types(actives, actives with prepositional phrases, passives, locatives,subject relative clauses, and object relative clauses). Imaging con-sisted of either CT or MR scans. For MR scans, lesions were drawnmanually onto a structural image which was normalized to MNIspace. For CT images, an experienced neurologist drew the lesiondirectly onto the normalized atlas. Voxel-based lesion symptommapping revealed a significant association between damage intemporo-parietal cortex and overall sentence comprehension,which remained significant after a measure of phonological mem-ory was used as a covariate. There was also a significant associationbetween damage in temporo-parietal cortex and comprehension ofboth sentences with canonical thematic role order (using previousterminology, these are the ‘‘baseline” sentences – active, subjectrelative) and sentences with non-canonical thematic role order(‘‘experimental” sentences – passive, object relative sentences).The VLSM analysis did not show any areas associated with the dif-ference between noncanonical and canonical sentence scores, butthe regional analysis found that the difference score was signifi-cantly correlated with damage in part of the inferior parietal lobe(BA 39).

Tyler, Wright, Randall, Marslen-Wilson, and Stamatakis (2010)studied 14 PWA and reported voxel-based lesion symptom map-ping of effects of sentence type on word monitoring times. Theypresented three types of stimuli – normal sentences, grammati-cally correct but nonsensical sentences (anomalous prose), andrandom word strings – and measured participants’ reaction timesto detect a word in each stimulus. Normal controls showed fasterRTs for target words that appeared late in normal sentences and

anomalous prose, which the authors said reflects the predictivevalue of the accruing syntactic and semantic representations,which is not available in random word strings. The authors arguedthat the purest measure of syntactic processing is the word posi-tion effect in syntactic prose, which reflects the predictive valueof syntactic structure alone. PWA showed a position effect onlyin normal sentences. Voxel-based lesion symptom mappingshowed a significant positive correlation between T1 signal integ-rity in left pars orbitalis and triangularis and the size of the posi-tion effect in anomalous prose. The authors interpret this as‘‘showing that increasing damage in left BA47/45 was associatedwith impaired syntactic processing (p. 3403).”2 There are a numberof issues that make this result hard to interpret. First, error rateswere not reported, so speed-accuracy trade-offs cannot be assessed.Second, a larger difference score could be due to many combinationsof slower and faster RTs for early and late targets. It would be usefulto see the effect of tissue integrity on RTs for early and late targetsseparately. Third, the absence of a correlation between tissue integ-rity in left BA47/45 and the size of the position effect in normal sen-tences is puzzling, because PWA did show an position effect innormal sentences (and not in anomalous prose) and because the syn-tactic processing that produces the position effect in anomalousprose contributes to the effect in normal sentences.

Tyler et al. (2011) studied the same 14 PWA as Tyler et al.(2010) and reported voxel-based lesion symptom mapping fortwo additional measures of syntactic comprehension. The firstwas accuracy in matching semantically reversible active and pas-sive sentences to pictures. Separate correlations of tissue integritywith syntactic errors in active and passive sentences showed thatmore damage in LIFG, LpMTG, LSTG and SMG was correlated withincreasing syntactic errors in passive sentences, but not active sen-tences (their Supplementary Materials, Fig. 3). The second measureof syntactic comprehension was performance on an acceptabilityjudgment task. Tyler et al. used the difference in error rates inaccepting correct ambiguous and matched unambiguous sentencesas the measure of syntactic comprehension. They found that moretissue damage in left middle temporal gyrus and left inferior fron-tal gyrus (BA 45, extending into BA 44 and BA 47) was related tosmaller differences in error rates in ambiguous compared to unam-biguous sentences. Tyler et al. based their interpretation of thisresult on the fact that error rates in controls were high (42.4%)for ambiguous sentences that were later disambiguated towardtheir less preferred interpretation. Tyler et al. argued that thesehigh error rates show that normal individuals were highly sensitiveto the preferred interpretation of the ambiguous segment, and con-cluded that smaller differences between error rates in ambiguousand unambiguous sentences in PWA indicated less normal sensi-tivity to syntactic structure and its interpretation. However, PWAdid better than controls on ambiguous sentences, which makes ithard to use their performance as evidence for a deficit in compre-hension. A more straightforward measure of syntactic comprehen-sion would simply be the error rates on each type of sentence.

Overall, the evidence regarding the effect of lesions in PWAwithchronic stroke on syntactic comprehension is limited. Limitationsarise with respect to the number of participants and/or the numberof sentence types studied and/or the number of examples of eachsentence type that was presented. Tyler et al. (2010, 2011)reported only 14 PWA; Caplan et al. (1996) reported 18.Thothathiri et al. (2012) reported 79 subjects, but only presentedfive examples of each sentence type, which is likely to miss reliablebut small differences in performance of different PWA. Caplan,Waters, and DeDe (2007) is intermediate in both regards,

t

3 The participants in this study are a subset of those reported in Caplan et al.013a, 2013b) and at meetings of several scientific societies (Academy of Aphasia,eurobiology of Language, CUNY sentence processing). The descriptions of PWA andehavioral methods are the same as in prior publications and are repeated here fore reader’s convenience.

18 D. Caplan et al. / Brain & Language 152 (2016) 14–27

presenting 10 examples of each sentence type to 42 PWA. Withtwo exceptions (Tyler et al., 2010), one of which is hard to inter-pret, the results are based on accuracy in end-of-sentence tasks,not on-line observations. No previous study has reported reliabilitystatistics.

The results of existing studies are inconsistent. Of the six stud-ies that meet criteria for identifying syntactic comprehension def-icits and use modern neuroimaging, two report null results (Caplanet al., 1996; Warren et al., 2009). With respect to positive findings,the effect of tissue damage in Tyler et al. (2011) differed from thatin Thothathiri et al. (2012), with Tyler et al. (2011) reporting effectsof lesion size in LIFG, LpMTG, LSTG and SMG on comprehension ofpassives and Thothathiri et al. (2012) reporting an effect of lesionsize only in the left inferior parietal lobe on comprehension of pas-sive and object relative sentences. Caplan, Waters, and DeDe(2007) found yet another pattern – there was no effect of lesionsize in specific ROIs on syntactic complexity scores that were sim-ilar to the measure in Thothathiri et al. (2012) in SPM, and FDG PETactivity in the insula predicted these scores in SPM. Tasks affectedresults. In Caplan, Waters, and DeDe (2007), lesion volume in Wer-nicke’s area, the inferior parietal lobe and the anterior inferior tem-poral lobe and PET activity in the inferior parietal lobe predictedsyntactic complexity scores in OM, but not in SPM. To our knowl-edge, there is no evidence for an equivalent effect of a focal lesionon syntactic comprehension performance in more than one task.Tyler et al. (2010, 2011) reported correlations of lesion size in Bro-ca’s area with performance of PWA in three tasks, but results arehard to interpret in two of the three tasks. The limited availabledata thus suggest that focal lesions affect syntactic comprehensionperformance differently as a function of task.

Four studies report syntactic comprehension following acutestroke. Three measured both infarct size using diffusion weightedimaging (DWI) and the apparent diffusion coefficient (ADC) andperfusion using perfusion weighted imaging (PWI). One definedareas of infarction using DWI imaging.

Davis et al. (2008) reported one PWAwith hypoperfusion in 85%of BA 45 and 47% of BA44, which resolved within 36 h. While theseregions were hypoperfused, the PWA performed at chance onanswering yes/no semantically reversible questions and matchingsemantically reversible active and passive sentences to videos.These deficits resolved completely after reperfusion.

Newhart et al. (2012) studied 53 PWA with strokes on their firsthospitalization day. They measured the size of infarction andhypoperfusion in Broca’s area (BA 44 and 45) and 12 other BAs(6, 10, 11, 18, 19, 20, 21, 22, 37, 38, 39, 40) on a three point scale:(1) part, (2) all, or (0) none. Comprehension was tested with bothsentence–picture matching and object manipulation, presenting10 reversible and 10 constrained sentences of each of four typesof sentences (active, passive, subject-cleft, object-cleft). Theydefined a deficit in syntactic comprehension where three criteriawere met: (1) performance was at or below chance level on passivereversible sentences on at least one test (SPM or object manipula-tion); (2) there was greater than 10 percentage points lower accu-racy on passive compared to active sentences and object-cleftcompared to subject-cleft sentences; and (3) there was greaterthan 10 percentage points lower accuracy on reversible comparedto irreversible sentences. Lesions in the angular gyrus were associ-ated with syntactic comprehension deficits by these criteria, andthere was a trend for lesions in BA45 to be associated with thesedeficits as well. However, these criteria are questionable becausethe difference between performance on passive and active sen-tences that was needed to establish a dissociation in performancewas chosen arbitrarily and is very low, and, when applied to pas-sives and actives, was apparently calculated for all sentences(reversible and constrained). Newhart et al. also compared perfor-mance of PWA with and without lesions in these ROIs. PWA with

ischemia in left BA 45 and the supplementary motor area (SMA)and were significantly more impaired than those without ischemiain these regions in comprehension of reversible passive sentencesin SPM. PWA with ischemia in left 39 were significantly moreimpaired than those without ischemia in these regions in compre-hension of passive reversible sentences in SPM and OM, and incomprehension of cleft object reversible sentences in SPM.

