The Response of Left Temporal Cortex to Sentences

R. Vandenberghe1, A. C. Nobre2, and C. J. Price3

Abstract Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 & The meaning of a sentence differs from the sum of the response occurred in the left anterior superior temporal meanings of its constituents. Left anterior temporal cortex sulcus and the left posterior middle temporal gyrus. A subset responds to sentences more strongly than to unconnected of voxels within the left anterior temporal pole responded words. We hypothesized that the anterior temporal response more to semantically random sentences and their scrambled to sentences is due to this difference in meaning (composi- versions than to normal sentences and the corresponding tional semantics). Using positron emission tomography scrambled versions (main effect of semantic randomness). (PET), we studied four experimental conditions (2 Â 2 Finally, the grammatical and the semantic factor interacted in factorial design): In one condition, subjects read normal a subset of voxels within the anterior temporal pole: Activity sentences. In a second condition, they read grammatically was higher when subjects read normal sentences compared correct sentences containing numerous semantic violations to their scrambled versions but not for semantically random (semantically random sentences). In a third condition, we sentences compared to their corresponding scrambled scrambled the word order within the normal sentences, and, versions. The effects of grammar and meaning and, most in a fourth condition, the word order was scrambled within importantly, the interaction between grammatical and seman- the semantically random sentences. The left anterior tempo- tic factors are compatible with the hypothesis that the left ral pole responded strongly to sentences compared to anterior temporal pole contributes to the composition of scrambled versions of sentences. A similar although weaker sentence meaning. &

INTRODUCTION et al., 1996). In the current study, we tried to determine By combining words into sentences, we can communi- which brain regions are involved in the composition of cate an unlimited variety of experiences and ideas. The sentence meaning from a syntactic combination of meaning of a sentence is a function of the individual meaningful words (Caplan, 1992; Taraban and McClel- meanings of its constituents and also of the way that land, 1990; Chomsky, 1986). they are syntactically combined (Partee, 1995). Composi- One candidate region is the left anterior temporal tional semantics refers to this difference in meaning pole. Left anterior temporal blood flow increases when between a sentence and an unstructured sequence of subjects listen to grammatically structured sequences of words (Fodor, 1995; Partee, 1995). pseudowords in comparison to a resting control con- Previous neuroimaging studies have defined the neu- dition but not when they listen to a list of unconnected ral substrate for syntactic operations by contrasting words (Mazoyer et al., 1993). It also increases when different types of sentences. They compared, for in- subjects read sentences compared to lists of uncon- stance, sentences with embedded clauses and with nected words (Stowe et al., 1998; Bottini et al., 1994), right-branching clauses (Stromswold, Caplan, Alpert, & nonwords (Bavelier et al., 1997), and fixation or Tamil Rauch, 1996), sentences with cleft-object and with cleft- pseudocharacters (Chee et al., 1999). This is surprising: subject phrases (Caplan, Alpert, & Waters, 1999) or Focal lesions of anterior temporal cortex can give rise to sentences of increasing complexity (Just, Carpenter, naming (Saykin et al., 1995; Hermann, Wyler, Somes, & Keller, Eddy, & Thulborn, 1996). In agreement with Clement, 1994; Devinsky, Perrine, Llinas, Luciano, & patient lesion studies (Zurif, Swinney, Prather, Solomon, Dogali, 1993) or semantic deficits (for review, see Saffran & Bushell, 1993; Caplan, 1992; Mesulam, 1990; Schwartz, & Sholl, 1999) but, to our knowledge, they do not lead Saffran, & Marin, 1980), these experiments successfully to clinically obvious sentence comprehension deficits. implicated the left frontal operculum in processing Recent evidence has implicated the left anterior tempo- syntax (Caplan et al., 1999; Just et al., 1996; Stromswold ral cortex in the processing of meaning. Left anterior temporal cortex is activated during lexical–semantic tasks where subjects have to retrieve proper names 1University Hospital Gasthuisberg, Leuven, 2Oxford Uni- (Gorno-Tempini et al., 1998; Damasio, Grabowski, Tranel, versity, 3University College London Hichwa, & Damasio, 1996) or where they actively

D 2002 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 14:4, pp. 550–560

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 compare the meanings of words or pictures (Vanden- would be higher when subjects read sentences in com- berghe, Price, Wise, Josephs, & Frackowiak, 1996). parison to scrambled word sequences (Mazoyer et al., Word-related responses in anterior temporal cortex 1993). Second, we expected that activity would be high- are affected by the semantic context provided by a er when the open class words were chosen randomly word list (Mummery, Shallice, & Price, 1999; Nobre, within a sentence or a word sequence (Nobre et al., Allison, & McCarthy, 1994). We hypothesized that the 1994). Finally, and most critically, we predicted that the anterior temporal response to sentences is related to effect of grammatical structure would depend upon the the composition of sentence meaning rather than to semantic content of the constituent words. syntactic operations in themselves. At least two previous studies (Ni et al., 2000; Dapretto Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 & Bookheimer, 1999) examined the anatomical relation- RESULTS ship between semantic and syntactic processing. Both The final image smoothness estimate was 9.2 Â 9.4 Â studies used fMRI, a technique that is less sensitive to 3 changes occurring in the anterior temporal pole (Devlin 15.0 mm . The search volume extended from z = À40 et al., 2000). One study compared between sentences mm to z = +68 mm. containing either a syntactic or a semantic anomaly (Ni et al., 2000). Syntactic anomalies activated the left Behavioral Data inferior frontal gyrus compared to semantic anomalies while the opposite was found in the left posterior Behavioral data from five of the positron emission temporal cortex (Ni et al., 2000). In a second study, tomography (PET) subjects were pooled with those subjects made a same–different judgement about the obtained from 10 additional subjects who did not under- meaning of sentences that either differed by the mean- go scanning. These subjects participated in an event- ing of one open class item or by their syntactic structure related potential version of the study where exactly (Dapretto & Bookheimer, 1999). Segregation was found identical materials were used. No data of the remaining in the left inferior frontal gyrus: Activity in Brodmann’s five PET subjects were available due to technical failure. area (BA) 47 was higher when sentences differed by one A repeated-measures 2 Â 2 factorial ANOVA was per- open class item while activity in BA44/45 was higher formed using grammatical structure and semantic ran- when sentences differed by their syntactic structure domness as factors. The behavioral measure was the (Dapretto & Bookheimer, 1999). The semantic condi- target detection response time. A target consisted of the tion in the study by Dapretto and Bookheimer (1999) repetition of two identical words within the visually involved a comparison between the meaning of lexical presented word series. There were no significant main items embedded in sentences. In contrast, we deter- effects. A significant interaction effect was found: Sub- mined which regions are involved in composing the jects responded significantly more slowly when they meaning of a sentence. read semantically random sentences (see Appendix: In this experiment, subjects read two types of sen- sequences of type Gram+SemÀ) compared to the cor- tences: normal sentences and grammatically correct responding scrambled versions (see Appendix: sequen- sentences containing numerous semantic violations ces of type GramÀSemÀ) but not when they read normal (semantically random sentences). The latter type of sentences (see Appendix: sequences of type Gram+ sentences conveyed a message that did not in any way Sem+) compared to the corresponding scrambled ver- correspond to potential real-life events. Nevertheless, sions (see Appendix: sequences of type GramÀSem+) both types of sentences could be written out formally as [F(2,460) = 4.2, p < .01] (Table 1). The behavioral data a sentence tree (Chomsky, 1986) indicating the subject, are in agreement with psychophysical studies indicating theverb,theobject,andsoforth.Foreachofthe that sentence context affects performance of tasks sentences, there was a corresponding scrambled ver- that are directed at the single-word processing level sion. We predicted that left anterior temporal activity (Marslen-Wilson & Tyler, 1980).

