Brain & Language 111 (2009) 125–139

Contents lists available at ScienceDirect

Brain & Language

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

Word order and Broca’s region: Evidence for a supra-syntactic perspective

Ina Bornkessel-Schlesewsky a,b,*, Matthias Schlesewsky c, D. Yves von Cramon d,e

a Department of Germanic Linguistics, University of Marburg, Marburg, Germanyb Research Group Neurotypology, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germanyc Department of English and Linguistics, Johannes Gutenberg-University, Mainz, Germanyd Department of Cognitive Neurology, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germanye Department of Neurology, Max Planck Institute for Neurological Research, Cologne, Germany

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

Article history:Accepted 12 September 2009Available online 22 October 2009

Keywords:Broca’s areaInferior frontal gyrusPars opercularisWord orderLinearizationReferentialityDefinitenessProminence

0093-934X/$ - see front matter � 2009 Elsevier Inc. Adoi:10.1016/j.bandl.2009.09.004

* Corresponding author. Address: Department of Geof Marburg, Wilhelm-Roepke-Strasse 6A, 35032 Mar6421 2824558.

E-mail address: iboke@staff.uni-marburg.de (I. Bor

It has often been suggested that the role of Broca’s region in sentence comprehension can beexplained with reference to general cognitive mechanisms (e.g. working memory, cognitive control).However, the (language-related) basis for such proposals is often restricted to findings on English.Here, we argue that an extension of the database to other languages can shed new light on the typesof mechanisms that an adequate account of Broca’s region should be equipped to deal with. Thisbecomes most readily apparent in the domain of word order variations, which we examined in Ger-man verb-final sentences using event-related fMRI. Our results showed that activation in the parsopercularis – a core subregion of Broca’s area – was not only modulated by the relative orderingof subject and object, but also by a further factor known to affect word order in a number of lan-guages, namely referentiality. Notably, the finding provides the first demonstration of a wordorder-related activation difference within subject-initial sentences in this region. Additional paramet-ric analyses using individual behavioral data as predictors further attest to the independence of thepars opercularis activation from: (a) sentence acceptability, and (b) difficulty in performing the exper-imental (judgment) task. We argue that these and related findings attest to the need for a processingmechanism that can manipulate predicate-independent, interacting and hierarchically structured rela-tional representations during real time comprehension. These properties pose a challenge to existingaccounts of pars opercularis function.

� 2009 Elsevier Inc. All rights reserved.

1. Introduction

Of the language-related areas of the human brain, Broca’s area(comprising the pars opercularis and triangularis of the left inferiorfrontal gyrus, IFG) is arguably the most famous. In recent years, theprecise role of this region in language comprehension has beensubject to a heated debate, the primary focus of which has beenits increased activation during the comprehension of sentences inwhich the object precedes the subject. An example for a ‘‘non-canonical” or ‘‘permuted” sentence of this type is the (italicized)object-relative clause in ‘Bill caught the burglar who the detectivechased’.

The involvement of Broca’s area in the processing of word orderpermutations has been demonstrated in a large number of neuro-imaging studies. While increased activation of this region for ob-ject-initial sentences was first reported for relative clauses in

ll rights reserved.

rmanic Linguistics, Universityburg, Germany. Fax: +49 (0)

nkessel-Schlesewsky).

English (e.g. Caplan, Alpert, Waters, & Olivieri, 2000; Constableet al., 2004; Just, Carpenter, Keller, Eddy, & Thulborn, 1996; Keller,Carpenter, & Just, 2001; Stromswold, Caplan, Alpert, & Rauch,1996), it has also been observed for clause-medial word order per-mutations (‘‘scrambling”) in German (e.g. Bornkessel, Zysset, Fried-erici, von Cramon, & Schlesewsky, 2005; Friederici, Fiebach,Schlesewsky, Bornkessel, & von Cramon, 2006; Grewe et al.,2005, 2006, 2007; Röder, Stock, Neville, Bien, & Rösler, 2002) andfor object-initial sentences in Hebrew (Ben-Shachar, Palti, & Grod-zinsky, 2004) and Japanese (Kinno, Kawamura, Shioda, & Sakai,2008).

In view of these very consistent results, it is undisputed thatBroca’s area – and particularly the pars opercularis of the left IFG– plays an important role in the processing of varying word orders.This function (‘‘linearization”) is very important for the compre-hension of natural language for at least two reasons: (a) since lan-guage unfolds over time, the order in which sentence constituentsare encountered in the speech stream imposes crucial constraintson how the comprehension process can proceed; and (b) becausean estimated 70% of natural languages exhibit a significant degreeof word order freedom (Steele, 1978), the possibility of variations

126 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

in this order should be considered the rule rather than the excep-tion. Approaches to the function of Broca’s region/the left parsopercularis differ as to whether they attribute its role in the pro-cessing of permuted word orders to language-internal (e.g. syntac-tic movement: Ben-Shachar et al., 2004; Grodzinsky, 2000;Grodzinsky & Friederici, 2006; Grodzinsky & Santi, 2008) or do-main-general operations (e.g. working memory: Caplan et al.,2000; Fiebach, Schlesewsky, Lohmann, von Cramon, & Friederici,2005; Kaan & Swaab, 2002; Müller, Kleinhans, & Courchesne,2003; or cognitive control: Thompson-Schill, Bedny, & Goldberg,2005). From a somewhat more general perspective, it has also beenproposed that the language-related function of Broca’s region canbe subsumed under broader cognitive mechanisms related to ‘‘ac-tion understanding” (Rizzolatti & Arbib, 1998). Language-specificapproaches are often criticised (e.g. Müller & Basho, 2004) becausethey fail to account for the observation that Broca’s region is alsoactivated by non-linguistic tasks, e.g. motor imagery (Binkofskiet al., 2000) and the processing of complex relational information(Kroger et al., 2002). One of the main challenges in accountingfor the involvement of the pars opercularis in the processing ofword order permutations thus lies in determining whether it canbe derived from some more general function of this cortical regionwithin higher cognition.

Here we will argue that, at least with respect to word ordervariations, domain-general approaches and highly specific lan-guage-internal approaches to pars opercularis function have morein common than is typically acknowledged. In our view, both typesof approaches underestimate the complexity of the activation pat-terns that this region shows in response to fine-grained linguisticdifferences. Specifically, we will argue that, while the basic findingof increased activation for object-before-subject orders can be de-rived by all models in a relatively straightforward manner, they areequally challenged by more complex word order patterns. In thefollowing, we first provide a brief summary of previous findingswhich suggest that the relative ordering of subject and objectmay not be the key to explaining pars opercularis activation forword order variations before describing an fMRI study which fur-ther corroborates this perspective.

1.1. Beyond subjects and objects: interacting information types inword order variations

Despite differing viewpoints on precisely which mechanismshould be held responsible for the additional processing costs inobject-before-subject orders, virtually all approaches appeal tothe inherent dependency between objects and subjects in derivingthese costs. For example, it is commonly assumed that compre-hending an object-initial sentence involves reconstructing thecanonical subject-initial order and that an initial object cannot beinterpreted until either the subject or the verb is encountered(see, for example, Kaan & Swaab, 2002).1 One might therefore as-sume that it is this type of dependency that engenders the increasedactivation of the pars opercularis in object-initial orders. If this wereindeed the correct functional characterization of word order-relatedpars opercularis activation, it would provide a candidate mechanismwhich domain-general models would need to explain. It is, of course,easily derived in a movement-based, language-internal approach,since object-initial orders can be modeled theoretically as involving

1 The idea that integration of the object is triggered by the subject has beenproposed for verb-final languages such as German, in which the base position of theobject intervenes between the subject and the verb (see Fiebach, Schlesewsky, &Friederici, 2002). In English, by contrast, the integration of the object can be assumedto be triggered by the verb, irrespective of whether this integration is mediated by atrace or by the direct association between the object and its subcategorizer (Pickering& Barry, 1991). Importantly, however, the processing of the subject is assumed to be anecessary prerequisite for the integration and interpretation of the object in all cases.

an additional movement operation which places the object in frontof the subject.2

However, recent results from German indicate that matters aresomewhat more complex: under certain circumstances, object-ini-tial orders systematically fail to show increased pars opercularisactivation in comparison to their subject-initial counterparts. Thisis the case, for example, when the object bears a higher-ranking the-matic role than the subject, as in (1) (from Bornkessel et al., 2005).

(1) Gestern wurde erzählt,

yesterday was told,dass dem Jungen die Lehrer auffallenthat [the boy]DAT.OBJ.SG [the teachers]NOM.SUBJ.PL

be.striking.toPL

‘Yesterday, someone said that the boy finds the teachersstriking.’

In (1), the verb auffallen (‘to be striking to’) assigns the higher-ranking Experiencer role to the grammatical object, the boy,whereas the grammatical subject, the teachers, bears the lower-ranking role of Theme (or Stimulus) (cf. Grimshaw, 1990; Primus,1999, for theoretical arguments; and Bornkessel, Schlesewsky, &Friederici, 2003, for empirical evidence). Thus, in sentences suchas (1), the object-initial order allows for an independent preferenceto be upheld, namely that arguments bearing higher-ranking the-matic roles should precede arguments bearing lower-ranking the-matic roles. This ordering tendency has been assumed to hold forGerman (Fanselow, 2000; Haider & Rosengren, 2003; Wunderlich,1997) as well as across a wide range of languages (Tomlin, 1986). Ithas also been confirmed empirically (see Haupt, Schlesewsky, Roe-hm, Friederici, & Bornkessel-Schlesewsky, 2008 for the results of arating study with 1120 native speakers of German). In an fMRIstudy, Bornkessel et al. (2005) demonstrated that subject-objectorder interacts with the order of thematic roles in the left parsopercularis: increased activation for object- vs. subject-initial or-ders was only observed for sentences with action verbs (in whichthe subject bears the higher-ranking thematic role than the object),but not for sentences with ‘‘object-experiencer” verbs of the typein (1).