Race, Ochfeld, and Hillis (2012) studied 38 PWA within 24 h ofhospital admission for the ability to answer 6 yes/no constrainedand 4 yes/no reversible questions. Lesion size (infarction andhypoperfusion) was measured in nine Brodmann areas (6, 21, 22,37, 39, 40, 44, 45, 46). Their Fig. 2 shows better performance onreversible questions in PWA without ischemia than in PWA withischemia in all areas; the difference was significant in BA6, 39and 40. Lesions in BA 39 and 40 were also associated with wordcomprehension deficits.

Magnusdottir et al. (2012) studied 50 PWA within 20 days ofinfarction (identified by DWI MRI) in a sentence–picture matchingtask using five examples of each of nine sentence types. All sen-tences were semantically reversible. Three had canonical thematicrole order and three had non-canonical thematic role order. VLSMshowed that overall performance was predicted by the presence ofa lesion in a large area of the posterior left hemisphere encompass-ing the middle and superior temporal gyri, inferior parietal cortex,angular gyrus, superior to inferior occipital cortex, fusiform gyrusand middle temporal lobe. Lesions in the superior and posteriorportions of this area were more predictive of performance on thecanonical sentences, and lesions in the inferior and anterior por-tions of this area were more predictive of performance on thenon-canonical sentences. Lesions in anterior superior and middletemporal gyri and the temporal pole predicted the differencebetween performance on canonical and non-canonical sentences.

These studies in acute stroke are also limited. All data consist ofend-of-sentence accuracy measures and two studies used smallnumbers of sentences. The results are not consistent across studies.Davis et al. (2008) reported effects of hypoperfusion in BA 44/45 onanswering reversible questions, not seen in Race et al. (2012) andMagnusdottir et al. (2012) found that lesions in STG and MTG weremost predictive of disturbances of comprehension that requiredsyntactic processing. Overall, the most consistent finding in bothchronic and acute stroke is that posterior lesions and hypoperfu-sion, in inferior parietal and superior and middle temporal lobe,affect syntactic comprehension, but much remains to be learnedabout the effects of lesions on this function.

The study reported here adds additional information aboutthese topics, based on MR scanning in 31 chronic PWA who werestudied for the ability to understand eleven sentence types in threetasks, providing on-line as well as end-of-sentence measures ofperformance.

2. Methods

2.1. Participants3

Thirty-one aphasic PWA (12F, 19M; age: 60.9 years (SD: 13.1);education: years 15.3 (SD: 2.9)) were tested. PWA were requiredto be aphasic as determined by a physician or speech-languagepathologist, right handed, have a single left hemisphere stroke,be able to perform the tests, and to have adequate singleword comprehension to not fail because of lexical semantic

(2Nbth

Table 1Sentence types. Verbs in bold underline and pronouns and NPs in bold italics servedas measures of on-line performance (see text).

Sentence type Example

Active (A) The girl hugged the boyPassive (P) The boy was hugged by the girlThree Noun Phrases (3NP) The niece said that the girl hugged the boyPronoun (PRO) The niece said that the girl hugged herReflexive (REF) The niece said that the girl hugged herselfCleft Subject (CS) It was the girl who hugged the boyCleft Object (CO) It was the boy who the girl huggedSubject Object (SO) The boy who the girl hugged washed the

womanSubject Subject (SS) The girl who hugged the boy washed the

womanSubject Subject and Pronoun

(SSPRO)The woman who hugged the girl washedher

Subject Object and Pronoun(SOPRO)

The woman who the girl hugged touchedher

Subject Subject and Reflexive(SSREF)

The woman who hugged the girl washedherself

Subject Object and Reflexive(SOREF)

The woman who the girl hugged touchedherself

Experimental-baseline contrasts used to examine effects of structures were asfollows:Passive: P–A.Object Extraction: CO-CS, SO-SS, SOPRO-SSPRO, SOREF-SSREF.Pronouns: PRO-3NP, SSPRO-SS, SOPRO-SO.Reflexives: REF-3NP, SOREF-SO, SSREF-SS.

D. Caplan et al. / Brain & Language 152 (2016) 14–27 19

disturbances. PWA were screened for disturbances of phonemediscrimination, auditory lexical decision and spoken word–picturematching, and were trained on the words in the sentences, if nec-essary. Forty-six age and education matched controls (mean age:64.2 years, SD: 11.3, range: 38–87; mean education: 16.3 years,SD: 2.3, range: 8–28; M:F = 20:26) were also tested.

2.2. Materials and procedures

2.2.1. Syntactic comprehensionThe ability to parse and interpret passives, object extracted

clefts, object extracted relative clauses, and sentences with reflex-ives and pronouns was tested by having PWA respond to pairs ofsentences in which the baseline sentence did not contain the con-struction/element in question or could be interpreted on the basisof a heuristic and the experimental sentence contained the struc-ture/element and required the assignment of a complex syntacticstructure to be understood. In addition, the ability to combineoperations was tested by presenting sentences with relativeclauses and either a reflexive or a pronoun. Twenty examples ofeach sentence type were presented. Table 1 lists the sentence typesused.

For analysis of AMW data, in each sentence pair, a ‘‘criticalword” in the experimental sentence was paired with a correspond-ing word in the corresponding baseline sentence. The criticalwords in the experimental sentences were ones at which an incre-mental parser performed an operations not needed in the baselinesentence, or where the demands of an operation were greater inthe experimental than in the baseline sentence. In the experimen-tal passive and corresponding baseline active sentences, the criticalword was the main verb. In the experimental sentences with objectextracted relative clauses and corresponding baseline sentenceswith subject relative extracted clauses, the critical word was therelative clause verb. In the experimental sentences with pronounsand reflexives, the critical word was the pronoun or reflexive, andin the corresponding baseline sentences with a third noun phrasein the position of the pronoun or reflexive, the critical word wasthe third noun phrase.

Subjects were tested in object manipulation (OM) and picturematching (SPM) tasks, the latter with both whole sentence (FullSPM) and self-paced listening (auditory moving windows –AMW-SPM) presentation conditions, using digitized computer-delivered auditory stimuli. Accuracy and end-of-sentence reactiontime in Full SPM and AMW-SPM were measured. Procedures wereas in previous studies (Caplan, DeDe, & Michaud, 2006; Caplanet al., 1985, 1996; Caplan, Waters, DeDe, Michaud, et al., 2007;Caplan, Waters, & DeDe, 2007; Caplan, Michaud, & Hufford,2013a, 2013b; Caplan & Hildebrandt, 1988; Caplan & Waters,2003) and will be summarized briefly. All sentence stimuli wererecorded by one of the authors (DC) at a normal, but slow, speakingrate, and digitized using SoundEdit software. Pictures were createdas JPEG files. Auditory and visual stimuli were used to create exper-iments in Superlab, which were presented on Macintosh iBookcomputers, with auditory files presented over headphones. In allstudies, sentences were presented in a pseudo-randomized order,so that three or more examples of the same sentence type neveroccurred in succession.

2.2.1.1. Enactment (object manipulation (OM)). In the enactmenttask, participants indicated thematic roles and co-indexation bymanipulating paper dolls. PWA were told that the purpose of theexperiment was to test their abilities to understand ‘‘who did whatto whom” in the sentences. They were instructed to indicate ‘‘whodid what to whom” by acting out the sentence using the items pro-vided. The experimenter emphasized that the PWA did not need toshow details of the action of the verb, but had to clearly

demonstrate which item was accomplishing the action and whichitem was receiving it. The experimenter then proceeded with thedigitized sentences and videotaped the participant’s responses.See Caplan et al. (1985) and Caplan and Hildebrandt (1988) for afuller description of the task.