Table 1. Behavioral Data

Gram+ Sem+ GramÀSem+ Gram+ SemÀ GramÀSemÀ Gram+Ctrl GramÀCtrl RT (msec) 609 (170) 595 (214) 630 (215) 568 (144) 571 (136) 625 (182) False alarms 0.09 (0.30) 0.41 (0.71) 0.16 (0.45) 0.27 (0.58) 0.47 (0.67) 0.48 (0.74) Omissions 0.04 (0.11) 0.06 (0.22) 0.04 (0.14) 0.08 (0.22) 0.06 (0.14) 0.09 (0.16)

Target detection response times, number of false alarms per condition, and number of omissions divided by number of targets. Abbreviations: RT: reaction time; Gram+Sem+: normal sentences; GramÀSem+: scrambled versions of normal sentences; Gram+SemÀ: semantically random sentences; GramÀSemÀ: scrambled versions of semantically random sentences. Behavioral parameters (means (SD)).

Vandenberghe, Nobre, and Price 551

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 A. Z= -36mm -28mm -20mm -12mm L

L L Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021

B. L

C. L

D. L

Figure 1. (A) Significance map for the contrast between the four experimental conditions and the two control conditions; p < .001 uncorrected. First column: sagittal, coronal, and transverse view (see-through). Second column: transverse sections through the group average MRI at a z level of À36, À28, À20, and À12 mm. (B) Main effect of grammatical structure: sentences minus scrambled word sequences; p < .05 uncorrected masked with the general network depicted in A. (C) Main effect of semantic randomness: semantically random sentences and their scrambled versions minus normal sentences and their scrambled versions; p < .05 uncorrected masked with the general network depicted in A. (D) Interaction effect: [Gram+Sem+ÀGramÀSem+]À[Gram+SemÀ ÀGramÀSemÀ]. Threshold p < .05 uncorrected masked with the general network depicted in A.

552 Journal of Cognitive Neuroscience Volume 14, Number 4

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 Table 2. Four Experimental Conditions Minus Control Conditions

x,y,z Gram+Sem+ GramÀSem+ Gram+SemÀ GramÀSemÀ Z Ext.

L ant. temporal pole À46,4,À20 2.99 0.63 2.67 1.51 5.68 221 À44,6,À28 3.24 0.80 2.53 1.97 5.65 L post. middle temporal g. À62,À54,4 1.59 1.74 2.44 1.24 5.52 128 À52,À54,12 1.74 1.24 1.77 0.80 5.08 L sup. temporal sulcus À66,À16,À8 3.16 1.05 2.01 1.19 5.25 53 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 L frontal operculum À48,22,4 0.92 1.04 2.33 1.26 4.03 14

Four experimental conditions minus the two control conditions. First column: anatomical name. When two peaks occurred within a same cluster, the anatomical name for the cluster is given only once. Second column: Talairach coordinates of the peak activation. Third to sixth column: % blood flow increase for each of the experimental conditions with respect to averaged control conditions. Seventh column: Z scores of the properly weighted contrast of all four experimental conditions to the control conditions. Eighth column: number of 2 Â 2 Â 4mm3 voxels belonging to a given cluster and exceeding p < .001 uncorrected. If a cluster contains two peaks, its extent is provided only once. Bold: p < .05 corrected; plain: p < .001 uncorrected. Abbreviations: g. = gyrus; post. = posterior; ant. = anterior; sup. = superior.

Experimental Minus Control Conditions compared to their scrambled versions (À46,6,À24; Z = 4.08) ( p < .001 uncorrected) (Figure 2A), as well As shown in Figure 1A, the four word-reading conditions as for semantically random sentences compared to activated the left anterior temporal pole, the posterior their corresponding scrambled versions (À42,À6,À20; end of the left middle temporal gyrus, and the middle Z = 3.88) (Figure 2B) ( p < .001 uncorrected). The left third of the left superior temporal sulcus in comparison anterior superior temporal sulcus and the left posterior to the consonant letter string control conditions ( p < .05 middle temporal gyrus were also more active when corrected) (Table 2). The left frontal operculum was also subjects read sentences compared to scrambled word more active at p < .001 uncorrected. sequences, although at a lower significance threshold ( p < .001 uncorrected) (Figure 1B; Table 3). Surprisingly, the region most frequently implicated in Main Effect of Grammatical Structure syntactic operations, the left inferior frontal cortex, was As shown in Figure 1B, sentences activated the left not more active during reading sentences compared anterior temporal pole compared to scrambled word to scrambled word sequences (Figure 1B; Table 2), sequences (main effect of grammatical structure; p < .05 even when the threshold was lowered to p <.05 corrected) (Table 3). This was true for normal sentences uncorrected. The inverse comparison, scrambled word