Further converging evidence that object-initial orders do notengender increased pars opercularis activation when the non-canonical order is motivated by an independent word order rulewas provided by Grewe et al. (2005). Here, the independent rulewas one of pronoun placement, namely that pronouns should pre-cede non-pronominal arguments in the medial region of the Ger-man clause (Bierwisch, 1963; Lenerz, 1977; Müller, 1999).Similarly to Bornkessel et al.’s (2005) results on thematic roles,Grewe et al.’s (2005) findings showed that object-initial ordersonly engendered increased activation in the left pars operculariswhen both subject and object were non-pronominal noun phrases,but not when the initial noun phrase was realized by a pronoun.

On the basis of these findings, Bornkessel et al. (2005) andGrewe et al. (2005) concluded that pars opercularis activation inthe domain of word order variations cannot be reduced to a singlefactor, but rather results from the interaction of multiple informa-tion types. They thus put forward the ‘‘linearization hypothesis”,according to which activation in the left pars opercularis is modu-lated by a range of non-syntactic ‘‘prominence scales”, with activa-tion increasing whenever a less prominent argument precedes amore prominent argument. The relevant scales, which are given

2 But note that this is only one possible theoretical characterization of object-before-subject orders. By contrast, a number of theories of grammar do not assumemovement operations (e.g. Lexical Functional Grammar: Bresnan, 2001; ‘‘SimplerSyntax”: Culicover & Jackendoff, 2005; Role and Reference Grammar: Van Valin, 2005;Construction Grammar: Goldberg, 2006).

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 127

in (2), are well-known from cross-linguistic research as they serveto constrain grammatical phenomena in a range of languages (cf.Comrie, 1989; Croft, 2003). It has recently been argued that theyalso play an important role in language processing, for examplein the assignment of thematic interpretations (‘‘who is acting onwhom”) during online comprehension (Bornkessel & Schlesewsky,2006; Bornkessel-Schlesewsky & Schlesewsky, 2009) or in theavoidance of similarity-based interference (Lee, Lee, & Gordon,2007). Furthermore, all of the prominence scales in (2) have beenargued to influence word order preferences both in German(Lenerz, 1977) and cross-linguistically (Tomlin, 1986), i.e. speakersfind sentences more acceptable when a more prominent argumentprecedes a less prominent argument rather than vice versa.

(2) Prominence scales relevant to word order processing (fromWolff, Schlesewsky, Hirotani, & Bornkessel-Schlesewsky,2008).

Note that ‘‘>” should be read as ‘‘is more prominent than”.+nominative > �nominative+pronoun > �pronoun+animate > �animate+specific/+definite > �specific/�definite+actor > �actor (i.e. higher thematic role > lower thematic role)

In accordance with the linearization hypothesis, an influence ofanimacy (2c) on word order-related activation within the left parsopercularis has been demonstrated for both English (Chen, West,Waters, & Caplan, 2006) and German (Grewe et al., 2006).

These previous findings are problematic for a movement-based account of pars opercularis function because they eitherinvolve information types that are not easily amenable to move-ment-based analyses (e.g. animacy, thematic roles) or becausemovement-based approaches make incorrect predictions with re-gard to the fine-grained pattern of activation differences be-tween the different word orders (e.g. pronouns). For a detaileddiscussion of the problems that these findings pose for move-ment-based accounts, see Bornkessel et al. (2005) and Greweet al. (2005, 2006). At the same time, they raise the bar for do-main-general explanations, since the differences between theword orders in question (e.g. in terms of acceptability or otherbehavioral processing measures) are considerably more subtlethan for ‘‘standard” word order manipulations involving onlysubjects and objects. Hence, they are not as easily amenable toone of the most straightforward domain-general explanations,namely that the word order variants engendering increased parsopercularis activation are simply less ‘‘natural” or ‘‘more diffi-cult” in some way.

3 Whereas bare plurals may lead to an ambiguity between a non-specific and aspecific (generic) reading (Carlson, 1977), it suffices for the purposes of the presentmanipulation that, even under a specific reading, they are still outranked by propernames on the definiteness/referentiality scales.

1.2. The present study

In the present study, we aimed to further corroborate the per-spective that word order processing (and its neural bases) requirea multi-factorial account. To this end, we examined a type of prom-inence information that has not been scrutinized from a func-tional-neuroanatomical perspective to date and investigatedwhether prominence-based activation modulations in the parsopercularis can also lead to activation differences between sub-ject-initial structures. To this end, we manipulated word orderand noun phrase definiteness/specificity (cf. 2d) in German. A moreelaborate version of the definiteness/specificity scale (Aissen,2003) is given in (3).

(3) Definiteness scale (Aissen, 2003, p. 437)personal pronoun > proper name > definite NP > indefinite spe-cific NP > non-specific NP

We chose not to manipulate definiteness directly, e.g. by com-paring noun phrases with definite and indefinite determiners, be-cause the indefinite determiner ein(e) (‘a’) is also compatiblewith a numeral reading (‘one’) in German. It could therefore beinterpreted as a quantifier (Fodor & Sag, 1982), thereby leadingto a confound during word order processing (cf. the results of therating study in Haupt et al., 2008, for a first indication of such aninfluence). Specificity is also difficult to manipulate directly in Ger-man as it is not morphologically expressed (in contrast to lan-guages such as Turkish, e.g. Comrie, 1989). For the prominencemanipulation, we thus drew upon a subpart of the definitenessscale, which, following Croft (2003), can be termed the ‘‘referen-tiality scale” (cf. 4). Note that, rather than representing a theoreti-cal claim about the nature of referentiality, this scale expresses across-linguistic generalization about different noun phrase typesand their participation in a range of morphosyntactic phenomena(see Croft, 2003, Chapter 5, for details).

(4) Referentiality scale (Croft, 2003, p. 130)pronoun < proper name < common noun

In accordance with the scale in (4), the present study contrastedsentences in which an argument that was higher in referentiality (aproper name) preceded an argument that was lower in referential-ity (a bare plural common noun) in terms of linear order or viceversa. Proper names are higher in referentiality than commonnouns—and thereby more prominent—because they are usuallyuniquely identifiable, whereas common nouns denote sets of enti-ties.3 Crucially, note that the purpose of the present study was not toexamine the neural representation of referentiality per se (e.g. byexamining whether proper names and bare plurals engender activa-tion in different neural regions) but rather to investigate how thereferentiality scale influences the processing of argument order.

Whereas, prima facie, one might question whether the referen-tiality distinction between proper names and common nouns alsoholds when a sentence is presented out of context, previous psy-cholinguistic studies have demonstrated robust differences be-tween proper names and common nouns even in isolatedsentences (Gordon, Hendrick, & Johnson, 2001; Gordon, Hendrick,& Levine, 2002; Lee et al., 2007). In particular, these previousexperiments suggest that similarity-based interference betweenco-arguments is reduced when one argument is a proper nameand the other is a common noun, thereby attesting to distinct rep-resentations for the two noun phrase types and to the robustnessof the referentiality contrast. By using names and bare plurals wewere thus able to manipulate argument prominence on the defi-niteness/specificity scale (3), while avoiding the potential con-founding influence of a quantificational reading that might resultfrom the use of indefinite determiners. The critical sentence typesresulting from this manipulation are shown in Table 1.

The experimental conditions in Table 1 instantiate a 2 � 2 de-sign involving the factors subject-object ORDER (subject-initial,A/B vs. object-initial, C/D) and REFerentiality (proper name pre-cedes plural nominal, A/C vs. plural nominal precedes propername, B/D). As a result of the referentiality manipulation (i.e. theuse of proper names and bare plurals as motivated above), the crit-ical sentences were locally ambiguous between a subject- and anobject-initial reading, with disambiguation effected via the num-ber agreement information of the clause-final verb. Previous find-ings suggest that this should not affect activation in the left parsopercularis: for sentences structures similar to those under

Table 1Critical sentence conditions in the present experiment. Stimulus segmentation isindicated by the vertical bars. Abbreviations used: SO = subject-before-object;OS = object-before-subject; REF = NP higher in referentiality (proper name) precedesNP lower in referentiality (bare plural); NREF = NP lower in referentiality (bare plural)precedes NP higher in referentiality (proper name); ACC = verb subcategorizing for anaccusative object; DAT = verb subcategorizing for a dative object; SG = singular;PL = plural. The variation between accusative and dative verbs was implemented as abetween-subjects factor (see Section 2). Whether a sentence is subject- or object-initial is determined via number agreement between subject and verb.

Condition Example

Common matrixclause

Gestern hat jeder gehört, . . .

yesterday has everyone heard . . .

‘Yesterday, everyone heard . . .’

A. SO-REF . . . dass | Reinhold | Autorinnen | auslacht/applaudiert.. . . that ReinholdSG authorsPL laughs-atACC.SG/applaudsDAT.SG

‘. . . that Reinhold laughs at/applauds authors.’

B. SO-NREF . . . dass | Autorinnen | Reinhold | auslachen/applaudieren.. . . that authorsPL ReinholdSG laugh-atACC.PL/applaudDAT.PL

‘. . . that authors laugh at/applaud Reinhold.’

C. OS-REF . . . dass | Reinhold | Autorinnen | auslachen/applaudieren.. . . that ReinholdSG authorsPL laugh-atACC.PL/applaudDAT.PL

‘. . . that authors laugh at/applaud Reinhold.’

D. OS-NREF . . . dass | Autorinnen | Reinhold | auslacht/applaudiert.. . . that authorsPL ReinholdSG laughs-atACC.SG/applaudsDAT.SG

‘. . . that Reinhold laughs at/applauds authors.’

128 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

examination here, Bornkessel et al. (2005) found that word order-related activation of this cortical region did not differ between lo-cally ambiguous sentences and unambiguous sentences and alsothat the relevant word order contrasts generalize over singularand plural subjects and objects (in both ambiguous and unambig-uous sentences).

Using this experimental design, we aimed to test the predictionof the linearization hypothesis (Bornkessel et al., 2005; Greweet al., 2005) that those parts of ‘‘Broca’s region” that show in-creased activation for object- vs. subject-initial orders (specificallythe left pars opercularis) should also be sensitive to referentiality.We thus predicted that the voxels within the pars opercularis thatare activated by the ORDER contrast (object-initial – subject-ini-tial) should also show increased activation for the REFerentialitycontrast (bare-plural-initial – proper-name-initial). Furthermore,if the linearization hypothesis is correct, differences in argumentprominence (cf. 2) should not only serve to attenuate activation in-creases for object-initial sentences, but should also lead to activa-tion differences between subject-initial conditions. This predictionwas also tested in the present study.