2.2.1.2. Sentence–picture matching with uninterrupted auditory pre-sentation (Full SPM). In this test, each sentence was played audito-rily with two black and white line drawings in full view of theparticipant, and the participant was required to choose the draw-ing that matched the sentence by pressing one of two keys onthe computer keyboard using fingers on the non-paretic hand.Drawings depicted the thematic roles in the sentence and an alter-nate set of thematic roles that depicted either the reverse set ofthematic roles or an incorrect antecedent of a pronoun or reflexive(Fig. 1). Correct and incorrect pictures were displayed equally fre-quently on the right and left side of the computer screen; order ofpresentation on each side was randomized within sentence type.Responses were scored for accuracy and reaction time (RT).

2.2.1.3. Sentence–picture matching with auditory self paced (auditorymoving window) presentation (AMW SPM). The method was basedon Ferreira, Anes, et al. (1996), Ferreira, Henderson, et al. (1996)and identical to that used in Caplan and Waters (2003) andCaplan, Waters, DeDe, Michaud, et al. (2007). The files were editedby placing a ‘‘tag” in the waveform at each word boundary and sav-ing segments (tag to tag portions of the waveform) as individualaudio files. The participant’s task was to pace his/her way throughthe sentence as quickly as possible, by pressing a computer key forthe successive presentation of each segment, and choose the cor-rect picture, as in Full SPM. If a participant pressed the buttonbefore the end of a segment, the segment was truncated at thepoint of the button press in order to discourage participants frompressing the button before they had heard and processed each seg-ment. Inter-response times between successive button presses, aswell as response time and accuracy on the SPM task, were

Fig. 1. Sample stimulus.

20 D. Caplan et al. / Brain & Language 152 (2016) 14–27

recorded. Word duration was subtracted from button responsetime to yield ‘‘corrected self paced listening times” to adjustedfor word length (for details, see Caplan et al., 2013b).

2.3. Short term working memory (ST-WM)

Ten tests of ST-WM were administered. Six were simple tests ofimmediate serial recall: Digit Span, Word Span; Span for Phonolog-ically Similar Words; Span for Phonologically Dissimilar Words;Span for Long Words; Span for Short Words. The contrast betweenperformance on Span for Phonologically Similar and DissimilarWords is a measure of the Phonological Similarity Effect, an indica-tion of the preservation of the Phonological Store in Baddeley’smodel. The contrast between performance on Span for Long andShort Words is a measure of the Word Length Effect, an indicationof the preservation of the rehearsal mechanism in Baddeley’smodel. Four tests required both retention and manipulation ofitems: Alphabet Span, Backwards Digit Span, Subtract-2 Span,and Sentence Span. These tests were considered to assess the Cen-tral Executive component of ST-WM, which has been related tosyntactic processing (Just & Carpenter, 1992, but see Caplan &Waters, 1999). We here report the results after performance onST-WM was regressed from measures of syntactic comprehension(see below). Because Sentence Span requires sentence comprehen-sion, which overlaps with the syntactic abilities being studied, itwas dropped from the measure of ST-WM. Materials and methodsfor the remaining three tests are described here (see Caplan et al.,2013b, for description of all tests).

Materials for Backwards Digit Span and Subtract-2 Span con-sisted of the digits 0 – 9. In Alphabet Span, they consisted of twosyllable middle frequency words. Participants listened to lists ofwords or digits presented over headphones at a rate of one stimu-lus per second. In Backwards Span, they reported the digits in thelist in the reverse order of presentation. In Subtract-2 Span, theyreported the result of subtracting two from each digit, in the orderin which the digits were presented. In Alphabet Span, theyreported the words in the stimulus list in alphabetical order. InSentence Span (Waters & Caplan, 1996), they listened to a seriesof sentences in a simple syntactic form (cleft subject) and indicatedwhether the sentence was acceptable or not. After seeing all sen-tences in a set they reported the final words of all of the sentencesin the set in the order they had occurred. RT and accuracy on thesentence processing component of the task as well as word recallwere measured.

Because of speech production disorders in some aphasic partic-ipants, participants made pointing responses in all tasks. InSubtract-2 Span and Backwards Span, the digits 0 through 9 weredisplayed on a computer screen. In Alphabet Span, the stimuluswords and two additional words drawn from the same pool ofwords were displayed on a computer screen.

For both control participants and PWA, each task began with apractice session consisting of three trials at Span 2. Following thepractice session, for controls, the task began at list length 2 andcontinued to list length 8, regardless of performance on lowerspans. For PWA, the task began at list length 2 and continued to listlength 5, regardless of performance on lower spans. The lower limitwas selected because many PWA performed at very low levels atlist lengths 4 or 5 on several tasks, leading to concerns about attri-tion due to frustration, with potentially significant loss of partici-pants in all aspects of the study. For PWA who performedcorrectly on at least 3/5 trials at List Length 5, 6 or 7 correct, testingcontinued to the next list length.

Performance was scored for the number of items recalled in thecorrect serial position. Each participant’s span score was deter-mined to the highest list length at which s/he achieved at least3/5 trials correct (regardless of performance on lower spans), plus0.5 to the score if s/he achieved 2/5 trials on the next list lengthafter the last one in which s/he achieved at least 3/5 correct. Thetotal number of items each subject reported correctly in the correctorder was also calculated. The number of correct responses in thelast list length tested in PWA who were not tested to list length 8was always less than 2, indicating that these measures of item andorder recall did not significantly underestimate the performancesof PWA who were not tested at longer list lengths.

2.4. Neuroimaging

Participants underwent a single MRI scanning session thatincluded T1-weighted-MRI and DT-MRI. MRI scanning was per-formed on a Siemens 3.0T Tim Trio system (Siemens, Erlangen,Germany), equipped with 32-channel head coil. Head movementwas restricted using expandable foam cushions. An automatedscout image was acquired and shimming procedures conductedto optimize field homogeneity that was then utilized for subse-quent slice positioning (van der Kouwe et al., 2005). Automaticalignment was done at acquisition time using ‘‘autoalign” (vander Kouwe et al., 2005). After scanning, each MR image data setwas saved and transferred to CD and maintained in duplicate copy.A high resolution 3D Multi-Echo (ME) MPRAGE sagittal sequence,which minimizes signal from the dura in the pial surface (vander Kouwe, Benner, Salat, & Fischl, 2008), was collected with anisotropic resolution of 1 mm3 (TR = 2.53 s, TI = 1.2 s,TE = 1.64/3.5/5.36/7.22 ms, bandwidth = 651 Hz/px, flip angle = 7�,FOV = 256 mm, 256 � 256 matrix, 176 slices, slice thick-ness = 1 mm). Data was checked by visual inspection for artifactssuch as ghosting and blurring principally due to motion.

Automated computational reconstruction of brain surface, cor-tical thickness maps, and segmentation of cortical and subcorticalstructures were undertaken using established protocols (Fischlet al., 2004; Salat et al., 2004) and followed previous work(Caplan, Waters, & DeDe, 2007). The T1-weighted ME-MPRAGEimages from each subject were motion-corrected, averaged, andnormalized for intensity using algorithms in the FreeSurfer soft-ware package (http://www.martinos.org/freesurfer) (Fischl &Dale, 2000; Fischl, Sereno, & Dale, 1999; Fischl et al., 2004). Inten-sity variations due to magnetic field inhomogeneities were cor-rected, normalized intensity images created, and the skullremoved from all normalized images using a skull-stripping algo-rithm (Ségonne et al., 2004). Segmentation of cortex was doneusing algorithms described in prior publications (Dale, Fischl, &Sereno, 1999; Fischl & Dale, 2000; Fischl et al., 1999). Minimalmanual editing was performed as necessary, with standard, objec-tive editing rules. High correlations between automated and man-ual methods have been demonstrated in a series of studies (Fischl& Dale, 2000; Fischl et al., 2002; Walhovd et al., 2005).