Table 3. Summary of the Main Effects and Interactions

x,y,z Gram+Sem+ GramÀSem+ Gram+SemÀ GramÀSemÀ Z Ext. Increases due to the presence of grammatical structure L ant. temporal pole À44,À6,À24 1.75 À0.68 2.64 0.31 5.14 290 L sup. temporal sulcus À64,À10,À12 2.77 0.32 1.45 0.25 3.85 38 L post. middle temporal g. À50,À58,12 1.73 0.44 1.53 0.27 3.59 91

Increases due to the presence of semantic randomness L ant. temporal pole À48,À4,À20 1.64 À0.53 2.80 1.19 3.22 33

Interaction between grammatical and semantic factor L ant. temporal pole À40,8,À28 3.26 0.83 2.48 2.28 2.45 33

First three rows: [Gram+Sem+ +Gram+SemÀ]À[GramÀSem+ +GramÀSemÀ]. Fourth row: [Gram+SemÀ+ GramÀSemÀ]À[Gram+Sem+ + GramÀSem+]. Fifth row: [Gram+Sem+ÀGramÀSem+]À[Gram+Sem – ÀGramÀSemÀ]. First column: anatomical name. Second column: Talairach coordinates of the peaks of activation for the main effects and the interaction effect. Third to sixth column: % blood flow difference during each condition compared to the averaged control conditions. Seventh column: Z scores for the contrast mentioned in the heading. Eighth column: number of 2 Â 2 Â 4mm3 voxels reaching the p < .05 uncorrected threshold for the contrast mentioned. Bold: p < .05 corrected; plain: p < .001 uncorrected masked with general activation pattern. Italic: p < .05 uncorrected masked with general activation pattern. Abbreviations: see Table 2.

Vandenberghe, Nobre, and Price 553

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 Figure 2. (A) Normal sentences minus their scrambled versions. (B) Semantically random sentences minus their scrambled versions. (C) Semantically random sentences minus normal sentences. (D) Scrambled versions of semantically random sentences minus scrambled versions of normal sentences. Threshold

p < .05 uncorrected masked Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 with the general network depicted in Figure 1A.

sequences minus sentences, did not yield any signifi- interacted (Figures 1D and 3). The interaction (peak at cant activation. À40,8,À28) was located anteriorly and inferiorly to the voxels that showed a main effect of semantic randomness Main Effect of Semantic Randomness As shown in Figure 1C, the left anterior temporal pole was activated more by semantically random sentences and their scrambled versions than by normal sentences and their scrambled versions ( p < .001 uncorrected) (Figure 1C; Table 3). This was true when scrambled versions of semantically random sentences were compared to scrambled versions of normal sentences (À48,À4,À20, Z = 2.75; Figure 2D) but not when the semantically random sentences were compared to the normal sentences ( p > .05 uncorrected; Figure 2C). Semantically random sentences activated the left posterior middle temporal gyrus (À54,À50,4, Z = 2.98; Figure 2C) compared to normal sentences. The scrambled versions of these sentences did not differ in this region (Figure 2D). The inverse comparison, semantically constrained sequences minus semantically random sequences, did not yield any significant activation. Figure 3. Activity profile in the left anterior temporal pole (À40,8,À28). First bar: normal sentences. Second bar: scrambled Interaction Between Grammar and Meaning versions of normal sentences. Third bar: semantically random sentences. Fourth bar: scrambled versions of semantically random Finally, in a subset of voxels within the left anterior sentences. Mean (±SE) % rCBF difference between the experimental temporal pole, the grammatical and the semantic factor and the averaged control conditions.

