PUBLISHED MANUSCRIPT Bailer, C., Tomitch, L.M.B. (2015). Syntactic processing: Insights from behavioral and neuroimaging studies. In A.C. Naschold, A. Pereira, R. Guaresi, V.W. Pereira (Eds.), Aprendizado da leitura e da escrita: a ciência em interfaces (pp.21-48). Natal, RN, Brazil: Edufrn. SYNTACTIC PROCESSING: INSIGHTS FROM BEHAVIORAL AND NEUROIMAGING STUDIES Cyntia Bailer1 Lêda Maria Braga Tomitch2 It is commonsensical that the ability to process language, and more precisely the ability to comprehend sentence structures, is considered to be inherently human. Reading comprehension depends, of course, on understanding the meaning of individual words, but essentially on understanding the relationship between the words in a sentence, in a paragraph, and in discourse as a whole. These relationships are central because “they signal who is doing what to whom in the sentence. They are indicated by syntactic structures whose basic architecture is considered to be hierarchical” (FRIEDERICI, 2004, p.789). This hierarchy comes from the fact that words must be put into a specific order and inflection (ANDERSON, 2011) so that we can comprehend what is being said or written. According to Ferreira and Engelhardt (2006, p.61), 1 Cyntia Bailer é doutoranda no Programa de Pós-Graduação em Inglês: Estudos Linguísticos e Literários da Universidade Federal de Santa Catarina, com mestrado no mesmo programa (Working memory capacity and attention to form and meaning in EFL reading, 2011). Neste momento, está fazendo doutorado sanduíche nos EUA, no Center for Cognitive Brain Imaging, da Universidade de Carnegie Mellon, com orientação da professora doutora Lêda Maria Braga Tomitch (UFSC) e co-orientação do professor doutor Marcel Just (CMU). Interessa-se por estudos psicolinguísticos e aplicados na área de compreensão e produção escritas, bem como de aquisição de segunda língua, e em particular, na implementação de processos cognitivos no cérebro. Suas publicações incluem artigos em periódicos tais como Revista Letras de Hoje, Revista Intercâmbio, Revista Ilha do Desterro, Revista Letrônica e Revista BELT (Brazilian English Language Teaching Journal). Contato: cyntiabailer@gmail.com 2 Lêda Maria Braga Tomitch é professora associada IV do Departamento de Língua e Literatura Estrangeiras da Universidade Federal de Santa Catarina. Tem mestrado e doutorado em linguística aplicada pelo Programa de Pós-Graduação em Inglês: Estudos Linguísticos e Literários da UFSC, tendo realizado pós-doutorado na área de psicologia cognitiva, na Universidade de Carnegie Mellon - EUA. Interessa-se por estudos na área da compreensão e produção escritas, envolvendo tanto os aspectos cognitivos como os instrucionais. Suas publicações incluem artigos em periódicos tais como Brain and Language, Psychology & Neuroscience, Revista DELTA, Revista Brasileira de Linguística Aplicada, Revista da ANPOLL, Lenguas Modernas, Revista Intercâmbio, Revista Ilha do Desterro, entre outros; organização do livro Aspectos cognitivos e instrucionais da leitura (2008), e co-organização do livro Linguagem e cérebro humano: contribuições multidisciplinares (2004), entre outros. Contato: leda@cce.ufsc.br 1 PUBLISHED MANUSCRIPT “syntax allows words to be combined to create unique combinations of meaning”. This means that although there are other sub-processes in reading which have to be successfully performed so that comprehension of any sentence or text can take place, as will be briefly discussed later in this text, it is vital that syntactic processing be realized appropriately. In view of the above, it is the objective of this chapter to bring an overview of the seminal and the most recent behavioral and neuroimaging studies on the implementation of syntactic processes in the brain. More specifically, the present study seeks to provide an understanding of the most recent findings in relation to how syntactic processing takes place behaviorally and also concerning its cortical representation. Scholars (as cited in ANDERSON, 2011; GAZZANIGA, IVRY & MANGUN, 2009) have proposed the tree structure to represent the structure of a sentence, where the sentence (S) is made up of a noun phrase (NP), which is the subject, and a verb phrase (VP), consisting of the verb (V) and a noun phrase that is the object (O). Figure 1 presents two phrase structures that illustrate the possible meanings of the ambiguous sentence They are cooking apples: (a) that those people (they) are cooking apples; (b) that those apples are for cooking. Figure 1: The tree structure (according to ANDERSON, 2011, p.329). 