D. Caplan et al. / Brain & Language 152 (2016) 14–27 21

Cortical parcellation divided the cortex into 48 parcellationunits (”PUs”) (Desikan et al., 2006; Fischl et al., 2002;Rademacher et al., 1992). The first step in the parcellation processwas the application of a standard coordinate system to each scan.Three user-specified points were identified the anterior commis-sure (AC), posterior commissure (PC) and interhemispheric plane(mid-sagittal point, MSP). The Y axis was determined by the ante-rior commissural–posterior commissural line, the Z axis was set bythe intersection of the interhemispheric plane with the AC–PC line,and the X axis was orthogonal to the Y and Z axes. Second, sulciwere identified and labeled by a neuroanatomically trained techni-cian. The end-points of selected sulci served as limiting planes thatwere marked on multiplanar orthogonal views to permit them tobe tracked three-dimensionally. Parcellation units were then cre-ated as the intersections of sulci and limiting planes. Volumes ofparcellation units were derived by summing the voxels in eachPU. The resulting volumes of PUs are individualized to each brainand are thus not subject to distortions associated with normaliza-tion to an atlas.4

In the lesioned hemisphere, segmentation was undertaken forthe same set of gray and white matter structures as in the unle-sioned hemisphere. The lesion was also segmented. Reconstructionof lesion-ablated structures was achieved by a combination ofintensity borders, special image analysis techniques to permit visu-alization of the lesion borders, and use of correspondence betweenpoints of reference in the two hemispheres. Two types of lesionborder may be obscured in these analyses, one between necrotictissue and (apparently) intact gray or white matter, and the otherbetween necrotic tissue and CSF-containing spaces. The first wasaddressed with intensity contours, as for segmentation in theintact brain. The second type of tissue border, which arises whenthe lesion extends to the cortical surface and has totally destroyedthe cortex and underlying white matter, or extends to the ventricle,was first visualized by selective high contrast ‘‘windowing” anddecreasing the brightness to near the threshold for visual detectionof the image. A ‘‘veil” of surviving meningeal connective tissue thatdelineates approximately the cortical surface configurationbecomes apparent under these conditions. The ependyma at theventricular boundary of a lesion was established in similar fashion.The inner border of the destroyed cortical ribbon (that is, thedestroyed cortical-white matter border) was estimated from theaverage cortical thickness in the opposite hemisphere and is gener-ally about 2–3 mm in width.

The lesioned hemisphere was parcellated with the same proce-dure used in intact brains. Two classes of anatomical landmarksupon which the parcellation system is based were applied to lesion

4 The remainder of the segmented hemisphere, including central white matterhippocampus and subcortical gray matter structures, was parcellated entirelycomputationally from the landmarks specified in the neocortical parcellation stepplus four additional sub-cortical points specified by the user. The diencephalon wasdivided into two broad regions, the thalamus–epithalamus and hypothalamus–subhalamus, by an X–Y plane passing through the AC–PC line. The dorsal thalamus wassubdivided into anterior, medial, lateral and posterior subdivisions; the caudate intohead, body and tail; the putamen into anterior and posterior subdivisions; and theamygdala was separated from its apparent continuity with the hippocampaformation. The internal–external segments of the globus pallidus and the remainderof the diencephalon were not subdivided. The central white matter was divided intofour concentric radial domains: (1) the corona radiata, subjacent to the neocortex; (2the external sagittal stratum (which includes ipsilateral longitudinal fiber bundles)(3) the interhemispheric fibers of the corpus callosum; and (4) the internal capsuleand the fornix–fimbria fibers. The corona radiata was parcellated by proximity intounits corresponding to the overlying neocortical parcellation units. The externasagittal stratum was parcellated into callosal fibers, cingulate bundle, and superiorinferior, and temporal strata containing longitudinal intrahemispheric fibers. Thecallosal fibers, cingulate bundle, and inferior and superior longitudinal fibers wereparcellated into six divisions along their antero-posterior course. The corpus callosumwas subdivided into splenium, body (anterior and posterior segments) and genu. Wedid not analyze these PUs in this study.

,

t

l

);

l,

parcellation – a set of points along the Y axis that specifies the loca-tion of X–Z co-ordinate planes, such as the tip of the genu and sple-nium of the corpus callosum, and points that specify locationsdefined by all three of the coordinate axes, such as the intersec-tions of precentral, central, or postcentral fissures with the inter-hemispheric or sylvian fissures. With few exceptions, the meancoordinate positions of these points for all planes (X, Y, and Z axes)are minimally variant (<2–3 mm) between the two hemispheres,allowing their locations in the intact hemisphere of a lesionedbrain to determine their approximate locations in the lesionedhemisphere. The application of segmentation and parcellationdirectly to images reduces mischaracterization of areas that aredue to the effect of distortions of structures seen in images afterstroke in normalization to templates.

The combined segmentation and parcellation proceduresresulted in a total of 48 cortical parcellation units, associated whitematter units (corona radiata), 27 deep white matter units, and 15subcortical gray matter regions in each hemisphere. Each of theseunits was segmented in two dimensions, as described above, butinvolved three dimensions, due to MR slice thickness. For eachbrain, the volume of each parcellation unit was thus expressed asthe number of voxels assigned to that structure. Lesion volumein each parcellation unit was expressed as the number of voxelsin the lesion in that unit. The extent of damage to each parcellationunit was expressed as the percent of each reconstructed parcella-tion unit that was occupied by a lesion. The percent of left hemi-sphere cortex occupied by a lesion were also calculated. Theresulting volumes of lesions are expressed as percentages of indi-vidual PWA’s PUs, and, like the volumes of PUs themselves, arenot subject to distortions associated with normalization.

3. Results

3.1. Behavioral results

Performance on the comprehension and short term/workingmemory tests is summarized for the 31 PWA reported here inTables 2–4. The results are highly reliable, with Pearson’s r andthe Spearman Brown reliability co-efficients for each sentence typein each task for the PWA all significant (all ps < .001).

The sentences were grouped to reduce the number of analyses.Four sets of sentence types were considered, grouping togethersentences with common linguistic elements:

1. Passive (P) vs Active (A).2. Relative Clauses: Object extraction (CO, SO, SOREF and SOPRO)

vs Subject extraction (CS, SS, SSREF and SSPRO).3. Pronouns (PRO, SOPRO, SSPRO) vs sentences with referential

NPs (3NP, SO, SS).4. Reflexives (REF, SO REF, SS REF) vs sentences with referential

NPs (3NP, SO, SS).

We investigated whether age and time since stroke affectedsyntactic comprehension, by correlating the behavioral measureswith age and time since stroke in the entire group of PWA and inthe 31 PWA who were scanned. None of these correlations weresignificant and these factors were dropped from further analyses.

3.2. Neurological results

We analyzed the effects of lesions on performance in parcella-tion units in which eight or more PWA had lesions. There were21 such PUs, listed in Table 5. Lesions were more frequent in areassupplied by the middle cerebral artery. The four areas that weremost frequently lesioned were the pre- and post-central gyri and

Table 2Performance (accuracy) of people with aphasia on sentence comprehension tests.

A P 3NP PRO REF CS CO SS SO SSPRO SOPRO SSREF SOREF

Sentence-picture matching with auditory moving window presentationScan PWA mean 82.2 77.2 85 73.4* 86.6 83 74.8* 77.6 66.6* 74.6 63.4* 74.4 65.8*

Scan PWA SD 18.4 22.2 13 21.4 21.4 19 23.6 17.4 20.8 22 21.6 20.2 19.8EC mean 94.8 94 97.8 93 98.6 95.8 94.4 93.4 91 96.6 88.4 95.8 83.6EC SD 6.2 8.4 5.8 4.4 3.8 6.8 6.2 9.8 8.8 6.8 14.8 7.8 13.2

Sentence-picture matching with full sentence presentationScan PWA mean 82.8 77* 81 73.6* 78.8 81.6 70.8* 70 67* 71 59.6*,+ 69.6 62*,+

Scan PWA SD 15.8 18.8 17.6 23.6 24.8 19.8 20.6 16 15.6 20 20.2 20.2 17.4EC mean 95.2 93.6 97 96.4 97.2 95 92.6 92.2 82.4 95.2 82.4 94.2 83EC SD 3.6 6.4 5.4 5.6 4.8 4.6 6.6 8 13.4 6 14.2 8.8 12.8

Object manipulationScan PWA mean 88.6 76* 69.8 70 69.2 87.2 73.6* 62.6 40.4* 57.4 39.8* 63.4 40.6*

Scan PWA SD 16.4 29.2 35.6 34.8 37.6 19.2 28.8 39 37.8 42.6 32 39.2 37.4EC mean 100 98.8 97.2 97.4 95.4 100 98.8 98.8 84.4 92 83.6 96.6 87.8EC SD 0 7.4 15 13.4 17.8 0 6 2.8 21 25.4 23.8 14.4 22

Sentence type labels as in Table 1.* Indicates a significant difference between the experimental sentence annotated and its baseline (e.g., PRO and 3NP in sentence picture matching with auditory moving

window presentation.+ Indicates a significant difference between SOPRO and SO and between SOREF and SO sentences.