554 Journal of Cognitive Neuroscience Volume 14, Number 4

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 (peak at À48,À4,À20) (Figure 1C vs. D). Analysis of the task, subjects view or hear sentences that have either simple main effects revealed that normal sentences one or two possible meanings. Examples of ambiguous activated this subset of voxels more than the corre- sentences are ‘‘The sailors liked the port in the evening,’’ sponding scrambled versions did (Z = 3.52) (Table 3; ‘‘Hortense defended the man she loved with all her Figure 3). This was not the case for semantically random heart’’ (MacKay et al., 1998). In H.M., medial temporal sentencescomparedtotheirscrambledversions cortex was bilaterally ablated together with more lateral ( p > .05 uncorrected) (Table 3; Figure 3). In other damage to the rostralmost aspects of superior and words, sentence structure enhanced blood flow only middle temporal gyri (Corkin, Amaral, Gonzales, when this structure produced a meaningful composi- Johnson, & Hyman, 1997). H.M. was impaired in the tion. If the semantic randomness of the sentence pre- detection of sentence ambiguity. His responses were Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 cluded the composition of a meaningful message, characterized by free associations to sentence constitu- grammatical structure had no effect. ents rather than by a comprehensive interpretation of the full sentence meaning (MacKay et al., 1998). Sen- tence ambiguity detection was also impaired in an addi- DISCUSSION tional set of four patients with medial temporal In closely adjacent and partially overlapping areas with- lobectomies extending into lateral temporal cortex in the left anterior temporal pole, we found a main (Schmolck et al., 2000). Sentence ambiguity detection, effect of grammatical structure (Figure 1B), a main however, was spared in three patients with lesions that effect of semantic randomness (Figure 1C), and an were restricted to medial temporal cortex (Schmolck interaction between the grammatical and the semantic et al., 2000). These neuropsychological data provide factor (Figure 1D) (Table 3). preliminary evidence for a role of lateral temporal cortex The anterior temporal sensitivity to grammatical in the interpretation of meaning at the sentence level. It structure confirms earlier reports of left anterior tem- is currently unclear to what degree this linguistic func- poral activation when sentences are compared to con- tion requires the integrity of the medial temporal cortex trol conditions lacking grammatical structure (Chee (MacKay et al., 1998), a structure to which the lateral et al., 1999; Stowe et al., 1998; Bavelier et al., 1997; temporal cortex sends afferents. The hippocampal for- Bottini et al., 1994; Mazoyer et al., 1993). Its sensitivity mation has a well-documented role as an associative to semantic randomness confirms earlier findings of structure in the formation and consolidation of memo- higher activity during reading of semantically unrelated ries (e.g., Mesulam, 2000; Rolls, 1996). wordsincomparisontosemanticallyrelatedwords The region that showed the predicted interaction effect (Mummery et al., 1999; Nobre et al., 1994). The effect was confined to the anterior and inferior portion of the of grammatical and semantic structure also interacted in anterior temporal region that showed a main effect of the anterior temporal pole. The interaction effect was in sentence structure (compare Figure 1B–D). As Figure 2A accordance with our a priori hypothesis that the ante- and B showed, a significant portion of anterior temporal rior temporal pole is involved in compositional seman- voxels responds to grammatical structure regardless of tics. We propose that the anterior temporal pole semantic randomness. In these voxels, a response to mediates between the representation of a sentence grammatical structure does not require the composition and the distributed activity that corresponds to the of a meaningful message. Strictly speaking, these voxels meaningful content of this sentence. This distributed could respond to grammatical structure for reasons activity differs from the sum of the distributed activities other than what can be attributed solely to the compo- that would be evoked by the individual sentence con- sition of sentence meaning. stituents (principle of compositionality). The anterior A second region, the left posterior temporal gyrus, was temporal role in composing meaning may extend be- also activated when subjects read sentences instead of yond single sentence comprehension: The left anterior lists of unconnected words, in accordance with earlier temporal pole is also activated when subjects read a studies (Stowe et al., 1998; Just et al., 1996). The left story compared to unconnected sentences (StGeorge, posterior temporal gyrus was also sensitive to the pres- Kutas, Martinez, & Sereno, 1999; Perani et al., 1998; ence of semantic anomaly in a sentence (Figure 2B and C), Fletcher et al., 1995). Taken together, these neuro- confirming an earlier report by Ni et al. (2000). A nearby imaging data suggest that the left anterior temporal or identical area has been activated during picture pole may have a role in binding an ensemble of mean- naming (Martin, Haxby, Lalonde, Wiggs, & Ungerleider, ingful components into one message. 1995; Martin, Wiggs, Ungerleider, & Haxby, 1996). A Our hypothesis leads to the prediction that damage of meta-analysis of functional imaging studies of picture the anterior temporal pole would impair sentence com- naming, word reading, or verbal fluency has implicated prehension. Medial temporal lobe excisions that extend this region in lexical phonological code retrieval (Levelt into the lateral rostral temporal cortex affect the detec- & Indefrey, 2000). The effect of sentence structure tion of sentence ambiguity (Schmolck, Stefanacci, & therefore apparently occurs in a region that is principally Squire, 2000; MacKay, Stewart, & Burke, 1998). In this dealing with single-word processing. This could be seen

Vandenberghe, Nobre, and Price 555

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 as an illustration of the distributed and parallel nature of (Caplan et al., 1999; Dapretto & Bookheimer, 1999; Just language processing (Bavelier et al., 1997; Marslen- et al., 1996; Stromswold et al., 1996). Wilson & Tyler, 1980). There could, however, be an alternative explanation. According to a study by Pinango and Zurif (in press), Wernicke aphasics show a deficit in METHODS sentence comprehension beyond their deficit in lexical– Subjects semantic retrieval. They have difficulties with sentences such as ‘‘The boy began the book’’ and ‘‘The girl Ten subjects aged between 27 and 72 years participated. jumped until dawn.’’ In these sentences, the lexical They were strictly right handed, had no neurological or meaning of the sentence constituents must be adapted psychiatric history, and had a normal brain magnetic Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 to the sentence context so that the final meaning of the resonance image (MRI). They gave their written in- sentences are ‘‘The boy began reading the book’’ and formed consent in accordance with the Declaration of ‘‘The girl jumped repeatedly until dawn.’’ This neuro- Helsinki. The effective dose equivalent for the PET scans psychological report may suggest that the left posterior was <5.0 mSv as approved by the Administration of middle temporal gyrus might have a role in the inter- Radioactive Substances Advisory Committee of the De- pretation of sentence meaning beyond its role in lexical partment of Health, UK. The experiment was approved semantics (Pinango & Zurif, 2001). by the joint ethical committee of the Institute of Neu- Activation in a third region, the anterior superior rology and the National Hospital for Neurology and temporal sulcus (À66,À16,À8), is close to an area Neurosurgery, London, UK. (À54,+6,À16) that is activated when subjects listen to intelligible sentences compared to spectrally rotated Stimuli and Task speech (Scott, Blank, Rosen, Wise, 2000). It is also relatively close to an area (À58,À30,0) that is activated when Word stimuli were presented on a 43-cm monitor sus- subjects perform an explicit semantic task with written pended on a movable gantry at a distance of approx- words but less so with pictures (Vandenberghe et al., imately 50 cm. Capital letters were shown in white upon 1996). Syntactic impairment of comprehension is seen a black background. Words were displayed one by one most consistently with lesions of the anterior part of left by means of serial visual presentation. Each stimulus was superior temporal gyrus (including BA22), according to a displayed for 325 msec. A 225 ± 100 msec blank interval study by Dronkers, cited by Hagoort, Brown, and Osterh- separated successive stimuli. The timing was relatively out (1999), even more consistently so than with inferior long because we intended to carry out a parallel electro- frontal lesions. The area of activation in our study falls physiological experiment using identical stimuli and within this zone of lesion overlap described by Dronkers timings. During each condition, 11 sequences of 5–11 (Hagoort et al., 1999). stimuli were presented. The last word of each word The left frontal operculum has been activated in a sequence was marked by a period signifying full stop. A number of functional imaging studies of sentence com- 2500-msec interval separated consecutive sequences. We prehension (Ben-Shachar, Hendler, Kahn, Ben-Bashat, & opted for word-by-word presentation (Kutas & Petten, Grodzinsky, 2001; Ni et al., 2000; Caplan et al., 1999; Just 1994) because presentation of a full word sequence on et al., 1996, Stromswold et al., 1996). As mentioned the screen could elicit eye movement patterns that before, these studies compared different sentence types. could differ depending on whether the sequences were None of the sentences in the current study belonged to sentences or scrambled sequences. the sentence types responsible for inferior frontal acti- Subjects held a key in their right hand. They were vation in these studies, although some grammatically instructed to press the key when two identical stimuli more complex features were present, such as pronoun followed each other immediately without intervening agreement for number and gender. The above studies, stimuli. Such a repetition occurred rarely (three times in which compared sentences that differed along one a 90-sec scanning period on average). This explicit syntactic dimension, provide strong evidence for the monitoring task did not require sentence verification role of the left inferior frontal gyrus in syntactic process- or comprehension. Thus responses to grammatical ing. This contrasts with the anterior temporal region, structure or semantic coherence occurred implicitly. which is activated when sentences are compared to Had the task required active processing of the sequence nonsentences (Stowe et al., 1998; Bavelier et al., 1997; as a whole rather than single words, this might have led Mazoyer et al., 1993) (Figure 1B). to strategic differences between the sentences and the To conclude, we propose that the left anterior tem- scrambled word sequences. For instance, subjects could poral cortex has a role in the composition of sentence have intentionally imposed a grammatical or semantic meaning binding together meaningful constituents into structure upon the scrambled word sequences in order one message. This function is distinct from that of to enhance encoding. the left inferior frontal gyrus, which previous studies The experimental design consisted of four experimen- have shown is involved in specific syntactic operations tal conditions and two control conditions, each condition