2 PUBLISHED MANUSCRIPT Some languages have a relatively fixed word order signaling the grammatical relations, as the SVO order (subject-verb-object) in English, while other languages, such as German, have a relatively free word order and signal subjects and objects by marking case (see FRIEDERICI, 2004, p.789 for examples). Specificities apart, language provides users with freedom of choice. Ferreira and Engelhardt (2006, p.61) explain that one thematic role structure can be expressed by different syntactic structures, such as the possibilities of grammatical encoding in a proposition involving a cat, a dog, and a state of fear: “my cat terrifies the dog next door, or the dog next door is terrified of my cat, or it’s my cat that terrifies the dog next door, and so on”. As users of a specific language, we create sentences that respect the constraints on our processing capacity and on grammar. According to Snijders, Vosse, Kempen, Van Berkum, Petersson and Hagoort, (2009, p.1493), “what makes language useful and creative is that words occur in all sorts of different contexts, with the varying combinations of words allowing for an infinite number of higher-level representations”. As already exposed, language comprehension requires more than just recognizing individual words. According to Gagné, Yekovich and Yekovich’s model (1993), reading comprehension involves four levels of comprehension – decoding, literal comprehension, inferential comprehension and comprehension monitoring3. The lowest level processes involve the decoding of the printed information to access meaning and literal comprehension, subdivided into lexical access and parsing. Lexical access refers to accessing the best interpretation of the word from all the options activated in our mental lexicons, whereas parsing, the focus of this study, involves using the syntactic and linguistic rules of the language for putting words together to 3 The component processes inferential comprehension and comprehension monitoring and their subdivisions will not be explained here due to space limits and the scope of the chapter. 3 PUBLISHED MANUSCRIPT form meaningful ideas, or propositions. Anderson (2011, p.360) explains that parsing involves understanding, with our knowledge of the structure of a language, that the sentences A doctor shot a lawyer and A doctor was shot by a lawyer represent different meanings. According to King and Just (1991, p.580), syntactic processing requires “the temporary storage of word representations during the left-to-right processing of a sentence”. The cognitive architecture of the language processing system, however, remains a controversy. According to Friederici (2004, p.789), “some researchers assume syntactic processes to be modular and to precede semantic processing in time” while “others view comprehension as a highly interactive, online process”. These two views differ in the time course syntactic and semantic information interact. The first, called serial/modular account, or syntax-first approach “does not allow interaction during the initial stage of phrase structure building, but only during a later stage of processing” (FRIEDERICI, 2004, p.790). This account is represented by the Garden-Path theory, proposed by Frazier (1979, as cited in PICKERING & VAN GOMPEL, 2006). The second view, called interactive, holds that interaction takes place during all processing stages, assuming that “all potentially relevant sources of information can be used immediately during sentence processing and can affect initial processing decisions” (PICKERING & VAN GOMPEL, 2006, p.460). Each of these views can count on “ample empirical evidence from behavioral studies conducted on healthy participants. Therefore, it is difficult to describe the cognitive architecture of the language processing systems on the basis of these data alone” (FRIEDERICI, 2004, p.790). Data collected by means of technological tools, such as fMRI (Functional Magnetic Resonance Imaging), PET (Positron Emission Tomography), and ERPs (Event-related Potentials) 4 PUBLISHED MANUSCRIPT serve as additional sources of evidence, contributing to the development of a framework that really represents what goes on in the human brain while comprehending language. In what follows, seminal behavioral and neuroimaging studies of syntactic processing are reviewed briefly with a focus on the patterns that emerge from the data and on the implications of these studies to the understanding of how this process takes place in the brain. This way the present chapter is divided into three subsections: behavioral studies of syntactic processing; neural substrates of syntactic processing; and final remarks. 1. Behavioral studies of syntactic processing There has been an increasing interest to understand whether “multiple syntactic representations are generated when a syntactically ambiguous string is encountered” (MACDONALD, JUST & CARPENTER, 1992, p.57). Serial/Modular models postulate that only one syntactic representation is constructed for the ambiguous string, and when the incoming information is incompatible with this interpretation, a different interpretation has to be sought, since the original one had to be abandoned. On the other hand, interactive models hypothesize that multiple syntactic representations are constructed for the syntactic ambiguity. These multiple representations are maintained until some information indicates which interpretation is right, the moment in which the incorrect representations are rejected. In a garden-path sentence such as The horse raced past the barn fell (p.57) readers commit to one interpretation (raced as the main verb) but when they encounter the verb fell, parsing fails, since it does not make sense. Reading the sentence for a second time, the reader may insert specific punctuation, change morphology, and/or even insert a word, so that the sentence becomes 5 PUBLISHED MANUSCRIPT understandable: The horse, racing past the barn, fell; or The horse raced past the barn and fell; and even The horse