Table 3Performance (RT) of people with aphasia on sentence comprehension tests.

A P 3NP PRO REF CS CO SS SO SSPRO SOPRO SSREF SOREF

Sentence-picture matching with auditory moving window presentationScan PWA mean 1838 2084 1845 1947 1523 1765 2151 2300 2673 2255 2560 1809 2062Scan PWA SD 1311 1504 1230 1421 965 1531 1643 1695 1963 1395 1782 1510 1605EC mean 623 867 846 866 641 669 806 768 994 886 1284 708 947EC SD 339 437 496 589 473 440 588 510 668 570 608 485 627

Sentence-picture matching with full sentence presentationScan PWA mean 4896 5339 6377 6138 5544 5621 6463⁄ 7252 7784 7419 7886 7061 7750Scan PWA SD 1650 1741 1705 1691 1292 1628 1869 1977 2085 2029 1843 1767 2036EC mean 3284 3591 4502 4465 4154 3990 4482 4831 4986 4795 5243 4565 4864EC SD 801 779 833 827 733 698 753 863 737 875 720 881 780

Sentence type labels as in Table 1

Table 4Performance on ST-WM tests.

Task Mean SD Minimum Maximum Maximum

ControlsWord 85.13 23.85 41 158 158Digits F 131.48 35.44 64 175 175WORD DIS 79.45 29.22 38 161 161WORD SIM 51.23 22.96 22 118 118WORD SHORT 74.88 17.62 41 121 121WORD LONG 71.55 22.23 38 124 124Digits B 85.43 39.54 25 167 167Sent 35.33 20.69 8 96 96Alpha 88.78 35.98 39 159 159Sub2 74.43 36.09 25 168 168

PWAWord 52.23 30.22 0 125 125Digits F 55.46 47.54 0 173 173WORD DIS 44.52 35.16 0 131 131WORD SIM 27.65 32.31 0 136 136WORD SHORT 40.06 31.25 0 126 126WORD LONG 43.52 34.39 0 160 160Digits B 30.77 37.53 0 166 166Sent 10.69 16.61 0 68 68Alpha 38.31 30.52 0 123 123Sub2 40.65 37.24 0 171 171

22 D. Caplan et al. / Brain & Language 152 (2016) 14–27

their opercular cortex, and the insula. All areas in perisylviancortex associated with language were ones in which eight or morePWA had lesions.

We used regression analyses to determine the effect of lesionsize in parcellation units on syntactic comprehension. In all analy-ses, the dependent variable was a measure of performance for theexperimental sentences in a sentence type group – accuracy, RT torespond in sentence picture matching, and corrected self-paced lis-tening times for critical words. To measure lesion effects on syn-tactic comprehension independent of short-term/workingmemory and the ability to assign propositional meaning throughthe use of heuristics, we used forward regression analyses in whichwe first entered measures of short term/working memory, and per-formance on baseline sentences. The measure of short term/work-ing memory was the total number of items reported in the correctserial positions in all STM/WM tasks. The measure of performanceon the baseline sentences corresponded to the measure of syntac-tic comprehension – accuracy, response RT, and corrected selfpaced listening times on words in baseline sentences that corre-sponded to critical words in experimental sentences. To reducethe possibility that a significant effect lesion size in a parcellationunit was due to effects in other parcellation units that were alsoinfarcted, we also entered total lesion volume (percent of lefthemisphere cortex) that was lesioned in the regressions. Afterthese three variables were entered, we then entered percent lesionin each parcellation unit. The effect of percent lesion in each par-cellation unit on performance on the experimental sentence mea-sure after the effects of the other three variables are removedreflects the effect of lesion size in that parcellation unit on the abil-ity to assign and interpret the syntactic structure while performinga comprehension task. We report all effects that reached the level

Table 5Distribution of lesions.

Parcellation unit Number of PWA with lesion Mean lesion percent SD Minimum Maximum

Lesion distributionOLs Occipital lateral gyri, superior division 8 8.5575 10.43732 0.23 27.5PHa Parahippocampal gyrus, anterior division 8 2.935 3.0357 0.04 8.73AG Angular gyrus 10 34.927 21.5499 4.34 76.95FP Frontal pole 10 5.979 6.4945 0.05 20.77T1p Superior temporal gyrus, posterior division 10 46.428 41.15907 1.23 100BFsbcmp Basal forebrain 12 13.45917 12.17854 0.03 43.56F3t Inferior frontal gyrus, pars triangularis 12 43.3125 35.91228 0.2 100H1 Heschl’s gyrus 12 75.41583 35.05663 4.53 100PP Planum polare 12 68.08833 36.26746 2.67 100SGa Supramarginal gyrus, anterior division 12 59.42333 42.43712 1.03 100F2 Middle frontal gyrus 13 20.14 26.14542 0.02 88.95FOC Fronto-orbital cortex 13 23.85308 24.68038 0.02 74.47PO Parietal operculum 13 75.29615 37.6085 3.15 100PT Planum temporale 13 68.01538 33.55446 2 100SGp Supramarginal gyrus, posterior division 15 31.03267 28.48812 0.22 84.51F3o Inferior frontal gyrus, pars opercularis 16 41.76813 39.29025 3.51 100FO Frontal operculum 16 66.06813 34.60829 0.16 100POG Postcentral gyrus 17 15.19176 16.56305 0.24 48.38CO Central operculum 21 57.03619 39.20564 0.05 100INS Insula 21 51.64381 41.17554 0.02 100PRG Precentral gyrus 23 8.9113 16.65685 0.01 69.6

D. Caplan et al. / Brain & Language 152 (2016) 14–27 23

of a trend. Bonferroni corrected p values can be obtained by multi-plying the p values in Table 6 by 3.

There were no significant effects of lesion size in any parcella-tion units on RTs to respond in the two tasks that involved sen-tence picture matching.

There were three significant effects of lesion volume in particu-lar parcellation units on corrected self paced listening times forcritical words in experimental sentences. Greater percent lesionin the posterior portion of the supramarginal gyrus led to longerself paced listening times for critical words in sentences withreflexives. In two cases, greater percent lesion led to shorter selfpaced listening times for critical words. This was the case forlesions in the middle frontal gyrus and on-line processing in sen-tences with reflexives, and for lesions in the orbital frontal cortexand these measures of on-line processing in sentences withpronouns.

Effects of lesion volume in PUs on accuracy on a sentence typeare listed in Table 6. There were eleven effects of lesion volume inparcellation units on accuracy for experimental sentences thatwere significant or at the level of a trend. Table 6 also lists theunique variance associated with lesion size on accuracy on the sen-tence type in difference tasks – the squared semi-partial correla-tion (part correlation) of the percent lesion in each parcellationunit – and the difference in part correlations for the PU on that sen-tence type in different tasks. Cohen, Cohen, West, and Aiken (2003)suggest that differences in unique variance of 2% should be consid-ered weak, 9% moderate and 25% strong.

Larger lesions in the posterior superior temporal gyrus wereassociated with lower accuracy in sentences with passives inobject manipulation. The difference between the unique variancein passives associated with lesion size in this PU in object manip-ulation and the unique variance in passives in sentence picturematching with either full or self paced presentation was large.

Larger lesions in the angular gyrus were associated with loweraccuracy in sentences with object extracted relative clauses in sen-tence picture matching with full sentence presentation. The differ-ence between the unique variance in object extracted relativeclauses associated with lesion size in this PU in sentence picturematching with full presentation and the unique variance in objectmanipulation was moderate.

Larger lesions in the angular gyrus were also associated withlower accuracy in sentences with pronouns in sentence picture

matching with auditory moving windows presentation. The differ-ences between the unique variance in pronouns in sentence pic-ture matching with auditory moving windows presentationassociated with lesion size in this PU and the unique variance inpronouns in either sentence picture matching with full presenta-tion or in object manipulation were moderate.

Larger lesions in the pars opercularis were associated withlower accuracy in sentences with pronouns in sentence picturematching with full sentence presentation. Larger lesions in the parsopercularis were also associated with lower accuracy in sentenceswith reflexives in sentence picture matching with full sentencepresentation and in object manipulation.