556 Journal of Cognitive Neuroscience Volume 14, Number 4

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 being scanned twice. Experimental conditions differed positioned by use of laser alignment beams. A trans- along two dimensions: presence or absence of grammat- mission scan with a Germanium–Gallium source was ical structure, and absence or presence of semantic taken to verify position and correct for attenuation. randomness. Some representative examples are given Each subject was scanned twice in each condition. For 15 in the Appendix. each condition, approximately 9 mCi O-labelled H2O In the first condition, we used normal sentences (see was flushed by an automatic pump into the subject via Appendix: Gram+Sem+) taken from a pool of affirmative an intravenous cannula over a period of 20 sec. The task sentences used in event-related potential studies by Ne- was initiated manually by the experimenter approxi- ville, Nicolas, Barss, Forster, and Garrett (1991) and by mately 5 sec before the intracranial radioactivity count Hagoort, Brown, and Groothusen (1993) (after trans- rate started to rise. Acquisition started automatically as Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 lation). Maximally, four thematic roles were assigned soon as the intracranial radioactivity count exceeded a within a sentence. No subordinate phrases or conjoint threshold equal to four times the intracranial back- sentences were used. Each sentence conveyed a plausi- ground radiation. The first 90 sec of image acquisition ble message. The messages were unrelated to each other. were used for further analysis. The task itself continued In the second condition, we scrambled randomly the for approximately 50–60 sec after onset of acquisition. word order within each normal sentence (see Appendix: An interval of at least 8 min separated two successive GramÀSem+). This manipulation disrupted the gram- injections. The order between the six conditions was matical structure but preserved the semantic relation- counterbalanced across subjects as much as possible ships between the individual open class words. with 10 subjects and 6 conditions. The brain tissue In the third condition, we preserved the grammatical radiation count rate was used as a measure of regional structure but filled in the open class items pseudoran- cerebral blood flow. T1-weighted, structural MRI scans domly (semantically random sentences; see Appendix: were performed on a Siemens 2 Tesla Magnetom Vision Gram+SemÀ). We started from the set of normal sen- scanner (Erlangen, Germany). tences. We pseudorandomly replaced the open class items from a given normal sentence with open class Image Analysis items taken from other normal sentences within the set. Items shared the grammatical category of the original The analysis was based on pooled data from all 10 item, verbs of the main clauses were kept in place and subjects. Calculations and image manipulation were subcategorization rules (Chomsky, 1986) were obeyed. carried out on Sun SPARC computers (Sun Microsys- In the fourth condition, we scrambled randomly the tems, Mountain View, CA) using Matlab (Mathworks, word order within each semantically random sentence Sherborn, MA). For data preprocessing and for statistical (see Appendix: GramÀSemÀ). analysis, Statistical Parametric Mapping (Wellcome De- In two additional control conditions, we replaced partment of Cognitive Neurology, London, UK: http// vowels of open class items with consonants according www.fil.ion.ucl.ac.uk/spm) version SPM96 was used to a fixed vowel-to-consonant mapping rule. Closed class (Friston, Ashburner, et al., 1995; Friston, Holmes, et al., items were preserved. In one control condition, the 1995). Images of all conditions were available for each closed class items remained in the same positions as subject. Scans from each subject were realigned using the those held in the original sentences. In the second first image as a reference (Friston, Ashburner, et al., control condition, word order was scrambled. 1995) and normalized to the Montreal Neurological A subject never saw the same word sequence in differ- Institute PET template in the Talairach space (Friston, ent versions. For a given word sequence, each subject Ashburner, et al., 1995; Talairach & Tournoux, 1988) with saw either its original format (i.e., a normal sentence), a voxel sizes set at 2 Â 2 Â 4mm3. Images were smoothed scrambled version of the normal sentence (lacking gram- with a Gaussian filter of 16 Â 16 Â 16 mm3. Data were matical structure), a semantically random sentence, or a analyzed using a fixed-effects model with a randomized scrambled version of this semantically random sentence block design with global brain activity as subject-specific (lacking grammatical structure). Stimuli were counter- covariate of no interest fixed at 50 ml/dl/min (Friston, balanced between conditions across subjects. Holmes, et al., 1995). Subject-specific effects, as well as their interactions with condition effects, were modelled explicitly (SPM96 option multistudy design, one subject Image Acquisition per study). Each subject underwent 12 brain scans performed on a First, we compared the four experimental conditions Siemens/CPS EXACT HR+ PET scanner (CTI, Knoxville, to the two control conditions. Second, we compared TN) (voxel size 2.1 Â 2.1 Â 2.1 mm3; transaxial spatial the normal and the semantically random sentences to resolution 6.4 mm full width of half maximum (FWHM) their scrambled versions (main effect of grammatical at 10 mm offset; axial spatial resolution 4.2 mm FWHM structure). Third, we compared the normal sentences at 0 mm offset; axial field of view: 15.1 cm). The and their scrambled versions to the semantically ran- subject’s head was immobilized with a helmet and dom sentences and their scrambled versions (main