2. Materials and methods

2.1. Participants

Thirty monolingually raised, right-handed native speakers ofGerman (16 females; mean age: 25.0 years; age range: 21–32 years) participated in the fMRI study after giving written in-formed consent. Two further participants were excluded from thefinal data analysis on account of movement artifacts and/or prob-lems in performing the task.

2.2. Materials

Thirty-six sets of the four critical sentence conditions (see Ta-ble 1) were constructed on the basis of lexical triplets consistingof a proper name, a bare plural noun, and a verb assigning accusa-tive object case. Each sentence was combined with a matrix clauseof the form shown in Table 1 (matrix clauses were always identical

for one lexical set). An additional 36 sets of the four critical condi-tions were constructed by replacing the verb in each of the first 36sets with a verb assigning dative object case. This variation (CASE)was implemented as a between-participants factor in order to con-trol for possible influences of different kinds of object case on wordorder processing (see Bornkessel, McElree, Schlesewsky, & Frieder-ici, 2004, for ERP data showing such differences). Note that allverbs allowed generic and indefinite readings for bare plural sub-jects and objects, hence ruling out a subject-object asymmetrybased on differences in ambiguity. The complete experimentalmaterials thus consisted of 144 sentences with accusative verbs(36 per condition) and 144 sentences with dative verbs (36 percondition). These were subsequently subdivided into four lists of72 sentences each, with two lists containing only sentences withaccusative verbs and two lists containing only sentences with da-tive verbs. Each list contained 18 sentences from each of the fourconditions in Table 1 and two sentences from each lexical set. Listpresentation was counterbalanced across participants. Each partic-ipant read the materials from a single list once and was therebyeither confronted with critical sentences containing accusativeverbs or with critical sentences containing dative verbs.

Each list of 72 critical sentences was pseudo-randomly com-bined with 72 filler sentences, 42 of which were ungrammatical(to balance responses for the acceptability judgment task, see be-low). Thirty-six null events (empty trials) were also included ineach list in order to improve the statistical evaluation of the data(Miezin, Maccotta, Ollinger, Petersen, & Buckner, 2000). The pseu-do-randomization ensured that not more than two sentences froma single condition were presented in a row, that at least twenty tri-als intervened between the repetition of lexical materials and thatcritical sentences, null events and fillers were distributed equallyamong the two functional runs (see below). The filler sentenceswere identical in all versions of the experiment and were of a sim-ilar structure to the critical sentences (including a matrix clause ofthe type exemplified in Table 1 and a subordinate clause intro-duced by the complementizer dass). Furthermore, the filler sen-tences included varying orders and noun phrases of varyingreferentiality so that they would be initially indistinguishable fromthe critical experimental sentences. Ungrammatical fillers includedviolations of case marking or agreement.

2.3. Procedure

Participants read the sentences via LCD goggles (Visuastim,Magnetic Resonance Technology, Northridge, CA). To control forreading strategies, sentences were presented in a segmented man-ner, with a presentation time of 600 ms for the matrix clause,500 ms for each of the arguments in the embedded clause and400 ms for the complementizer dass and the clause-final verb.The inter-stimulus interval (ISI) was always 100 ms. Trials beganwith the presentation of an asterisk (300 ms plus 200 ms ISI) andended with a 500 ms pause, after which a question mark signaledto participants that they should judge the acceptability of the pre-ceding sentence. In the instructions, it was emphasized that partic-ipants should not judge whether the sentences described aplausible event, but rather whether the chosen way of expressingthis event is possible in German or not. Participants were also in-structed to rely on their intuitions rather than on rules of ‘‘goodstyle” that they might remember from school (where ‘‘subject-first” is often enforced as a prescriptive stylistic rule). The rationalefor using a judgment task was twofold: (a) it ensured comparabil-ity with a number of previous studies on word order and argumentprominence in German (Grewe et al., 2005, 2006, 2007); (b) itserved to avoid possible activations that have been shown to occurin response to a probe sentence in a comprehension task (Caplan,Chen, & Waters, 2008). The participants performed the judgment

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 129

task by pressing one of two push-buttons with their right indexand middle fingers (maximal response time: 2300 ms). After theresponse, the screen remained blank for the remainder of the trial.The assignment of fingers to acceptable and unacceptable wascounterbalanced across participants.

Trials were presented with variable onset delays of 0, 400, 800,1200 or 1600 ms, thereby leading to an oversampling of the actualimage acquisition time of 2000 ms by a factor of 5 (Miezin et al.,2000). All trials (180 per participant) had a length of 8 s, thusresulting in a total measurement time of 24 min, which was sepa-rated into two functional runs.

Participants completed a short practice session before enteringthe scanner.

2.4. fMRI data acquisition

The experiment was carried out on a 3T scanner (Medspec 30/100, Bruker, Ettlingen). Twenty axial slices (19.2 cm FOV, 64 by64 matrix, 4 mm thickness, 1 mm spacing), parallel to the AC-PCplane and covering the whole brain were acquired using a singleshot, gradient recalled EPI sequence (TR 2000 ms, TE 30 ms, 90� flipangle). Two functional runs of 360 time points were collected, witheach time point sampling over the 20 slices. Prior to the functionalruns, 20 anatomical T1-weighted MDEFT (Norris, 2000; Ugurbilet al., 1993) images (data matrix 256 � 256, TR 1.3 s, TE 10 ms)and 20 T1-weighted EPI images with the same geometrical param-eters as the functional data were acquired.

2.5. fMRI data analysis

The fMRI data were analyzed using the LIPSIA software package(Lohmann et al., 2001). Functional data were corrected for motionusing a matching metric based on linear correlation. To correct forthe temporal offset between the slices acquired in one scan, a cu-bic-spline-interpolation based on the Nyquist-Shannon-Theo-rem was applied. A temporal highpass filter with a cutofffrequency of 1/112 Hz was used for baseline correction of the sig-nal and a spatial Gaussian filter with 5.65 mm FWHM was applied.

To align the functional data slices onto a 3D stereotactic coordi-nate reference system, a rigid linear registration with 6� of freedom(three rotational, three translational) was performed. The rota-tional and translational parameters were acquired on the basis ofthe MDEFT and EPI-T1 slices to achieve an optimal match betweenthese slices and the individual 3D reference data set (acquired foreach subject during a previous scanning session). The MDEFT vol-ume data set with 160 slices and 1 mm slice thickness was stan-dardized to the Talairach stereotactic space (Talairach &Tournoux, 1988). The same rotational and translational parameterswere normalized, i.e., transformed to a standard size via linearscaling. The resulting parameters were used to transform the func-tional slices using trilinear interpolation, so that the resulting func-tional slices were aligned with the stereotactic coordinate system.This normalization process was improved by a subsequent non-lin-ear normalization (Thirion, 1998), which serves to adjust struc-tural-anatomical differences between different brains by using aprocedure known as ‘‘demon matching”. Here, an anatomical 3Ddata set is deformed such that it matches another 3D data set thatserves as a fixed reference image.

The statistical evaluation was based on a least-squares estima-tion using the general linear model for serially autocorrelatedobservations (see Aguirre, Zarahn, & d’Esposito, 1997; Fristonet al., 1995; Worsley & Friston, 1995; Zarahn, Aguirre, & d’Esposito,1997). The design matrix was generated with a box-car functionconvolved with the hemodynamic response function (constructedby a gamma density function, Glover, 1999). The response delaywas set to 6 s and the undershoot delay to 16 s. The model equa-

tion, including the observation data, the design matrix and the er-ror term, was convolved with a Gaussian kernel of 4 s. FWHMdispersion to deal with the temporal autocorrelation (Worsley &Friston, 1995). Single-participant contrast-images were enteredinto a second-level random effects analysis for each of the con-trasts. The group analysis consisted of a one-sample t-test acrossthe contrast images of all subjects (Holmes & Friston, 1998). Subse-quently, t values were transformed into Z scores. To protect againstfalse positive activations, only regions with a Z score greater than3.1 (p < 0.001 uncorrected) and with a volume greater than 288mm3 (eight measured voxels) were considered (Braver & Bongiol-atti, 2002; Forman et al., 1995).

3. Results

3.1. Behavioral data

The behavioral task revealed the following mean acceptabilityrates/standard deviations: SO-REF (95.65%/5.56); SO-NREF(94.17%/11.23); OS-REF (79.44%/22.70); OS-NREF (43.33%/37.61).A repeated measures ANOVA revealed significant main effects of(subject/object) ORDER (F(1, 28) = 63.41; p < .0001) and REFeren-tiality (F(1, 28) = 27.95; p < .0001) as well as a significant interac-tion ORDER � REF (F(1, 28) = 17.59; p < .001). Resolving thisinteraction by ORDER showed a significant effect of REF for ob-ject-initial (F(1, 28) = 23.16; p < .0001) but not for subject-initialsentences (p > .26).

A closer examination of the individual participant data showedthat the mean acceptability rating for the OS-NREF condition(43.33%) resulted from a relatively high variability of judgmentsacross participants. Thus, as visualized in Fig. 1, several partici-pants showed a very high mean acceptability rating for this condi-tion, while others reliably judged it as unacceptable. This pattern ofinter-individual variability did not, however, generalize to the OS-REF condition, i.e. it was not the result of certain participantsaccepting and certain participants rejecting object-initial struc-tures in general (see Fig. 1). Rather, the data pattern speaks in favorof an acceptability continuum, in which the addition of a furtherinfluencing factor (e.g. a violation of the referentiality principle)may lead to a reversal of a participant’s judgment behavior. Notethat this data pattern is by no means unprecedented in the domainof word order variations in German: a range of studies has demon-strated that speakers’ intuitions in this domain are graded ratherthan absolute (e.g. Keller, 2000; Pechmann, Uszkoreit, Engelkamp,& Zerbst, 1994; Schlesewsky, Bornkessel, & McElree, 2006). Fur-thermore, as shown by Keller (2000), different word order con-straints can act in a cumulative manner. Thus, it is not surprisingthat referentiality has a stronger effect in the object-initial sen-tences. Finally, a recent ERP study using an identical experimentalmanipulation (Haupt et al., 2008, Experiment 2) revealed a verysimilar pattern of acceptability judgments, but showed by meansof a comprehension task that the low acceptability of the OS-NREFsentences does not preclude a correct understanding of the sen-tences. Taken together, these previous results indicate that theacceptability pattern observed here is by no means atypical andthat it should not be taken to indicate that participants did notunderstand the sentences (for a dissociation between acceptabilityand comprehensibility in German word order variations, see alsoFriederici et al., 2006). Nevertheless, in order to determine theinfluence of individual judgment patterns on our results, we con-ducted parametric analyses of our fMRI data using the behavioraldata as predictors (see below).