Larger lesions in the pars triangularis were associated withlower accuracy in sentences with reflexives in object manipulation.

Larger lesions in the frontal pole were associated with loweraccuracy in sentences with reflexives in object manipulation.

Larger lesions in the basal forebrain were associated with loweraccuracy in sentences with reflexives in sentence picture matchingwith auditory moving windows presentation.

Larger lesions in the posterior portion of the supramarginalgyrus were associated with higher accuracy in sentences withreflexives in sentence picture matching with full sentence presen-tation and with auditory moving windows presentation. The differ-ence between the unique variance sentences with reflexivesassociated with lesion size in this PU in sentence picture matchingwith full sentence presentation and the unique variance in sen-tences with reflexives in object manipulation was moderate.

4. Discussion

This study addresses some of the concerns regarding the data-base of deficit-lesion studies of syntactic comprehension raisedin the introduction to this paper. The range of paring/interpretiveoperations assessed is wide and is tested in depth – the study pre-sented four different types of syntactic operations, three of whichusing several sentence types containing the structure or requiringthe operation, twenty sentences of each type, and two tasks, one ofwhich had two presentation conditions. Accuracy, RT, and self-paced listening times were collected. The analyses focused on theability to construct and interpret particular syntactic structuresafter taking into account PWA’s ability to understand words, to

Table 6Significant effects of percent lesion in parcellation units on accuracy on experimental sentences. The table shows the results of forwards regressions in which the dependentvariable is accuracy on experimental sentence and the in dependent variables are total WM span, accuracy on baseline sentence, percent lesion in the LH cortex, and percentlesion in each parcellation unit (entered last).

24 D. Caplan et al. / Brain & Language 152 (2016) 14–27

understand matched baseline sentences that did not contain theexperimental structures, STM/WM, and the effect of total lesionsize.

A limitation of this study was that, although the number of PWAstudied was larger than in many studies, it was relatively small.There is a trade-off between the extent to which an individualPWA can be tested and the number of PWA that can be tested ina study; this study increased depth and breadth of testing at theexpense of greater sample size. Because of the relatively smallsample size, we consider all results here to be suggestive, as isthe case, in our view, for all studies of this subject in the literature.Our reporting of results that reach the statistical level of a trend isconsistent with the view that results here are suggestive. We notethat the most important results are reliable at conventional levels,including those with Bonferroni correction.

We begin by discussing two results that appear to be aberrant –the findings that larger lesions in two areas (middle frontal gyrusand orbital frontal cortex) were associated with shorter self pacedlistening times and that larger lesions in the supramarginal gyruswere associated with higher accuracy in sentences with reflexivesin both sentence picture matching tasks.

The finding that larger lesions in some PUs were associated withfaster self paced listening times seems paradoxical but can poten-tially be understood by considering the distribution of listeningtimes in PWA and controls. Using the same regression as herewithout the neurological variables, Caplan et al. (in preparation)found that some PWA had residual self paced listening times forcritical words in experimental sentences that were very shortand outside the normal range. In most cases, these very short

residual self paced listening times were associated with higherthan normal rates of errors. This suggests that both very longand very short duration of on-line processing can be pathological.Very long duration of on-line processing is usually taken an indica-tion that some incremental process is inefficient in a PWA (thesame interpretation as is made in neurologically normal individu-als). The inefficient process could be parsing, interpretation and/orperforming a comprehension task. Very short duration of on-lineprocessing that is associated with low comprehension perfor-mance has not been studied. Caplan et al. (in preparation) sug-gested that it results from abnormal functioning of the controlsystem, which must decide how much time to allow incrementalprocesses to operate before requesting the next input. Longer timesallocated to processing allow for more complete analysis butincrease the time material must be maintained in memory. Veryshort durations of on-line processing that are associated withlow comprehension performance could result from several abnor-mal functions of the executive control system. The control systemmight set abnormally low criteria for success in the task, leading toa speed-accuracy trade-off that increases speed of examination ofthe current word at the expense of final accuracy in comprehen-sion. There may be limitations in the ability to maintain task goals– a limitation of ‘‘exective attention.” Other control abnormalitiescould be imagined. The finding that lesion size in middle frontalgyrus and orbital frontal cortex was associated with faster selfpaced listening times suggests that these regions are involved inthese control processes. The fact that the negative relationshipbetween lesion size and self paced listening times was only foundfor one sentence type suggests that control processes operate

D. Caplan et al. / Brain & Language 152 (2016) 14–27 25

differently for different structures. Without obtaining data onincremental processing in a second task, it is impossible to knowif they also operate differently in different tasks.

The second finding that is hard to explain is that larger lesionsin the posterior supramarginal gyrus were associated with higheraccuracy in sentences with reflexives. One possible explanationfor this finding could be that this region is responsible for addingentities into the discourse representation, an operation that isrequired for the third NP in the baseline sentences but not in theexperimental sentences (which contain reflexives). However, if thiswere the case one would expect lesion size in this area to be neg-atively correlated with accuracy on the baseline sentences, whichwas not found to be the case. We therefore cannot account for thisfinding. We can only offer the speculation that one effect of activityin this region is to inhibit activity in another that is involved incomprehension of these sentences.

Turning to results that comport with expectations, the findingthat larger lesions in the posterior portion of the supramarginalgyrus were associated with longer self-paced listening times forreflexives is one of the few instances of an effect of lesion on ameasure of on-line syntactic processing. It points to a role for thisarea in processing reflexives in this task. As reviewed above, docu-mentation of relations between on-line measures of syntactic com-prehension and lesions have been hard to document. This is likelypartly due to the paucity of studies that have sought these relation-ships and the limited power of studies that have. This study alsohas limited power because of the relatively small number ofPWA that were studied. Studies with greater power may revealmore such relations.

The finding that larger lesions in some areas resulted in loweraccuracy on experimental sentences (relative to baseline sen-tences) was also expected. This study replicated previous findingsthat lesions in Broca’s area (pars opercularis and triangularis of theinferior frontal gyrus) andWernicke’s area (the posterior half of thesuperior temporal gyrus) were associated with syntactic compre-hension disorders. There were also effects of lesions outside theclassic language areas – in the frontal pole and the basal forebrain– on syntactic comprehension.

The effect of lesion location differed for different syntacticstructures. Larger lesions in pSTG were associated with lower accu-racy on passives, larger lesions in the angular gyrus were associ-ated with lower accuracy on object extracted relative clauses,and larger lesions in the supramarginal gyrus, pars opercularisand triangularis, the frontal pole, and the basal forebrain wereassociated with lower accuracy on sentences with pronouns orreflexives. This suggests that these regions support the construc-tion and/or interpretation of particular syntactic operations.

In no case was the effect of lesion size in a PU on accuracy for asentence type found in all tasks. In two cases – the effect of lesionsin the angular gyrus on object relatives (found in sentence picturematching with full sentence presentation) and on sentences withpronouns (found in sentence picture matching with auditory mov-ing windows presentation) – the difference between the uniquevariance associated with the lesion in the task in which the lesioneffect on the sentence type was significant and at least one task inwhich the lesion effect on the sentence type was not significantwas moderate. In one case – the effect of lesions in the posteriorsuperior temporal lobe on passives (found in object manipulation)– the effect of the lesion was significant (not just a trend) and thedifference between the unique variance associated with the lesionin the task in which the lesion effect on the sentence type was sig-nificant and a task in which the lesion effect on the sentence typewas not significant was strong.

These results indicate that the effects of lesions on the ability todemonstrate that a PWA has assigned and interpreted particularsyntactic structures differ in different tasks. The difference

between tasks was not an artifact of properties of performanceor the lesions. Table 2 shows that performance was well belowceiling and had large variance on all sentence types in all tasks.Similarly, the fact that there were significant effects of lesion sizein these areas and the comprehension of a structure in one taskindicates that the lack of effects of lesion size in these areas andcomprehension of a structure in another task is not due to limitedvariance in lesion size in those areas. The function that becomesincreasingly deficient as lesions in these areas become larger is alsonot simply the ability to evaluate the congruence between the the-matic roles depicted in pictures and those in a spoken sentence, orto enact thematic roles, because those abilities are required in allthe sentences presented in the SPM and OM tasks.