Vandenberghe, Nobre, and Price 557

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 effect of semantic randomness). Finally, we defined the Examples of Semantically Random Sentences interactions between the presence of grammatical struc- (Gram+SemÀ) ture and the presence of semantic randomness. All 1. THE BRANCHES OF THE TRAFFIC HARMED THE inference was based on a conjunction analysis: Activa- SHOPKEEPER’S FURTHER TEA. tions were considered significant only when they were 2. A DUTCH CHAP STAYS WITH THE SKATERS IN present across 10 subjects in the absence of a significant THE EXPENSIVE BLISTER. interaction between conditions and subjects (Price & 3. THE NETWORKS DREAMT ABOUT A REACTION. Friston, 1997). The significance threshold was p < .05 4. THE EXTENSIVE PRIZE IS CAREFULLY LOCKED corrected for the entire brain search volume. We de-

IN THE DOCUMENTARY. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 fined a general activation pattern by subtracting the two 5. YOUTHS RESENTED A SKETCH OF THE FOREST. control conditions from the four experimental condi- 6. MY GRANDMOTHER SEARCHED FOR SAND- tions ( p < .001 uncorrected). We subsequently exam- WICHES IN THE INTERNATIONAL GUN. ined within this activation pattern which voxels showed 7. THE CIRCUS SCHOOL DECIDES TO OPERATE main or interaction effects at a significance threshold of ON THE ADVERTISING OFFICER. p < .05 uncorrected (masking). 8. MY RATS PREPARED WAVES. 9. THE DISCUSSION MANAGED TO AVOID THE RANSOM JUST IN TIME. APPENDIX 10. A WEEPING CAR CAR ARRIVES AT THE GIRL. Examples of Normal Sentences (Gram+Sem+) 11. CASES COST MORE MUSHROOMS THAN EX- 1. THE BABY IS STAYING WITH HIS GRAND- PECTED. MOTHER. 2. HARDLY ANY SURVIVORS WERE FOUND AFTER Examples of Scrambled Versions of 11 THE EARTHQUAKE. Semantically Random Sentences (GramÀSemÀ) 3. THE SHOPKEEPER DESCRIBES HER EXPERI- ENCE WITH THE WASHING PRODUCT. 1. THE DO THE ACCEPT SHOPKEEPER NAPKIN. 4. MY MOTHER KNIT A TIE. 2. OF RESULTS THE NEED SWIMMER THE 5. THE KIDS ENJOY STORIES ABOUT THE FARM. RESERVE. 6. OUR NEIGHBOURS CELEBRATED THEIR MAR- 3. FOR PRAY BAD THE ZOO WASH PRESENT RIAGE IN A HOTEL. THEIR. 7. THE LION ATTACKED THE CIRCUS ARTIST 4. STATION EAGER TO BE THE LIVE. DURING A PUBLIC PERFORMANCE. 5. THE PICTURE HOUSE RECENTLY THE IN ELECT 8. THE LADY SOLD A PORTRAIT OF HER FATHER. PUPIL PUT. 9. THE TERRIER IS BARKING AT THE POSTMAN. 6. LAND BEDROOM PARISH THE CITY NEAR THE. 10. THE TEAMS WERE PLAYING INDOORS BECAUSE 7. PEACE TEACHER IN SEE PEOPLE RUNNER OF. OF BAD WEATHER. 8. THE LITTLE WEALTH OPERATION AT STATUE 11. PICTURES OF THE SUSPECT CIRCULATE WITH- BLUSH MY. IN THE SECURITY SERVICE. 9. WOMAN THE OF A SHIP PRINT THE SUMMARY. 10. DONKEY TERRIER MENU THE HIS THE UNDER Examples of Scrambled Versions of 11 Normal BE TRAINER HIDE. Sentences (GramÀSem+) 11. CRIME OF CLOSE THE EXERCISE BE BECAUSE. 1. THROUGH AFRICAN SAVANNA OF ELEPHANTS STEP THE TROOPS. Examples of Control Stimuli with Preserved 2. TO BIRDS IN RESERVE THE LIKE BREED. Positions of Closed Class Items 3. DISH MY GRANDMOTHER A NEXT LAY HER NAPKIN TO FOLD. 1. CCRH RMNNW WILL BE LSPLCSHED VSVWN. 4. OF BUY OCEAN DIRECTOR A PAINTING THE. 2. THE MRCMNNTS ABOUT THE RCDWW 5. HER EYES DURING CLOSE THE SHE CEREMONY. MDZCRED THE TRSNF. 6. AT THE BE GRADE STORIES WAR STUDENTS 3. THE MRWZN’S NMRTXRNZL CNZRTXRN YAWN THE FOURTH. RSSRPXSES HER BSSHZND. 7. WITH DISAPPEAR PROVISIONS THE DONKEY 4. THE BRM VCGES HIS LCRGDRCNNF A WZHCT ON HIS BACK. AS A TRNSNNP. 8. RETIRE THE IN LIVE HOUSE BE FARMER THIS. 5. WNLR NRRCED RCSCVRRS CDVE FROM THE 9. PAPERS RESULTS INVESTIGATION CONTAIN RRCNRCNHRS. THE OF THE. 6. THE ZHCRLRL ZTDCCD SNFRSES TO KRCND 10. UPSET OF WOMEN THE MICE SIGHT THE. SNSL. 11. SELL A A BE INVALUABLE DUCHESS CHAIN THE. 7. THE RRCWS SNNT IN THE CZSLTN TRRNW.