Turning now to the analysis of the reaction times, this yieldedthe following mean values/standard deviations: SO-REF (595 ms/168); SO-NREF (653 ms/194); OS-REF (822 ms/253); OS-NREF

Fig. 1. Individual participant acceptabilities for the two object-initial conditions in the present experiment. Error bars indicate standard errors of the mean.

130 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

(817 ms/293). A repeated measures ANOVA again revealed a signif-icant main effect of ORDER (F(1, 28) = 65.46; p < .0001) and a sig-nificant interaction of ORDER and REF (F(1, 28) = 5.42; p < .05).Planned comparisons for each of the two levels of ORDER showeda significant effect of REF for subject-initial (F(1, 28) = 18.96;p < .001) but not for object-initial structures (F < 1).

3.2. fMRI data: order and referentiality

In a first step, we calculated the contrast between object- andsubject-initial sentences in order to isolate the network involvedin the processing of non-canonical structures and to thereby en-able us to examine the referentiality effect within this network.The results of this contrast are shown in Figs. 2 and 3 and Table 2.

As predicted, the present study revealed increased activation forobject- vs. subject-initial sentences in the pars opercularis of the

left IFG, with activations extending into the adjacent deep frontaloperculum/anterior insula and the inferior frontal junction area(IFJ). Further activations were observed in the right deep frontaloperculum, left intraparietal sulcus (IPS) and in the head of the leftcaudate nucleus. This network is very similar to previous findingsfor word order permutations in the medial portion of the Germanclause (e.g. Bornkessel et al., 2005; Friederici et al., 2006; Greweet al., 2005; Röder et al., 2002).

In a second step, we calculated the main effect of referentiality(i.e. the contrast bare-plural-initial order, NREF – proper-name-ini-tial order, REF). As shown in Fig. 4 and Table 3, sentences in whichthe less prominent bare plural argument preceded the more prom-inent proper name yielded increased activation in the left inferiorfrontal junction area, extending into pars opercularis of the inferiorfrontal gyrus, and in the left intraparietal sulcus. By contrast, in-creased activation for sentences with a proper name preceding a

Fig. 3. Averaged activation for the word order contrast (object-initial > subject-initial sentences; z > 3.09).

Fig. 2. Averaged activation in the pars opercularis of the left IFG for the word ordercontrast (object-initial > subject-initial sentences; z > 3.09). The bar plot shows theaverage percent signal change (�2 to +2 s relative to the point of maximal signalchange) for the activation maximum within the pars opercularis (�53 11 5).Abbreviations: SO = subject-before-object; OS = object-before-subject; REF = NPhigher in referentiality (proper name) precedes NP lower in referentiality (bareplural); NREF = NP lower in referentiality (bare plural) precedes NP higher inreferentiality (proper name).

Table 2Talairach coordinates, maximal z-values and volumes of the activated region for thelocal maxima in the ORDER (object- vs. subject-initial) contrast. Only activations witha z-value >3.09 and a volume of at least 288 mm3 (eight measured voxels) wereconsidered. Local maxima were defined as the largest z-value exceeding 3.09 within a10 mm radius.

Region Talairachcoordinates

Max.z-value

Volume(mm3)

ORDER contrast (OS vs. SO)L. inferior frontal gyrus (IFG), pars

opercularis�53 11 5 4.35

L. deep frontal operculum/anterior insula �34 23 5 4.73 11,966L. inferior frontal junction area (IFJ) �40 5 32 4.77R. deep frontal operculum/anterior insula 31 20 6 4.49 1693L. pre-SMA �7 23 41 4.89 4300L. intraparietal sulcus (IPS) �32 �58

414.22 915

L. caudate nucleus �14 1 18 3.78 309

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 131

bare plural was observable in the posterior cingulate gyrus and thefrontomedian wall (BA 10). An additional conjunction analysis be-tween the ORDER and REF contrasts showed common activationonly in left inferior frontal regions, namely in the left IFG (parsopercularis) and left IFJ (maximum: �41 9 30).

Finally, we examined the possible main effect of case by com-paring the activation across all four critical sentence conditionsin the two participant groups (ACC vs. DAT). This analysis revealedno significant differences.

As the primary aim of this study was not to examine effects ofreferentiality per se, but to investigate the impact of this lineariza-tion factor within the network that is sensitive to word order per-mutations (i.e. object-before-subject orders), we conducted aregion of interest analysis for each of the regions identified usingthe order contrast (see Table 2). For each of these regions, we ex-tracted the time course of the underlying BOLD signal. The percentsignal change (relative to the mean signal intensity) for the voxelwith the highest z-value (in the group contrast) and the 26 adja-cent voxels was averaged for each condition and participant.4 Thetime course of the null events was subtracted from the time coursesfor the critical sentence conditions (Burock, Buckner, Woldorff, Ro-sen, & Dale, 1998).

The extracted time courses (mean percent signal change for atime window from –2 to + 2 s relative to the maximal signalchange) were subjected to repeated measures ANOVAs involving

4 ROIs were defined as the activation maximum and the 26 voxels adjacent to it inaccordance with the procedure adopted in previous studies examining word orderand linearization parameters (Bornkessel et al., 2005; Grewe et al., 2005, 2006, 2007).

the within-subject factors word order (ORDER; subject-object vs.object-subject), referentiality (REF; proper-name-initial vs. bare-plural-initial) and the between-subjects factor object case (CASE;accusative vs. dative; see Section 2.2). The results of these analysesare reported in Table 4. The average time courses themselves arevisualized in Fig. 5.

As is evident from Table 4, three regions within the overallnetwork sensitive to linear order also proved sensitive to

Fig. 4. Averaged activation for the referentiality contrast (NREF-conditions > REF-conditions; z > 3.09).

Table 3Talairach coordinates, maximal z-values and volumes of the activated region for thelocal maxima in the Ref(erentiality) contrast. Only activations with a z-value >3.09and a volume of at least 288 mm3 (eight measured voxels) were considered.

Region Talairachcoordinates

Max. z-value

Volume(mm3)

REF contrast (NREF vs. REF)L. inferior frontal junction area (IFJ) �41 9 30 3.96 1431L. intraparietal sulcus (IPS) �29 �66 39 4.97 675

REF contrast (REF vs. NREF)R. posterior cingulate gyrus 1 �21 36 3.79 648R. frontomedian wall (BA 10) 1 54 3 4.16 2592

132 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

referentiality: the pars opercularis of the left IFG, the left IFJ, andleft IPS. In order to test the strongest prediction of the linearizationhypothesis, namely that referentiality should lead to an activationdifference between subject-initial word orders in the pars opercu-laris (see the introduction), we computed an additional plannedcomparison between these conditions. As predicted, the two sub-ject-initial orders indeed differed significantly from one another:F(1, 28) = 7.66, p < .01 (corrected for multiple comparisons usingthe modified Bonferroni procedure; Keppel, 1991).

Table 4Summary of the global statistical analysis for the averaged percent signal change for the voxthe ORDER contrast (cf. Table 2). Each cell gives the significance level for an effect (n.s. =significant effects. In all cases, the degrees of freedom were df1 = 1, df2 = 28. Abbreviationsfor a dative object; SG = singular; PL = plural.

Region ORDER REF CASE R

L. IFG, pars opercularis ��� (25.71) � (6.26) n.s. nL. deep frontal operculum/anterior insula ��� (18.68) n.s. n.s. nL. IFJ ��� (18.57) �� (10.95) n.s. nR. deep frontal operculum/anterior insula ��� (17.40) n.s. n.s. n

L. pre-SMA ��� (33.15) m. (4.15) n.s. nL. IPS ��� (14.39) �� (8.30) n.s. n

L. caudate nucleus ��� (20.77) n.s. n.s. n

3.3. fMRI data: parametric analysis using individual behavioral data(acceptability, standard deviation of acceptability, reaction time)

In order to examine the nature of the activations observed fororder and referentiality more closely, we performed a series ofadditional analyses in which we used the individual behavioraldata acquired in the scanner as parametric predictors. This analysiswas motivated: (a) by the inter-individual variability in the accept-ability of the object-initial structures (see above), and (b) by theconjecture formulated in Grewe et al. (2005) that the activationof the deep frontal operculum/anterior insula in the context ofword order variations may arise from the relative ease or difficultyof performing an acceptability judgment task. For the pars opercu-laris, by contrast, the linearization hypothesis would not predictthat the activation is simply due to a correlation with behavioralparameters. The parametric analyses tested these hypotheses byexamining for which brain regions the activation patterns in thepresent study could be predicted on the basis of behavioral param-eters alone.

To estimate the influence of all possible aspects of the judgmenttask, we conducted parametric analyses (Büchel, Wise, Mummery,Poline, & Friston, 1996) using each participant’s mean acceptabilityrating, mean reaction time and standard deviation of acceptabilityfor each of the four critical conditions as predictor values (in sep-arate models). Each of these analyses used a regressor consistingof the behavioral parameter of interest per participant per condi-tion, e.g. for the acceptability, the mean acceptability for each par-ticipant and for each of the four critical conditions (see Volz,Schubotz, & von Cramon, 2003, for a similar application in a non-linguistic context). These parametric analyses should thereforeidentify brain regions correlating with the demands of the behav-ioral task, as measured by individual performance. While increasedreaction times may, in principle, reflect either increased difficultyin understanding the sentence or in performing the judgment task,we also chose the individual standard deviation (of acceptability)per condition as we reasoned that this value would provide a goodmeasure of how confident participants were in judging a particularsentence condition as acceptable or unacceptable. As noted above,is well-known that speakers’ judgments about their native lan-guage are often not categorical, but that they may rather also dis-play a certain degree of ‘‘gradience” (for comprehensivediscussions of this phenomenon, see for example Fanselow, Féry,Vogel, & Schlesewsky, 2006; Featherston, 2007, and the commen-taries on this article) and word order permutations of the typeexamined here are particularly prone to such gradient judgmentbehavior. Thus, the standard deviation of the individual acceptabil-ity ratings per condition should provide a good estimate of howcertain individual participants were in judging a particularsentence type as either acceptable or unacceptable. Furthermore,

el with the maximal activation and the 26 adjacent voxels in the regions identified vianot significant; m = p < .07; � = p < .05; �� = p < .01; ��� = p < .001) and the F-value forused: ACC = verb subcategorizing for an accusative object; DAT = verb subcategorizing

EF � ORDER CASE � ORDER CASE � REF CASE � REF � ORDER

.s. n.s. n.s. n.s.