who raced past the barn fell. Ferreira and Clifton (1986, p.352) used temporarily ambiguous sentences such as The evidence examined by the lawyer turned out to be unreliable and The defendant examined by the lawyer turned out to be unreliable. Both sentences omit complementizers (who, that) and a verb (was). The verb examined can be part of a reduced relative clause, in which it is a past participle, or part of a main clause, in which it is a past tense verb. Readers are likely to interpret examined as the main verb and encountering the word by makes processing difficult. Results indicated that syntactic processes are modular, since participants did not take the disambiguating information into account in their interpretation. According to the authors (p.365), their “findings do constitute support for the existence of an informationally encapsulated syntactic processor”, meaning that syntax comes first, since people analyze sentences taking into account the syntactic information before integrating it with semantic and pragmatic information, for instance. King and Just (1991) argue that among a huge body of research about syntactic processing, “only a few researchers have examined individual differences in the processing of syntactic information” (p.580). Their study investigated how syntactic processing and comprehension are influenced by a reader’s working memory capacity (WMC) for language. They used a classic example of syntactic construction that demands resources from working memory (WM)4: center-embedded relative clauses. The object-relative clause The reporter that the senator attacked admitted the error publicly after the hearing (p.584) imposes two types of difficulty: (1) the embedded clause interrupts the main clause, making WM retain the interrupted clause 4 For a review on working memory and individual differences see Bailer (2011, p.7-16). Available at <http://www.tede.ufsc.br/teses/PLLE0484-D.pdf> 6 PUBLISHED MANUSCRIPT representation; and (2) Reporter (in the example) is the subject of the main clause and at the same time the grammatical object of the embedded clause. According to the authors, subjects make a large amount of errors when trying to rephrase this type of sentence. The authors compared objective-relative clauses with subject-relative clauses, as The reporter that attacked the senator admitted the error publicly after the hearing (p.584). These sentences are considered easier to process, since the head noun (reporter) plays the same role in both clauses. Even though the main clause is still interrupted, this type of sentence requires no assignment of conflicting roles to the actor and no perspective shift. Two experiments were conducted: the first imposed an extraneous memory load during processing by presenting the sentences (as the examples above) one word at a time; and the second experiment supplied pragmatic information that presumably would make it easier to comprehend object-relative sentences. In this second experiment, four types of object-relative sentences (p.592) were used: (1) both verbs pragmatically biased The robber that the fireman rescued stole the jewelry; (2) only the relative-clause verb biased The robber that the fireman rescued watched the program; (3) only the main clause verb biased The robber that the fireman detested stole the jewelry; (4) neither verb biased The robber that the fireman detested watched the program. Overall results indicated that object-relative clauses were more demanding, since lower spans took longer to read and their comprehension accuracy was also poorer than that of higher spans, particularly when there is no pragmatic information available. In an earlier study, Daneman and Carpenter (1983, p.579) suggested that poor readers may have inefficient processes that “interfere with the amount of additional information they can store and maintain” to process sentences. Just and Carpenter (1992) repeated Ferreira and Clifton’s experiment (1986), classifying participants as high and low span readers, and adapting the sentences by 7 PUBLISHED MANUSCRIPT adding an explicit syntactic cue, as in the examples The evidence that was examined by the lawyer shocked the jury and The defendant who was examined by the lawyer shocked the jury. According to Just and Carpenter (1992, p.128), both groups used the explicit syntactic cue in a similar way, but only the high spans “have the capacity to take the pragmatic information into account”. The authors argue that syntactic processing for low span readers is essentially modular, because their WMC is not sufficiently large to use non-syntactic information immediately. In contrast, syntactic processing for high span readers is considered interactive, since they have the sufficient resources to take different kinds of information into account immediately. MacDonald, Just and Carpenter (1992) reconcile serial and interactive positions by proposing the Capacity Constrained Parsing Model. According to this model, sentences without ambiguities are integrated into the ongoing representation of the text but when the reader encounters ambiguities, s/he initially constructs multiple representations. Each of the multiple representations is assumed to have an activation level proportional to its frequency, syntactic complexity and, pragmatic