The selective relation of lesions in different areas to sentenceswith different syntactic elements and structures in different tasksthus indicates that the effect of a lesion in those areas is not toaffect parsing and/or interpretive operations themselves. Rather,the effect of a lesion in these areas is on the combination of theparsing and interpretive operations in that sentence type and theuse of the information derived from those operations to performa task. If the effect of a lesion in an area was on parsing and/orinterpretive operations themselves, lesion size in that area wouldcorrelate negatively with accuracy in the affected sentences in alltasks.

The finding that larger lesions in these areas led to lower accu-racy but not to longer self paced listening times for critical wordsin these sentences delineates a temporal window within whichthe operations that are affected by lesions in these areas occur.They must occur after a PWA presses the response key to hearthe next word in a sentence and before s/he selects a responsefor the task. The fact that this window follows initial incrementalprocessing is consistent with the view that lesions in these areasaffect operations that select responses on the basis of both themeaning of a sentence and its syntactic elements and structure.

The view that brain areas support the combination of compre-hension and task-related operations is consistent with a ‘‘situated”model of cognitive and language, that maintains that knowledge isattained through activity associated with social, cultural and phys-ical contexts and is represented in relation to such activity. In thearea of syntactic comprehension, situated language processing hasreceived support from results such as finding that the nature ofitems in an array affects the on-line interpretation of syntacticallyambiguous sentences (Farmer, Anderson, & Spivey, 2007;Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy, 1995). Most the-ories assume that situated processing results from the integrationof multiple brain areas. However, it could also be supported byfunctional specialization for particular combinations of cognitiveand task-related operations, which could arise as a result of thedevelopment of a skill. The repeated combination of parsing andinterpretive operations with task-related operations could lead tothe development of skilled operations in different domains, suchas extracting the meaning of sentences from sound and mappingthem onto the visual world, which could be supported by particu-lar brain areas, at least in part. A potentially interesting finding isthat the regions in which there is the strongest evidence for task-specific, structure-specific effects of lesions (significant effects oflesion size on one structure in one task and moderate to large dif-ferences in unique variance in that structure accounted for bylesion size in different tasks) are all posterior, involving the poste-rior half of STG and the adjacent AG. This requires further study.

Regional specialization for combinations of operations does notpreclude the brain supporting skilled integrated performance inother ways, such as through the interaction of many brain regionsor novel sets of brain regions being activated in novel ways ‘‘on thefly” under some circumstances (e.g., to accomplish new tasks). Itonly maintains that one aspect of brain organization is that

26 D. Caplan et al. / Brain & Language 152 (2016) 14–27

operations that are frequently performed together can come to uti-lize and eventually to require particular brain areas.

To summarize, the major result of this study is the finding thatthe size of lesions in small areas of the left hemisphere were neg-atively related to accuracy in particular operations required forsyntactic comprehension in particular tasks. This suggests thatthese areas are specialized for these operations in these tasks.The view of cortical specialization suggested here has not beenproposed in other areas of cognition, to our knowledge. If itreceives additional support from other studies in the area of syn-tactic comprehension, consideration of task effects on neural corre-lates of other cognitive functions would become an interestingtopic to explore. The view of cortical specialization suggested hereraises the question of what combinations of operations can be co-localized this way, and what the conditions are that produce suchco-localization.

Acknowledgment

This research was supported by NIDCD grant DC00942 to DavidCaplan.

References

Amici, S., Brambati, S. M., Wilkins, D. P., Ogar, J., Dronkers, N. L., Miller, B. L., et al.(2007). Anatomical correlates of sentence comprehension and verbal workingmemory in neurodegenerative disease. The Journal of Neuroscience: The OfficialJournal of the Society for Neuroscience, 27(23), 6282–6290.

Balogh, J., Zurif, E., Prather, P., Swinney, D., & Finkel, L. (1998). Gap-filling and end-of-sentence effects in real-time language processing: Implications for modelingsentence comprehension in aphasia. Brain and Language, 61, 169–182.

Bishop, D. V. M. (1979). Test for reception of grammar (2nd ed.). Age and CognitivePerformance Research Centre, University of Manchester.

Burkhardt, P., Piñango, M. M., & Wong, K. (2003). The role of the anterior lefthemisphere in real-time sentence comprehension: Evidence from splitintransitivity. Brain and Language, 86, 9–22.

Caplan, D. (2015). Disturbances of Syntactic Comprehension. Presented atUniversity of Copenhagen. Workshop on Morphology and Syntax.

Caplan, D., Baker, C., & Dehaut, F. (1985). Syntactic determinants of sentencecomprehension in aphasia. Cognition, 21, 117–175.

Caplan, D., DeDe, G., & Michaud, J. (2006). Task-independent and task-specificsyntactic deficits in aphasic comprehension. Aphasiology, 20, 893–920.

Caplan, D., & Gould, J. (2008). Language and speech. In M. Zigmond, F. E. Bloom, S. C.Landis, J. L. Roberts, & L. R. Squire (Eds.), Fundamental neuroscience (3rd ed.. NewYork, NY: Academic Press.

Caplan, D., & Hildebrandt, N. (1988). Disorders of syntactic comprehension.Cambridge, MA: M.I.T. Press (Bradford Books).

Caplan, D., Hildebrandt, N., & Makris, N. (1996). Location of lesions in strokepatients with deficits in syntactic processing in sentence comprehension. Brain,119, 933–949.

Caplan, D., Michaud, J., & Hufford, R. (2013a). Dissociations and associations ofperformance in syntactic comprehension in aphasia and their implications forthe nature of aphasic deficits. Brain and Language, 27(1), 21–33.

Caplan, D., Michaud, J., & Hufford, R. (2013b). Short-term memory, workingmemory, and syntactic comprehension in aphasia. Cognitive Neuropsychology,30(2), 77–109.

Caplan, D., & Waters, G. S. (1999). Verbal working memory and sentencecomprehension. Behavioral and Brain Sciences, 22(1), 77–94.

Caplan, D., & Waters, G. S. (2003). On-line syntactic processing in aphasia: Studieswith auditory moving windows presentation. Brain and Language, 84(2),222–249.

Caplan, D., Waters, G. S., & DeDe, G. (2007). Specialized verbal working memoryfor language comprehension. In A. Conway, C. Jarrold, M. Kane, A. Miyake, &J. Towse (Eds.), Variation in working memory. Oxford, UK: Oxford UniversityPress.

Caplan, D., Waters, G. S., DeDe, G., Michaud, J., & Reddy, A. (2007). A study ofsyntactic processing in aphasia I: Behavioral (psycholinguistic) aspects. Brainand Language, 101, 103–150.

Caramazza, A., & Zurif, E. B. (1976). Dissociation of algorithmic and heuristicprocesses in language comprehension: Evidence from aphasia. Brain andLanguage, 3, 572–582.

Chomsky, N. (1986). Knowledge of language. New York, NY: Praeger.Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press.Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple

regression/correlation analysis for the behavioral sciences (3rd ed.). NJ: Erlbaum.Culicover, P., & Jackendoff, R. (2005). Simpler syntax. Cambridge, MA: MIT Press.Dale, A. M., Fischl, B., & Sereno, M. I. (1999). Cortical surface-based analysis. I.

Segmentation and surface reconstruction. NeuroImage, 9(2), 179–194.

Davis, C., Kleinman, J. T., Newhart, M., Gingis, L., Pawlak, M., & Hillis, A. E. (2008).Speech and language functions that require a functioning Broca’s Area. Brainand Language, 105, 50–58.

Desikan, R. S., Ségonne, F., Fischl, B., Quinn, B. T., Dickerson, B. C., Blacker, D., et al.(2006). An automated labeling system for subdividing the human cerebralcortex on MRI scans into gyral based regions of interest. NeuroImage, 31(3),968–980.

Dick, F., Bates, E., Wulfeck, B., Utman, J. A., Dronkers, N., & Gernsbacher, M. A. (2001).Language deficits, localization, and grammar: Evidence for a distributive modelof language break-down in aphasic patients and neurologically intactindividuals. Psychological Review, 108, 759–788.

Dickey, M., Choy, J., & Thompson (2007). Real time comprehension of wh-movement in aphasia: Evidence from eyetracking, while listening. Brain andLanguage, 100, 1–22.

Dickey, M. W., & Thompson, C. K. (2009). Automatic processing of wh- and NP-movement in agrammatic aphasia: Evidence from eyetracking. Journal ofNeurolinguistics, 22(6), 563–583. PMC2748948.

Dronkers, N. F., Wilkins, D. P., Van Valin, R. D., Redfern, B. B., & Jaeger, J. J. (2004).Lesion analysis of the brain areas involved in language comprehension.Cognition, 92(1–2), 145–177.