.s. n.s. n.s. n.s.

.s. n.s. n.s. n.s.

.s. � (5.16) n.s. n.s.ACC DAT�� (17.13) n.s. n.s.

.s. n.s. n.s. n.s.

.s. � (4.56) n.s. n.s.ACC DAT�� (11.67) n.s. n.s.

.s. n.s. n.s. n.s.

Fig. 5. Averaged BOLD timecourses for the regions of interest defined via the word order contrast (see Table 2). Abbreviations: SO = subject-before-object; OS = object-before-subject; REF = adhering to the referentiality principle; NREF = violating the referentiality principle.

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 133

Fig. 6. Results of the parametric analyses using individual behavioral data. The figure shows averaged brain activations (z > 3.09) correlating with individual standarddeviations of acceptability ratings (panel A), with individual mean reaction times (panel B) and with individual mean acceptability ratings (panel C).

Table 5Talairach coordinates, maximal z-values and volumes of the activated regions in theparametric analyses using individual acceptability ratings, reaction times andstandard deviations of acceptability for the four critical conditions. Only activationswith a z-value >3.09 and a volume of at least 288 mm3 (eight measured voxels) wereconsidered.

Region Talairachcoordinates

Max.z-value

Volume(mm3)

AcceptabilityR. post-central sulcus 43 �27 27 3.84 378

Reaction timeL. deep frontal operculum �38 18 3 4.45 1863

Standard deviation (acceptability)L. deep frontal operculum �38 27 3 3.98 540R. deep frontal operculum 28 24 3 3.61 486

134 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

because of the gradient acceptability of the structures under con-sideration, there were essentially no right or wrong answers inthe judgment task. Therefore, reaction times for all trials enteredthe parametric analysis (i.e. not just ‘‘acceptable” responses). Theresults of the parametric analyses are shown in Fig. 6 and summa-rized in Table 5.

As Fig. 6 and Table 5 show, the measures that can be inter-preted as reflecting the relative degree of difficulty in performingthe judgment task (reaction time, standard deviation of accept-ability) correlated with an increased activation of deep fronto-opercular cortices. (Mean) acceptability per se, by contrast, isnot a good predictor for activation in the overall network of re-gions that proved sensitive to the word order manipulation inthe present study.

4. Discussion

The present findings provide the first demonstration that sen-tence-level linearization principles can engender activation differ-ences between subject-initial sentences in the pars opercularis5 of

5 Note that, in the sense used here, ‘‘pars opercularis” refers exclusively to thelateral convexity of the opercular part of the inferior frontal gyrus.

the left IFG as part of a (mainly) left-lateralized fronto-parietal-basalganglia network. In the following, we discuss the consequences ofthis finding for a characterization of the processing mechanismsunderlying linear order in language.

4.1. Referentiality as a linearization parameter

The finding of referentiality-based activation differences in anetwork responsive to the linearization of arguments is significantin several respects. Firstly, referentiality cannot be considered anentirely syntactic parameter. Rather, linearization effects due tothis information type are closely tied to the interface between syn-tax, semantics and pragmatics, as they are essentially a subcase ofthe more general preference to order old (or known) informationbefore new information (e.g. Choi, 1999; Lenerz, 1977). The presentfindings thus lend further support to a ‘‘supra-syntactic” account ofthe functional neuroanatomy of word order processing (see alsoGrewe et al., 2006). A second and related point concerns the obser-vation of referentiality-based activation differences in the parsopercularis between two subject-initial sentence structures. Asoutlined in the introduction, a crucial prediction of the lineariza-tion hypothesis was that order-based activation differences in thisregion are, in principle, independent of the relative positioning ofsubject and object. The finding of activation differences betweensubject-initial structures thus provides strong converging supportfor this claim. Crucially, as the two critical subject-initial condi-tions did not differ in acceptability (see Haupt et al., 2008, for con-verging acceptability data which attest to the reliability of thisobservation), the activation increase for the SO-NREF conditioncannot be attributed to lower acceptability or a decrease in ‘‘natu-ralness”. Furthermore, the present results demonstrate that word-order-based activation differences cannot generally be reduced to adependency between the first (‘‘permuted”) element encounteredand a second element on which it is dependent (as sketched outfor object-before-subject orders in the introduction): increasedprocessing cost for a violation of the referentiality principle cannotarise until the relative referentiality status of both arguments hasbeen evaluated. This can only take place once the second argumenthas been reached, since an initial argument that is low on the ref-erentiality scale is fully compatible with either an intransitive

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 135

(one-argument) structure or with a transitive structure with twoarguments that are low in referentiality. As we will show below,these properties are associated with profound consequences for acharacterization of the processing mechanisms underlying in-creases in neural activation for permuted word orders.

4.2. A network for linear order?

While our prediction – as based on the linearization hypothesis– focused mainly on the left pars opercularis, the complete neuralnetwork for the processing of linear order in language of coursealso encompasses a number of further regions. In the presentstudy, the ORDER manipulation engendered activation in a greaternetwork of regions comprising bilateral deep frontal opercular cor-tices, left basal ganglia (caudate nucleus), left pre-SMA, left inferiorfrontal junction area (IFJ) and left intraparietal sulcus (IPS). Whileall of these regions have been observed in one or more previousstudies examining clause-medial word variations in German(Bornkessel et al., 2005; Friederici et al., 2006; Grewe et al., 2005,2006, 2007; Röder et al., 2002), individual studies typically onlyyielded activation in a subset of the overall network observed here.As we will outline in the following, we believe that a more preciseconsideration of which (combinations of) regions were observed inwhich studies can help to shed further light on the factors involvedin modulating subparts of the overall ‘‘linearization network”.

4.2.1. Frontoparietal (IFJ-IPS) networkWith regard to the present study, recall that – in addition to the

left pars opercularis - only the left IFJ and left IPS regions alsoproved sensitive to the referentiality manipulation. The only otherstudy on word order variations in similar sentence structures thatobserved a left-lateralized network consisting of the pars opercu-laris, IFJ and IPS was reported by Bornkessel et al. (2005). The maincommonality between this experiment and the present study wasthat both employed ambiguous sentence structures, in which theorder of subject and object was only disambiguated at the end ofthe critical embedded sentences. By contrast, all other experimentson word order permutations in comparable sentence structures inGerman used only unambiguously case marked sentences in whichthe word order was clear from the very beginning (Friederici et al.,2006; Grewe et al., 2005, 2006, 2007; Röder et al., 2002). As localambiguities between a subject- and an object-first reading arepreferentially resolved towards a subject-first order (Frazier & Flo-res d’Arcais, 1989), initially ambiguous object-initial sentences ofthe type used here and in Bornkessel et al. (2005) require reanaly-sis processes in order for the correct reading to be computed. Asreanalysis is typically described as a late and controlled processduring language comprehension (e.g. Friederici & Alter, 2004), itappears highly plausible that precisely this type of mechanismshould require the recruitment of a larger network of regions thatare well-known for their involvement in cognitive control. Notably,the association between a subnetwork involving the IFJ/IPS andcognitive control has been established in a range of experiments(e.g. Brass & von Cramon, 2002, 2004; Bunge, Hazeltine, Scanlon,Rosen, & Gabrieli, 2002).6 We therefore suggest that the networkshowing a sensitivity to both subject-object order and referentiality

6 An account along these lines can also derive the sensitivity of the frontoparietal(IFJ-IPS) network to further information types important to linearization (thethematic hierarchy in Bornkessel et al., 2005; and referentiality in the present study).While these information types do not impact upon the initial subject-preference forlocally ambiguous sentences (cf. Haupt et al., 2008, and Kretzschmar, Bornkessel-Schlesewsky, Staub, Roehm, & Schlesewsky, submitted for publication, for ERP andeye-tracking evidence in this regard), they can aid the reanalysis process by serving asadditional cues for the object-initial order (see Bornkessel et al., 2004, for a detaileddescription). They are therefore relevant for determining the overall ‘‘cost” ofreanalysis and, thereby, the degree of cognitive control required.

in the present experiment (left pars opercularis, IFJ and IPS) sub-serves the processing of linear order under conditions giving riseto increased demands on cognitive control. An account along theselines might also explain the involvement of the caudate nucleus,which has also been implicated in language processing under ‘‘con-trolled” circumstances (Friederici, 2006). However, this region didnot show an effect of referentiality.

4.2.2. Pre-SMA and deep frontal operculum/anterior insulaWhen activations in the pre-SMA and deep frontal operculum/

anterior insula are observed in studies examining word order vari-ations in German, these tend to co-occur (Friederici et al., 2006;Grewe et al., 2005). In particular, the likelihood for an involvementof these regions appears to increase with the use of gradient stim-uli (i.e. word order permutations that give rise to significantly de-graded acceptability ratings) and an acceptability judgment task.On the basis of this observation and following a proposal advancedin Grewe et al. (2005), we assume that this subnetwork may beparticularly involved in the evaluation of linguistic structures un-der conditions of increased uncertainty. Such an assumption ishighly compatible with the results of studies investigating thefunctional neuroanatomy of decision-making in the presence ofuncertainty (e.g. Volz et al., 2003).