558 Journal of Cognitive Neuroscience Volume 14, Number 4

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 8. THE GCLLZGV RNZWS CMNRCZZ DURING THE sentences in Mandarin and English studied with fMRI. LNVZLSZBCN HPMMCS. Neuron, 23, 127–137. Chomsky, N. (1986). Knowledge of language: Its nature, 9. THE DCWRW ZKSED FOR ZVDCCN ABOUT origin, and use. Westport, CT: Praeger. XZTES. Corkin, S., Amaral, D., Gonzalez, R., Johnson, K., & Hyman, B. 10. THE GSDJN DNZR AN ZRLCCTN ABOUT RCCMN. (1997). H.M.’s medial temporal lobe lesion: Findings from 11. THE BCSRDNTCNND CPZH LPZPS IN HIS MRI. Journal of Neuroscience, 17, 3964–3979. DNBRRRM WITH THE TZC. Damasio, H., Grabowski, T., Tranel, D., Hichwa, R., & Damasio, A. (1996). A neural basis for lexical retrieval. Nature, 380, 499–505. Dapretto, M., & Bookheimer, S. (1999). Form and content: Examples of Control Stimuli with Scrambled Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 Positions of Closed Class Items Dissociating syntax and semantics in sentence comprehension. Neuron, 24, 427–432. 1. THE RNSCRSSLY SRHSN DURING NCDS THE CLL Devinsky, O., Perrine, K., Llinas, R., Luciano, D., & Dogali, M. KZLN. (1993). Anterior temporal language areas with early onset 2. LPZP A NNNNCVG ZRDHNNCN THE RRSTNCHRZ of temporal lobe epilepsy. Annals of Neurology, 34, 727–731. TO DNSPNRTZP RNZNLY NMTPP. Devlin, J., Russell, R., Davis, M., Price, C., Wilson, J., Moss, H., 3. RRSNTCW AT MCZ RSRVCVRSS MRNP HIS THE Matthews, P., & Tyler, L. (2000). Susceptibility-induced loss THE. of signal: Comparing PET and fMRI on a semantic task. 4. BRRKE TO TSZTCRN WZM ON NCWWRDS Neuroimage, 11, 589–600. WZCLRZM LRRHCGZNS THE THEIR. Fletcher, P., Happe, F., Frith, U., Baker, S., Dolan, R., Frackowiak, R., & Frith, C. (1995). Other minds in the brain: 5. ON A HIS THE RNTSZCN LBCSTNR SRLN RTNZT A functional imaging study of ‘‘theory of mind’’ in story KCHNR. comprehension. Cognition, 57, 109–128. 6. LVWL TZSP MY GRSGHT MCRF FOR OUR. Fodor, J. (1995). Comprehending sentence structure. In 7. LVWK NMPLRWMMS THE AT THE IN RNZHTNW D. Osherson & M. Liberman (Eds.), Language: An invitation FCNCSH THE. to cognitive science (pp. 209–246). Cambridge: MIT Press. Friston, K., Ashburner, J., Frith, C., Poline, J., Heather, J., & 8. CTZSM-SZLES HIS TNSPNR THE RBP BXSZDLE. Frackowiak, R. (1995). Spatial realignment and normaliza- 9. TFCNH TFCNH PNSCRCDTCRN THE LRPCCN tion of images. Human Brain Mapping, 2, 165–189. CCRCSTZLE OF THE A. Friston, K., Holmes, A., Worsley, K., Poline, J., Frith, C., 10. STRRPSNNM THEIR ZVDCCN ZCCNTP ABOUT Heather, J., & Frackowiak, R. (1995). Statistical parametric CNSSRZNCN THE. maps in functional imaging: A general approach. Human Brain Mapping, 2, 189–210. 11. THE SCP CRTLNCL GWRR THE NFRCCZZ SPCH. Gorno-Tempini, M., Price, C., Josephs, O., Vandenberghe, R., Cappa, S., Kapur, N., & Frackowiak, R. (1998). The neural systems sustaining face and proper-name processing. Brain, Reprint requests should be sent to Rik Vandenberghe, M.D., 121, 2103–2118. Ph.D., Neurology Department, University Hospital Gasthuis- Hagoort, P., Brown, C., & Groothusen, J. (1993). The syntactic berg, Herestraat 49, 3000 Leuven, Belgium, or via e-mail: positive shift (SPS) as an ERP measure of syntactic [email protected]. processing. Language and Cognitive Processes, 8, 439–483. Hagoort, P., Brown, C., & Osterhout, L. (1999). The neurocognition of syntactic processing. In C. Brown & P. Hagoort (Eds.), The neurocognition of language REFERENCES (pp. 273–316). New York: Oxford University Press. Hermann, B., Wyler, A., Somes, G., & Clement, L. (1994). Bavelier, D., Corina, D., Jezzard, P., Padmanabhan, S., Clark, V., Dysnomia after left another anterior temporal lobectomy Karni, A., Prinster, A., Braun, A., Lalwani, A., Rauschecker, J., without functional mapping: Frequency and correlates. Turner, R., & Neville, H. (1997). Sentence reading: A Neurosurgery, 35, 52–57. functional MRI study at 4 Tesla. Journal of Cognitive Just, M., Carpenter, P., Keller, T., Eddy, W., & Thulborn, K. Neuroscience, 9, 664–686. (1996). Brain activation modulated by sentence compre- Ben-Shachar, M., Hendler, T., Kahn, I., Ben-Bashat, D., & hension. Science, 274, 114–116. Grodzinsky, Y. (2001). Grammatical transformations Kutas, M., & van Petten, C. (1994). Psycholinguistics electrified: activate the language regions: An fMRI study. Neuroimage, Event related brain potential investigations. In M. Gernsba- 13, S505. cher (Ed.), Handbook of psycholinguistic (pp. 83–143). San Bottini, G., Corcoran, R., Sterzi, R., Paulesu, E., Schenone, P., Diego: Academic Press. Scarpa, P., Frackowiak, R., & Frith, C. (1994). The role of the Levelt, W., & Indefrey, P. (2000). The speaking mind/brain: right hemisphere in the interpretation of figurative aspects Where do spoken words come from? In A. Marantz, Y. of language: A positron emission tomography study. Brain, Miyashita, & W. O’Neil (Eds.), Imaging, language, brain 117, 1241–1253. (pp. 77–92). Cambridge: MIT Press. Caplan, D. (1992). Language: Structure, processing and MacKay, D., Stewart, R., & Burke, D. (1998). H.M. revisited: disorders (pp. 63–104). Cambridge: MIT Press. Relations between language comprehension, memory Caplan, D., Alpert, N., & Waters, G. (1999). PET studies of and the hippocampal system. Journal of Cognitive syntactic processing with auditory sentence presentation. Neuroscience, 10, 377–394. Neuroimage, 9, 343–351. Marslen-Wilson, W., & Tyler, L. (1980). The temporal structure Chee, M., Caplan, D., Soan, C., Sicram, N., Tan, E., Thiel, T., of spoken language understanding. Cognition, 8, 1–71. & Weekes, B. (1999). Processing of visually presented Martin, A., Haxby, J., Lalonde, F., Wiggs, C., & Ungerleider, L.