Farmer, T. A., Anderson, S. E., & Spivey, M. J. (2007). Gradiency and visual context insyntactic garden-paths. Journal of Memory and Language, 57(4), 570–595.

Ferreira, F., Anes, M. D., & Horine, M. D. (1996). Exploring the use of prosody duringlangauge comprehension using the auditory moving window technique. Journalof Psycholinguistic Research, 25, 273–290.

Ferreira, F., Henderson, J. M., Anes, M. D., Weeks, P. A., Jr., & McFarlane, D. K. (1996).Effects of lexical frequency and syntactic complexity in spoken languagecomprehension: Evidence from the auditory moving window technique. Journalof Experimental Psychology: Learning, Memory, and Cognition, 22, 324–335.

Fischl, B., & Dale, A. M. (2000). Measuring the thickness of the human cerebralcortex from magnetic resonance images. Proceedings of the National Academy ofSciences, 97, 11044–11049.

Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., et al. (2002).Whole brain segmentation: Automated labeling of neuroanatomical structuresin the human brain. Neuron, 33(3), 341–355.

Fischl, B., Sereno, M. I., & Dale, A. M. (1999). Cortical surface-based analysis. II:Inflation, flattening, and a surface-based coordinate system. NeuroImage, 9(2),195–207.

Fischl, B., van der Kouwe, A., Destrieux, C., Halgren, E., Ségonne, F., Salat, D. H., et al.(2004). Automatically parcellating the human cerebral cortex. Cerebral Cortex,14(1), 11–22.

Grodzinsky, Y. (1990). Theoretical perspectives on language deficits. Issues in thebiology of language and cognition. MIT Press (pp. 192). MIT Press.

Hale, J. (2001). A probabilistic Earley parser as a psycholinguistic model. InProceedings of NAACL (Vol. 2, pp. 159–166).

Hanne, S., Sekerina, I. A., Vasishth, S., Burchert, F., & De Bleser, R. (2011). Chance inagrammatic sentence comprehension: What does it really mean? Evidencefrom eye movements of German agrammatic aphasic patients. Aphasiology, 25,221–244.

Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension.Psychological Review, 99, 122–149.

Karbe, H., Herholz, K., Szelies, B., Pawlik, G., Wienhard, K., & Heiss, W. D. (1989).Regional metabolic correlates of Token test results in cortical and subcorticalleft hemispheric infarction. Neurology, 39(8), 1083–1088.

Kempler, D., Curtiss, S., Metter, E. J., Jackson, C. A., & Hanson, W. R. (1991).Grammatical comprehension, aphasic syndromes and neuroimaging. Journal ofNeurolinguistics.

Levy, R. (2008). Expectation-based syntactic comprehension. Cognition, 106(3),1126–1177.

Lewis, R. L., & Vasishth, S. (2005). An activation-based model of sentence processingas skilled memory retrieval. Cognitive Science, 29, 375–419.

Love, T., Swinney, D., Walenski, M., & Zurif, E. (2008). How left inferior frontal cortexparticipates in syntactic processing: Evidence from aphasia. Brain and Language,107, 203–219.

Love, T., Swinney, D., & Zurif, E. (2001). Aphasia and the time-course of processinglong distance dependencies. Brain and Language, 79, 169–170.

Magnusdottir, S., Fillmore, P., den Ouden, D. B., Hjaltason, H., Rorden, C.,Kjartansson, O., et al. (2012). Damage to left anterior temporal cortex predictsimpairment of complex syntactic processing: A lesion-symptom mappingstudy. Human Brain Mapping, 34, 2715–2723.

Meyer, A. M., Mack, J. E., & Thompson, C. K. (2012). Tracking passive sentencecomprehension in agrammatic aphasia. Journal of Neurolinguistics, 25,31–43.

Naeser, M. A., & Hayward, R. W. (1978). Lesion localization in aphasia with cranialcomputed tomography and the Boston Diagnostic Aphasia Exam. Neurology, 28(6), 545–551.

Newhart, M., Trupe, L. A., Gomez, Y., Cloutman, L. L., Molitoris, J. J., Davis, C. L., et al.(2012). Asyntactic comprehension, working memory, and acute ischemia inBroca’s area versus angular gyrus. Cortex, 48(10), 1288–1297.

Race, D., Ochfeld, E., & Hillis, A. E. (2012). Lesion analysis of cortical regionsassociated with the comprehension of nonreversible and reversible yes/noquestions. Neuropsychologia, 50(8), 1946–1953.

Rademacher, J., Galaburda, A. M., Kennedy, D. N., Filipek, P. A., & Caviness, V. S.(1992). Human cerebral cortex: Localization, parcellation, and morphometrywith magnetic resonance imaging. Journal of Cognitive Neuroscience, 4(4),352–374.

D. Caplan et al. / Brain & Language 152 (2016) 14–27 27

Salat, D. H., Buckner, R. L., Snyder, A. Z., Greve, D. N., Desikan, R. S. R., Busa, E., et al.(2004). Thinning of the cerebral cortex in aging. Cerebral Cortex, 14, 721–730.

Ségonne, F., Dale, A. M., Busa, E., Glessner, M., Salat, D., Hahn, H. K., et al. (2004). Ahybrid approach to the skull stripping problem in MRI. NeuroImage, 22(3),1060–1075.

Shankweiler, D., Crain, S., Gorell, P., & Tuller, B. (1989). Reception of language inBroca’s aphasia. Language and Cognitive Processes, 4, 1–33.

Swinburn, K., Porter, G., & Howard, D. (2004). Comprehensive aphasia test. Hove:Psychology Press.

Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy, J. E. (1995).Integration of visual and linguistic information in spoken languagecomprehension. Science, 268, 1632–1634.

Thompson, C. K., & Choy, J. (2009). Pronominal resolution and gap-filling inagrammatic aphasia: Evidence from eyetracking. Journal of PsycholinguisticResearch, 38, 255–283.

Thothathiri, W., Kimberg, D., & Schwartz, M. (2012). The neural basis of reversiblesentence comprehension: Evidence from voxel-based lesion symptom mappingin aphasia. Journal of Cognitive Neuroscience, 24(1), 212–222.

Tramo, M. J., Baynes, K., & Volpe, B. T. (1988). Impaired syntactic comprehensionand production in Broca’s aphasia: CT lesion localization and recovery patterns.Neurology, 38(1), 95–98.

Tyler, L. (1985). Real-time comprehension processes in agrammatism: A case study.Brain and Language, 26, 259–275.

Tyler, L. K., Marslen-Wilson, W. D., Randall, B., Wright, P., Devereux, B. J., Zhuang, J.,et al. (2011). Left inferior frontal cortex and syntax: Function, structure andbehaviour in left-hemisphere damaged patients. Brain, 134(2), 415–431.

Tyler, L. K., Ostrin, R. K., Cooke, M., & Moss, H. E. (1995). Automatic access of lexicalinformation in Broca’s aphasics: Against the automaticity hypothesis. Brain andLanguage, 48(2), 131–162.

Tyler, L. K., Wright, P., Randall, B., Marslen-Wilson, W. D., & Stamatakis, E. A. (2010).Reorganisation of syntactic processing following LH brain damage: Does RHactivity preserve function? Brain, 133(11), 3396–3408.

van der Kouwe, A. J. W., Benner, T., Fischl, B., Schmitt, F., Salat, D. H., Harder, M., et al.(2005). On-line automatic slice positioning for brain MR imaging. NeuroImage,27(1), 222–230.

van der Kouwe, A. J. W., Benner, T., Salat, D. H., & Fischl, B. (2008). Brainmorphometry with multiecho MPRAGE. NeuroImage, 40(2), 559–569.

Walhovd, K. B., Fjell, A. M., Reinvang, I., Lundervold, A., Dale, A. M., Eilertsen, D. E.,et al. (2005). Effects of age on volumes of cortex, white matter and subcorticalstructures. Neurobiology of Aging, 26(9), 1261–1270.

Warren, J. E., Crinion, J. T., Lambon Ralph, M. A., & Wise, R. J. S. (2009). Anteriortemporal lobe connectivity correlates with functional outcome after aphasicstroke. Brain, 132(12), 3428–3442.

Waters, G. S., & Caplan, D. (1996). The measurement of verbal working memorycapacity and its relation to reading comprehension. Quarterly Journal ofExperimental Psychology, 49A, 51–74.