The results of our parametric analyses provide converging sup-port for this ‘‘decision-based” account: of the regions that provedsensitive to our critical manipulation (order and referentiality),only the deep frontal operculum covaried with any of the behav-ioral predictors. Specifically, individual standard deviations ofacceptability predicted the bilateral activation of fronto-opercularcortices, while individual reaction times predicted the activationpattern of the left deep frontal operculum. While we do not wishto speculate with respect to a possible functional significance ofthe lateralization differences observed for the two predictors, theirjoint contribution to the fronto-opercular activation observed heresuggests that a participant’s confidence in judging a particular con-dition to be acceptable or unacceptable may be the key factorunderlying this activation. Whereas reaction time differences perse could be attributed to a number of factors (including individualdifferences between ‘‘slow responders” and ‘‘fast responders”, forexample), higher individual per-condition standard deviations foracceptability are clearly indicative of a higher variability in judg-ment behavior – as would be expected under conditions of judg-ment uncertainty. This, in turn, is compatible with longerreaction times. In view of these two sources of converging evi-dence, it appears plausible to associate increased activation inthe frontal operculum that arises in studies using a linguistic judg-ment task (e.g. Grewe et al., 2005) with increased judgment uncer-tainty. Interestingly, the ‘‘output” of this evaluation process,namely the final acceptability value per participant and condition,predicts neither the fronto-opercular activation nor activation inany of the regions within the greater language network. This obser-vation underscores the point raised above that the word order-re-lated activation patterns observed here cannot be explained on thebasis of acceptability differences between the critical conditions,i.e. the activation increases for the OS conditions are not due tothe participants who tended to reject these conditions. The diver-gence between mean acceptability and judgment uncertainty ap-pears natural under the perspective that confidence inperforming the judgment task need not decrease with a decreasein acceptability, i.e. participants can be very confident in rejectingcertain sentence structures. These observations are thus in linewith Grewe et al.’s (2005) finding that the acceptability of a criticalsentence condition correlates neither with the activation pattern ofthe frontal operculum nor with that of the pars opercularis.

A possible conclusion from these findings appears to be that thedeep frontal operculum – as well as the pre-SMA, with which it

136 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

appears to act in concert – provide an interface between linguisticproperties and more general task or decision-related mechanisms.This assumption is corroborated both by the high correlation be-tween fronto-opercular activation and behavioral parameters andby the lower degree of sensitivity of the frontal operculum/pre-SMA to the critical linguistic manipulation employed here: in con-trast to the pars opercularis, these regions did not show an effect ofreferentiality. It is also supported by findings in the literature oftask-demand-related activations within the frontal operculum/anterior insula, which have been reported, for example, for in-creased interference cost (Wager et al., 2005) or (for the right ante-rior insula) as correlating with valence in explicit and implicitevaluative judgments (Cunningham, Raye, & Johnson, 2004).

4.2.3. The pars opercularis of the left IFGWhen all available studies of (clause-medial) word order varia-

tions in German are considered, the pars opercularis of the left IFGstands out as the only region that is consistently activated, inde-pendently of the task or the modality of presentation (visual orauditory) employed. Furthermore, and perhaps even more impor-tantly for present purposes, only the left pars opercularis has pro-ven sensitive to all of the linearization parameters hithertoexamined: subject > indirect object > direct object (Friedericiet al., 2006; Röder et al., 2002); pronoun > non-pronominal argu-ment (Grewe et al., 2005); animate > inanimate (Grewe et al.,2006); higher thematic role > lower thematic role (Bornkesselet al., 2005); and higher referentiality > lower referentiality in thepresent study. It is this sensitivity to all critical word order-relatedfactors (see Lenerz, 1977; and Müller, 1999, for theoretical discus-sion) in combination with the independence from the experimen-tal environment (task, modality) that leads us to view the parsopercularis as the neural region that is at the core of lineariza-tion-related processes during sentence comprehension. While acti-vation in this region is, of course, typically observed as part of alarger network in any given study, the overall makeup of the net-work – with the exception of the pars opercularis – appears largelydetermined by the particular experimental setting in which thecritical sentences are presented.

7 Note that similar considerations serve to exclude an explanation of our findings interms of semantic differences with regard to the interpretation of bare plural NPs indifferent structural positions. According to Diesing’s (1992) account, the bare pluralsshould be ambiguous between an indefinite and a generic reading in the REFconditions (since they could either be within the VP domain or outside of it). Bycontrast, they should unambiguously require a generic reading in the NREF conditionsbecause, here, they precede the proper name, which must be outside the VP domain.(But see Frey, 2001, for problems with a one-to-one mapping between theinterpretation of bare plurals and their position within the German middlefield.)

4.3. Linearization: consequences for pars opercularis function

As outlined above, the left pars opercularis shows a sensitivityto all parameters known to influence argument linearization inGerman. Importantly, the set of information types involved (the-matic roles, animacy, definiteness/specificity, pronouns vs. non-pronominals) is not arbitrary: as noted briefly in the introduction,they have been identified as a major source of cross-linguistic sim-ilarities and differences with respect to a wide range of morpho-syntactic phenomena (e.g. Comrie, 1989; Croft, 2003). Mostgenerally, these factors serve to change the (conceptual) ‘‘promi-nence” status of an argument (Actors are more prominent thanUndergoers, animate entities are more prominent than inanimateentities etc.), with conceptual prominence ideally correspondingto morphosyntactic prominence. For example, a prominent argu-ment is more likely to be realized as a subject rather than an object(cf. Aissen, 1999, and the references cited therein). In addition, andmore importantly for present purposes, prominent argumentspreferentially precede their coarguments in terms of linear order(Tomlin, 1986). The pars opercularis of the left IFG thus appearsto play a crucial role in basic – and universal – processes of linear-ization in language.

The empirical observations in the domain of linearization yielda number of important consequences with respect to the functionof the pars opercularis in the processing of linear order in language.In particular, the overall data pattern poses a challenge to promi-

nent language-specific (syntactic) and domain-general approachesto the function of this cortical region.

Challenges for a syntactic (movement-based) account (cf. Grod-zinsky & Santi, 2008) have already been documented in previousrelated work (e.g. Bornkessel et al., 2005; Grewe et al., 2006,2007) and were briefly outlined in the introduction. The presentfindings serve to further corroborate this perspective. Importantly,existing movement-based accounts of referentiality effects (e.g.Diesing, 1992) assume an influence of referentiality on the surfacestructure of a sentence, but not on the base positions of the argu-ments (i.e. the subject is assumed to be generated in a higher struc-tural position than the object independently of referentiality).Specifically, it is assumed that referential arguments (includingproper names) must move out of the verb phrase domain in orderto be interpretable (cf. Carlson, 2003). Consequently, both REF con-ditions should require a single movement operation (of the refer-ential argument, whether it is subject or object), whereas bothNREF conditions should require two movement operations (thesecond argument, i.e. the proper name, must have moved since itis referential, and, consequently, the first argument must also haveundergone movement in order to ensure its linear precedence).Hence, a movement-based account would only predict a main ef-fect of referentiality and therefore cannot derive the pattern of parsopercularis activation observed here (see also Uszkoreit, 1986).7

Turning now to prominent domain-general approaches, work-ing memory-based accounts encounter difficulties when the line-arization parameter in question does not induce an inherentdependency between two arguments. A dependency between thearguments (e.g. of an initial object upon a following subject) mayclearly lead to increased working memory load because the firstargument must be maintained until the second is encountered.However, increased pars opercularis activation is observable evenwhen no such dependency applies (e.g. in the case of argumentsthat differ in terms of referentiality; see also Grewe et al., 2006,for similar findings involving animacy). Note that possible notionsof working memory related to differences in informational load(Almor, 1999) do not apply here, since no reactivation of anteced-ents was required in the critical sentences.

A second possibility is that the involvement of the pars opercu-laris in the processing of word order variations stems from the in-creased involvement of cognitive control mechanisms in thesestructures (Thompson-Schill et al., 2005). This hypothesis was ap-plied specifically to sentence processing by Novick, Trueswell, &Thompson-Schill (2005), who argue that Broca’s region engagesin the resolution of conflicts among competing stimulus represen-tations and is involved in reanalysis processes. On the one hand,this proposal appears partially compatible with the linear orderphenomena discussed here: pars opercularis activation increaseswhen there is an incompatibility between the different types ofprominence information. However, the proposal that the degreeof conflict should be higher in ambiguous structures appears prob-lematic, as Bornkessel et al. (2005) found very similar interactionsbetween subject-object order and thematic information for locallyambiguous and unambiguous structures. Thus, while main effectsof ambiguity have been observed in the left pars opercularis andthe adjacent pars triangularis, BA 45 (Bornkessel et al., 2005; Fie-bach, Vos, & Friederici, 2004; Stowe, Paans, Wijers, & Zwarts,

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 137

2004; see also Rodd, Davis, & Johnsrude, 2005, for findings on lex-ical/semantic ambiguity), the need for reanalysis does not increasethe activation of this region beyond the level observed for the com-parable (dispreferred) unambiguous structure. Rather, we have ar-gued above that the degree of cognitive control required in wordorder processing modulates the overall neural network withinwhich pars opercularis activation can be observed.

Finally, as the linearization principles at issue here are verb-independent, they also pose a challenge for ‘‘action understand-ing”-based approaches to Broca’s area function (Rizzolatti & Arbib,1998; for a more recent review see Rizzolatti, Fogassi, & Gallese,2002). Here, it is proposed that filler-slot associations betweenarguments and predicates, which may be used to represent motorcommands, can also serve to model language. More specifically,Broca’s area is viewed as representing ‘‘‘verb phrases’ and con-straints on the noun phrases that can fill the slots, but not detailsof the noun phrases themselves” (Rizzolatti & Arbib, 1998, p.192). It is, however, not at all clear how representations of this typecould serve to model linearization principles of the type animate-before-inanimate, definite-before-indefinite and pronoun-before-non-pronominal, which depend entirely on the properties of thearguments themselves.