Vandenberghe, Nobre, and Price 559

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021 (1995). Discrete cortical regions associated with knowledge Rolls, E. (1996). A theory of hippocampal function in memory. of color and knowledge of action. Science, 270, 102–105. Hippocampus, 6, 601–620. Martin, A., Wiggs, C., Ungerleider, L., & Haxby, L. (1996). Saffran, E. M., & Sholl, A. (1999). Clues to the functional Neural correlates of category-specific knowledge. Nature, and neural architecture of word meaning. In C. Brown & 379, 649–652. P. Hagoort (Eds.), The neurocognition of language Mazoyer, B., Tzourio, N., Frak, V., Syrota, A., Murayama, (pp. 241–272). New York: Oxford University Press. Levrier, O., Salamon, G., Dehaene, S., Cohen, L., & Mehler, J. Saykin, A., Stafiniak, P., Robinson, L., Flannery, K., Gur, R., (1993). The cortical representation of speech. Journal of O’Connor, M., & Sperling, M. (1995). Language before and Cognitive Neuroscience, 5, 467–479. after temporal lobectomy: Specificity of acute changes and Mesulam, M. (1990). Large-scale neurocognitive networks and relation to early risk factors. Epilepsia, 36, 1071–1077. distributed processing for attention, language and memory. Schmolck, H., Stefanacci L., & Squire, L. (2000). Detection Annals of Neurology, 28, 597–613. and explanation of sentence ambiguity are unaffected by Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/4/550/1757525/08989290260045800.pdf by guest on 18 May 2021 Mesulam, M. (2000). Behavioral neuroanatomy. In M. M. hippocampal lesions but are impaired by larger temporal Mesulam (Ed.), Principles of behavioral and cognitive lobe lesions. Hippocampus, 10, 759–770. neurology (pp. 1–120). New York: Oxford University Press. Schwartz, M., Saffran, E., & Martin, O. (1980). The word order Mummery, C., Shallice, T., & Price, C. (1999). Dual-process problem in agrammatism: II. Comprehension. Brain and model in semantic priming: A functional imaging Language, 10, 249–262. perspective. Neuroimage, 9, 516–525. Scott, S., Blank, C., Rosen, S., & Wise, R. (2000). Identification Neville, H., Nicolas, J., Barss, A., Forster, K., & Garrett, M. of a pathway for intelligible speech in the left temporal lobe. (1991). Syntactically based sentence processing classes: Brain, 123, 2400–2406. Evidence from event-related brain potentials. Journal of StGeorge, M., Kutas, M., Martinez, A., & Sereno, M. (1999). Cognitive Neuroscience, 3, 151–165. Semantic integration in reading: Engagement of the right Ni, W., Constable, R., Mencl, W., Pugh, K., Fulbright, R., hemisphere during discourse processing. Brain, 122, Shaywitz, S., Shaywitz, B., Gore, J., & Shankweiler, D. (2000). 1317–1325. An event-related neuroimaging study distinguishing form Stowe, L., Broere, C., Paans, A., Wijers, A., Mulder, G., Vaalburg, and content in sentence processing. Journal of Cognitive W., & Zwarts, F. (1998). Localizing components of a Neuroscience, 12, 120–133. complex task: Sentence processing and working memory. Nobre, A., Allison, T., & McCarthy, G. (1994). Word recognition NeuroReport, 9, 2995–2999. in the human inferior temporal lobe. Nature, 372, 260–263. Stromswold, K., Caplan, D., Alpert, N., & Rauch, S. (1996). Partee, B. (1995). Lexical semantics and compositionality. Localization of syntactic comprehension by positron In L. Gleitman & M. Liberman (Eds.), An invitation to emission tomography. Brain and Language, 52, 452–473. cognitive science: Language (pp. 311–360). Cambridge: Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic MIT Press. atlas of the human brain. New York: Thieme. Perani, D., Paulesu, E., Galles, N., Dupoux, E., Dehaene, S., Taraban, R., & McClelland, J. (1990). Parsing and comprehen- Bettinardi, V., Cappa, S., Fazio, F., & Mehler, J. (1998). The sion: A multiple-constraint view. In D. Balota, G. F. d’Arcais, bilingual brain: Proficiency and age of acquisition of the & K. Rayner (Eds.), Comprehension processes in reading second language. Brain, 21, 1841–1852. (pp. 231–263). Mahwah, NJ: Erlbaum. Pinango, M., & Zurif, E. (2001). Semantic operations in aphasic Vandenberghe, R., Price, C., Wise, R., Josephs, O., & Fracko- comprehension: Implications for the cortical organization of wiak, R. (1996). Functional anatomy of a common semantic language. Brain and Language, 79, 297–308. system for words and pictures. Nature, 383, 254–256. Price, C., & Friston, K. (1997). Cognitive conjunction: A new Zurif, E., Swinney, D., Prather, P., Solomon, J., & Bushell, C. approach to brain activation experiments. Neuroimage, 5, (1993). An on-line analysis of syntactic processing in Broca’s 261–270. and Wernicke’s aphasia. Brain and Language, 45, 448–464.

560 Journal of Cognitive Neuroscience Volume 14, Number 4

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/08989290260045800 by guest on 26 September 2021