To summarize, existing domain-general models of pars opercu-laris function cannot easily account for the high sensitivity of thiscortical region to linguistic linearization principles. In this respect,these models turn out to be surprisingly similar to what is arguablythe most specific language-internal approach, namely the syntacticmovement-based account. Importantly, we do not mean to suggestthat linearization effects are principally incompatible with do-main-general mechanisms, but only that – to the best of ourknowledge – there are currently no such models that can derivethe fine-grained data patterns in the domain of word order pro-cessing. In this context, we would like to stress once more thatthe phenomena under consideration here are by no means‘‘exotic”: as flexible argument order is very common cross-linguis-tically (Steele, 1978), the mechanisms examined here are funda-mental to the human ability to comprehend language. Thus, inorder for the full explanatory potential of non-linguistically-ori-ented models of pars opercularis function to become clear, we sug-gest that it would be fruitful for these models to focus less on therather restricted range of constructions that is available in lan-guages of the English type (which can also be accounted for in amovement-based approach) and more on the rich possibilities thatare available cross-linguistically (which cannot).

5. Conclusion

The present study was the first to show word order-related acti-vation differences between subject-initial sentences in the parsopercularis of the left IFG. Specifically, activation in this region in-creased when an argument that was lower in referentiality pre-ceded an argument that was higher in referentiality—a findingwhich cannot be explained on the basis of acceptability differencesbetween the critical sentence types. Together with previous find-ings, this result suggests that the word order-related processingmechanisms subserved by the pars opercularis: (a) apply indepen-dently of any inherent dependency between two arguments, (b)are verb-independent and (c) do not necessarily require increaseddegrees of cognitive control. Furthermore, we have argued that theoverall neural network involved in the processing of linear order inlanguage is determined by a number of factors. While a fronto-parietal (IFJ/IPS) subnetwork comes into play when additional cog-nitive control is required, fronto-opercular/anterior insular corticesand the pre-SMA support the processing of gradient stimuli andlinguistic judgments.

Acknowledgments

Parts of the research reported here were supported by the Ger-man Research Foundation (Grant BO 2471/2-1/SCHL 554/2-1). Theauthors would like to thank Katrin Bock for assistance in stimulusconstruction and Mandy Naumann, Annett Wiedemann and Si-mone Wipper for the data acquisition.

References

Aguirre, G. K., Zarahn, E., & d’Esposito, M. (1997). Empirical analyses of BOLD fMRIstatistics II. Spatially smoothed data collected under null-hypothesis andexperimental conditions. Neuroimage, 5, 199–212.

Aissen, J. (1999). Markedness and subject choice in optimality theory. NaturalLanguage and Linguistic Theory, 17, 673–711.

Aissen, J. (2003). Differential object marking: Iconicity vs. economy. NaturalLanguage and Linguistic Theory, 21, 435–483.

Almor, A. (1999). Noun-phrase anaphora and focus: The informational loadhypothesis. Psychological Review, 106, 748–765.

Ben-Shachar, M., Palti, D., & Grodzinsky, Y. (2004). Neural correlates of syntacticmovement: Converging evidence from two fMRI experiments. Neuroimage, 21,1320–1336.

Bierwisch, M. (1963). Grammatik des deutschen Verbs [=‘‘A grammar of the verb inGerman”]. Berlin: Akademie-Verlag.

Binkofski, F., Amunts, K., Stephan, K. M., Posse, S., Schormann, T., Freund, H.-J., et al.(2000). Broca’s region subserves imagery of motion: A combinedcytoarchitectonic and fMRI study. Human Brain Mapping, 11, 273–285.

Bornkessel, I., McElree, B., Schlesewsky, M., & Friederici, A. D. (2004). Multi-dimensional contributions to garden path strength: Dissociating phrasestructure from case marking. Journal of Memory and Language, 51, 495–522.

Bornkessel, I., & Schlesewsky, M. (2006). The extended argument dependencymodel: A neurocognitive approach to sentence comprehension acrosslanguages. Psychological Review, 113, 787–821.

Bornkessel, I., Schlesewsky, M., & Friederici, A. D. (2003). Eliciting thematicreanalysis effects: The role of syntax-independent information during parsing.Language and Cognitive Processes, 18, 268–298.

Bornkessel, I., Zysset, S., Friederici, A. D., von Cramon, D. Y., & Schlesewsky, M.(2005). Who did what to whom? The neural basis of argument hierarchiesduring language comprehension. NeuroImage, 26, 221–233.

Bornkessel-Schlesewsky, I., & Schlesewsky, M. (2009). The role of prominenceinformation in the real time comprehension of transitive constructions: A cross-linguistic approach. Language and Linguistics Compass, 3, 19–58.

Brass, M., & von Cramon, D. Y. (2002). The role of the frontal cortex in taskpreparation. Cerebral Cortex, 12, 908–914.

Brass, M., & von Cramon, D. Y. (2004). Decomposing components of task preparationwith functional magnetic resonance imaging. Journal of Cognitive Neuroscience,16, 609–620.

Braver, T. S., & Bongiolatti, S. (2002). The role of frontopolar cortex in subgoalprocessing during working memory. Neuroimage, 15, 523–536.

Bresnan, J. (2001). Lexical functional grammar. Oxford: Blackwell.Büchel, C., Wise, R. J. S., Mummery, C. J., Poline, J.-B., & Friston, K. J. (1996). Nonlinear

regression in parametric activation studies. Neuroimage, 4, 60–66.Bunge, S. A., Hazeltine, E., Scanlon, M. D., Rosen, A. C., & Gabrieli, J. D. E. (2002).

Dissociable contributions of prefrontal and parietal cortices to responseselection. Neuroimage, 17, 1562–1571.

Burock, M. A., Buckner, R. L., Woldorff, M. G., Rosen, B. R., & Dale, A. M. (1998).Randomized event-related experimental designs allow for extremely rapidpresentation rates using functional MRI. Neuroreport, 9, 3735–3739.

Caplan, D., Alpert, N., Waters, G., & Olivieri, A. (2000). Activation of Broca’s area bysyntactic processing under conditions of concurrent articulation. Human BrainMapping, 9, 65–71.

Caplan, D., Chen, E., & Waters, G. (2008). Task-dependent and task-independentneurovascular responses to syntactic processing. Cortex, 44, 257–275.

Carlson, G. N. (1977). A unified analysis of the English bare plural. Linguistics andPhilosophy, 1, 413–456.

Carlson, G. (2003). Interpretive asymmetries in major phrases. In A.-M. DiSciullo(Ed.), Asymmetry in linguistic theory (pp. 299–311). Amsterdam: John Benjamins.

Chen, E., West, W. C., Waters, G., & Caplan, D. (2006). Determinants of BOLD signalcorrelates of processing object-extracted relative clauses. Cortex, 42, 591–604.

Choi, H.-W. (1999). Optimizing structure in context: Scrambling and informationstructure. Stanford, CA: CSLI Publications.

Comrie, B. (1989). Linguistic universals and language typology (2nd ed.). Oxford:Blackwell.

Constable, R. T., Pugh, K. R., Berroya, E., Mencl, W. E., Westerveld, M., Ni, W., et al.(2004). Sentence complexity and input modality effects in sentencecomprehension: An fMRI study. Neuroimage, 22, 11–21.

Croft, W. (2003). Typology and universals (2nd ed.). Cambridge: CambridgeUniversity Press.

Culicover, P. W., & Jackendoff, R. (2005). Simpler syntax. Oxford: Oxford UniversityPress.

Cunningham, W. A., Raye, C. L., & Johnson, M. K. (2004). Implicit and explicitevaluation: fMRI correlates of valence, emotional intensity and control in theprocessing of attitudes. Journal of Cognitive Neuroscience, 16, 1717–1729.

138 I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139

Diesing, M. (1992). Indefinites. Cambridge, MA: MIT Press.Fanselow, G. (2000). Optimal exceptions. In B. Stiebels & D. Wunderlich (Eds.),

Lexicon in focus (pp. 173–209). Berlin: Akademie Verlag.Fanselow, G., Féry, C., Vogel, R., & Schlesewsky, M. (Eds.). (2006). Gradience in

grammar: Generative perspectives. Oxford: Oxford University Press.Featherston, S. (2007). Data in generative grammar: The stick and the carrot.

Theoretical Linguistics, 33, 269–318.Fiebach, C. J., Schlesewsky, M., & Friederici, A. D. (2002). Separating syntactic

memory costs and syntactic integration costs during parsing: The processing ofGerman wh-questions. Journal of Memory and Language, 47, 250–272.

Fiebach, C. J., Schlesewsky, M., Lohmann, G., von Cramon, D. Y., & Friederici, A. D.(2005). Revisiting the role of Broca’s area in sentence processing: Syntacticintegration versus syntactic working memory. Human Brain Mapping, 24, 79–91.

Fiebach, C. J., Vos, S. H., & Friederici, A. D. (2004). Neural correlates of syntacticambiguity in sentence comprehension for low and high span readers. Journal ofCognitive Neuroscience, 16, 1562–1575.

Fodor, J. D., & Sag, I. (1982). Referential and quantificational indefinites. Linguisticsand Philosophy, 5, 355–398.

Forman, S., Cohen, J., Fitzgerald, M., Eddy, W. F., Mintun, M., & Noll, D. (1995).Improved assessment of significant activation in functional magnetic resonanceimaging (fMRI): Use a cluster-size threshold. Magnetic Resonance in Medicine,33, 636–647.

Frazier, L., & Flores d’Arcais, G. B. (1989). Filler-driven parsing: A study of gap fillingin Dutch. Journal of Memory and Language, 28, 331–344.

Frey, W. (2001). Über Indefinita und ihre Platzierung im Mittelfeld [= ‘‘Onindefinites and their placement in the middlefield”]. ZAS Papers in Linguistics,21, 33–44.

Friederici, A. D. (2006). What’s in control of language? Nature Neuroscience, 9,991–992.

Friederici, A. D., & Alter, K. (2004). Lateralization of auditory language functions: Adynamic dual pathway model. Brain and Language, 89, 267–276.

Friederici, A. D., Fiebach, C. J., Schlesewsky, M., Bornkessel, I., & von Cramon, D. Y.(2006). Processing linguistic complexity and grammaticality in the left frontalcortex. Cerebral Cortex, 16, 1709–1717.

Friston, K. J., Holmes, A. P., Worsley, K. J., Poline, J. B., Frith, C. D., & Frackowiak, R. W.(1995). Statistical parametric maps in functional imaging: A general linearapproach. Human Brain Mapping, 2, 189–210.

Glover, G. H. (1999). Deconvolution of impulse response in event-related BOLDfMRI. Neuroimage, 9, 416–429.

Goldberg, A. E. (2006). Constructions at work: The nature of generalization in language.Oxford, UK: Oxford University Press.

Gordon, P. C., Hendrick, R., & Johnson, M. (2001). Memory interference duringlanguage processing. Journal of Experimental Psychology: Learning, Memory andCognition, 27, 1411–1423.

Gordon, P. C., Hendrick, R., & Levine, W. H. (2002). Memory-load interference insyntactic processing. Psychological Science, 13, 425–430.

Grewe, T., Bornkessel, I., Zysset, S., Wiese, R., von Cramon, D. Y., & Schlesewsky, M.(2005). The emergence of the unmarked: A new perspective on the language-specific function of Broca’s area. Human Brain Mapping, 26, 178–190.

Grewe, T., Bornkessel, I., Zysset, S., Wiese, R., von Cramon, D. Y., & Schlesewsky, M.(2006). Linguistic prominence and Broca’s area: The influence of animacy as alinearization principle. Neuroimage, 32, 1395–1402.

Grewe, T., Bornkessel-Schlesewsky, I., Zysset, S., Wiese, R., von Cramon, D. Y., &Schlesewsky, M. (2007). The role of the posterior superior temporal sulcus inthe processing of unmarked transitivity. Neuroimage, 35, 343–352.

Grimshaw, J. B. (1990). Argument structure. Cambridge, MA: MIT Press.Grodzinsky, Y. (2000). The neurology of syntax: Language use without Broca’s area.

Behavioral and Brain Sciences, 23, 1–71.Grodzinsky, Y., & Friederici, A. D. (2006). Neuroimaging of syntax and syntactic

processing. Current Opinion in Neurobiology, 16, 240–246.Grodzinsky, Y., & Santi, A. (2008). The battle for Broca’s region. Trends in Cognitive

Sciences, 12, 474–480.Haider, H., & Rosengren, I. (2003). Scrambling: Nontriggered chain formation in OV

languages. Journal of Germanic Linguistics, 15, 203–267.Haupt, F. S., Schlesewsky, M., Roehm, D., Friederici, A. D., & Bornkessel-Schlesewsky,

I. (2008). The status of subject-object reanalyses in the languagecomprehension architecture. Journal of Memory and Language, 59, 54–96.

Holmes, A. P., & Friston, K. J. (1998). Generalisability, random effects and populationinference. Neuroimage, 7, 754.

Just, M. A., Carpenter, P. A., Keller, T. A., Eddy, W. F., & Thulborn, K. R. (1996). Brainactivation modulated by sentence comprehension. Science, 274, 114–116.

Kaan, E., & Swaab, T. Y. (2002). The brain circuitry of syntactic comprehension.Trends in Cognitive Sciences, 6, 350–356.

Keller, F. (2000). Gradience in grammar: Experimental and computational aspects ofdegrees of grammaticality. Unpublished doctoral dissertation, University ofEdinburgh.

Keller, T. A., Carpenter, P. A., & Just, M. A. (2001). The neural bases of sentencecomprehension: A fMRI examination of syntactic and lexical processing.Cerebral Cortex, 11, 223–237.

Kinno, R., Kawamura, M., Shioda, S., & Sakai, K. L. (2008). Neural correlates ofnoncanonical syntactic processing revealed by a picture-sentence matchingtask. Human Brain Mapping, 29, 1015–1027.

Keppel, G. (1991). Design and analysis (3rd ed.). Englewood Cliffs, NJ: Prentice Hall.Kretzschmar, F., Bornkessel-Schlesewsky, I., Staub, A., Roehm, D., & Schlesewsky, M.

(submitted for publication). Prominence facilitates ambiguity resolution: Onthe interaction between referentiality, thematic roles and word order in

syntactic reanalysis. In P. de Swart & M. Lamers (Eds.), Case, word order, andprominence. Psycholinguistic and theoretical approaches to argument structure.Berlin: Springer.

Kroger, J. K., Sabb, F. W., Fales, C. L., Bookheimer, S. Y., Cohen, M. S., & Holyoak, K. J.(2002). Recruitment of anterior dorsolateral prefrontal cortex in humanreasoning: A parametric study of relational complexity. Cerebral Cortex, 12,477–485.

Lee, Y., Lee, H., & Gordon, P. C. (2007). Linguistic complexity and informationstructure in Korean: Evidence from eye-tracking during reading. Cognition, 104,495–534.

Lenerz, J. (1977). Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: GunterNarr Verlag.

Lohmann, G., Müller, K., Bosch, V., Mentzel, H., Hessler, S., Chen, L., et al. (2001).Lipsia – A new software system for the evaluation of functional magneticresonance images of the human brain. Computerized Medical Imaging andGraphics, 25, 449–457.

Miezin, F. M., Maccotta, L., Ollinger, J. M., Petersen, S. E., & Buckner, R. L. (2000).Characterizing the hemodynamic response: Effects of presentation rate,sampling procedure, and the possibility of ordering brain activity based onrelative timing. Neuroimage, 11, 735–759.

Müller, G. (1999). Optimality, markedness and word order in German. Linguistics,37, 777–818.

Müller, R.-A., & Basho, S. (2004). Are nonlinguistic functions in ‘‘Broca’s area”prerequisites for language acquisition? FMRI findings from an ontogeneticviewpoint. Brain and Language, 329, 336.

Müller, R.-A., Kleinhans, N., & Courchesne, E. (2003). Linguistic theory andneuroimaging evidence: An fMRI study of Broca’s area in lexical semantics.Neuropsychologia, 41, 1199–1207.

Norris, D. G. (2000). Reduced power multislice MDEFT imaging. Journal of MagneticResonance Imaging, 11, 445–451.

Novick, J. M., Trueswell, J. C., & Thompson-Schill, S. L. (2005). Cognitive control andparsing: Reexamining the role of Broca’s area in sentence comprehension.Cognitive, Affective and Behavioral Neuroscience, 5, 263–281.

Pechmann, T., Uszkoreit, H., Engelkamp, J., & Zerbst, D. (1994). Word order in theGerman middle field: Linguistic theory and psycholinguistic evidence.Computational Linguistics at the University of the Saarland, 43.

Pickering, M. J., & Barry, G. (1991). Sentence processing without empty categories.Language and Cognitive Processes, 6, 229–259.

Primus, B. (1999). Cases and thematic roles. Tübingen: Niemeyer.Rizzolatti, G., & Arbib, M. A. (1998). Language within our grasp. Trends in

Neurosciences, 21, 188–194.Rizzolatti, G., Fogassi, L., & Gallese, V. (2002). Motor and cognitive functions of the

ventral premotor cortex. Current Opinion in Neurobiology, 12, 149–154.Rodd, J., Davis Matthew, H., & Johnsrude, I. S. (2005). The neural mechanisms of

speech comprehension: fMRI studies of semantic ambiguity. Cerebral Cortex, 15,1261–1269.

Röder, B., Stock, O., Neville, H., Bien, S., & Rösler, F. (2002). Brain activationmodulated by the comprehension of normal and pseudo-word sentences ofdifferent processing demands: A functional magnetic resonance imaging study.Neuroimage, 15, 1003–1014.

Schlesewsky, M., Bornkessel, I., & McElree, B. (2006). Decomposing gradience:Quantitative vs. qualitative distinctions. In G. Fanselow, C. Féry, R. Vogel, & M.Schlesewsky (Eds.), Gradience in grammar: Generative perspectives(pp. 124–142). Oxford: Oxford University Press.

Steele, S. (1978). Word order variations: A typological study. In J. H. Greenberg (Ed.),Universals of human language, Vol. 4, Syntax (pp. 585–623). Stanford, CA:Stanford University Press.

Stowe, L. A., Paans, A. M. J., Wijers, A. A., & Zwarts, F. (2004). Activations of ‘‘motor”and other non-language structures during sentence comprehension. Brain andLanguage, 89(290–299).

Stromswold, K., Caplan, D., Alpert, N., & Rauch, S. (1996). Localization of syntacticcomprehension by positron emission tomography. Brain and Language, 52,452–473.

Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic atlas of the human brain.Stuttgart: Thieme.

Thirion, J. P. (1998). Image matching as a diffusion process: An analogy withMaxwell’s demons. Medical Image Analysis, 2, 243–260.

Thompson-Schill, S. L., Bedny, M., & Goldberg, R. F. (2005). The frontal lobes andthe regulation of mental activity. Current Opinion in Neurobiology, 15, 219–224.

Tomlin, R. S. (1986). Basic word order: Functional principles. London: Routledge(Croom Helm).

Ugurbil, K., Garwood, M., Ellermann, J., Hendrich, K., Hinke, R., Hu, X., et al. (1993).Imaging at high magnetic fields: Initial experiences at $T. Magnetic ResonanceQuarterly, 9, 259–277.

Uszkoreit, H. (1986). Constraints on order. Linguistics, 24, 883–906.Van Valin, R. D. Jr., (2005). Exploring the syntax-semantics interface. Cambridge:

Cambridge University Press.Volz, K. G., Schubotz, R. I., & von Cramon, D. Y. (2003). Predicting events of varying

probability: Uncertainty investigated by fMRI. Neuroimage, 19, 271–280.Wager, T. D., Sylvester, C.-Y. C., Lacey, S. C., Nee, D. E., Franklin, M., & Jonides, J.

(2005). Common and unique components of response inhibition revealed byfMRI. Neuroimage, 27, 323–340.

Wolff, S., Schlesewsky, M., Hirotani, M., & Bornkessel-Schlesewsky, I. (2008). Theneural mechanisms of word order processing revisited: Electrophysiologicalevidence from Japanese. Brain and Language, 107, 133–157.

I. Bornkessel-Schlesewsky et al. / Brain & Language 111 (2009) 125–139 139

Worsley, K. J., & Friston, K. J. (1995). Analysis of fMRI time series revisited – again.Neuroimage, 2, 173–181.

Wunderlich, D. (1997). Cause and the structure of verbs. Linguistic Inquiry, 28,27–68.

Zarahn, E., Aguirre, G. K., & d’Esposito, M. (1997). Empirical analyses of BOLD fMRIstatistics in spatially unsmoothed data collected under null-hypothesisconditions. Neuroimage, 5, 179–197.