plausibility. WMC influences how long the multiple representations can remain activated. As stated by the authors, “for temporary ambiguities resolved with a highly preferred interpretation, both high and low span subjects should comprehend well because all subjects have carried the preferred interpretation” (MACDONALD, JUST & CARPENTER, 1992, p.60). For unpreferred interpretations, high spans should exhibit better comprehension than low spans, since low spans are more quickly taxed by the cognitive burden of maintaining multiple representations while reading the sentence, although both high and low spans should present a large increase in reading times. These authors carried out four experiments with sentences as The experienced soldiers who were told about the dangers conducted the midnight raid (unambiguous), The 8 PUBLISHED MANUSCRIPT experienced soldiers warned about the dangers before the midnight raid (main verb resolution), and The experienced soldiers warned about the dangers conducted the midnight raid (relative clause resolution). Results support and extend previous research “by demonstrating that individual differences in working-memory capacity can produce different parsing outcomes for syntactically ambiguous sentences, which in turn can lead to differences in comprehension” (p.88). The model proposed by the authors unifies single and multiple representation models of parsing, pointing to the adaptability of the mechanisms to the availability of WM resources. Swets, Desmet, Hambrick and Ferreira (2007) conducted two studies to investigate the effects of WM in syntactic ambiguity resolution with native speakers of English and Dutch (246 participants in study 1 and 292 in study 2). The stimuli were composed of ambiguous sentences The maid of the princess/who scratched herself in public/was terribly embarrassed (p.68). Results showed that for both groups of participants “the decision about which constituent of a complex noun phrase is modified by an ambiguous relative clause is influenced by the availability of working memory resources” (p.74). These results (KING & JUST, 1991; JUST & CARPENTER, 1992; MACDONALD, JUST & CARPENTER, 1992; SWETS et al., 2007) suggest that individual differences in WMC is an important factor in syntactic processing, since, in King and Just’s words (1991, p.599), “complex cognitive phenomena arise from the interaction of many different processes and subprocesses, each of which makes demands on limited cognitive resources”. To close this subsection, it is relevant to bring to light that serial models posit that first people compute the syntax, then the word meanings while interactive models 9 PUBLISHED MANUSCRIPT suggest that syntax and semantics are computed simultaneously. Miyake, Just and Carpenter (1994), in their study on lexical ambiguities, propose that lexical and syntactic ambiguities may be processed in a similar way, but differ in terms of the capacity demands associated with the maintenance of multiple interpretations: retaining two different syntactic representations in working memory may be much more capacity-demanding than retaining two different word meanings (p.197-198). 2. Neural substrates of syntactic processing The neural basis of syntactic processing has been inspected through lesion studies and more recently through technological tools, such as PET, fMRI and ERPs5. According to Bookheimer (2002), the lesion-deficit approach has led to a large-module philosophy, that the language system is composed primarily of two domain regions: Broca’s area (inferior frontal) and Wernicke’s area (posterior superior temporal). As Broca’s is close to the primary motor area, lesions to this region involve impairment in “articulation, sequential production of speech, sentence production, syntax, naming, and comprehension of some complex syntactic structures” (p.152). Luria (1966, as cited in BOOKHEIMER, 2002) noticed that Broca’s aphasics made comprehension errors in complex syntactic structures such as passive sentences. In a sentence as The gigantic dog was bitten by the little old lady (GAZZANIGA, IVRY & MANGUN, 2009, p.418), such patients have trouble in assigning syntactic structures - they understand that the lady was bitten by the dog, but when the sentence is written in the active voice, these patients do not present difficulties. Evidently, “comprehension was intact at the word level, but meaning at the sentence level was lost under conditions in which function words or knowledge of the syntactic structure were essential for comprehension” (BOOKHEIMER, 2002, p.156). 5 There has been 20 years since the publication of the first PET study interested in the neural basis of speech: Mazoyer et al. in 1993. 10 PUBLISHED MANUSCRIPT The observation of agrammatic Broca’s aphasics has led researchers to suggest that Broca’s area is the most important part of the neural network responsible for syntactic processing. Patients with other types of aphasia and whose lesions involve other areas, however, often show impairments of syntactic comprehension. In addition, more recently, neuroimaging studies show that “damage to Broca’s area need not give rise to Broca’s aphasia, that Broca’s area has been implicated in semantic processing, working memory, and a range of nonlinguistic tasks” (BERETTA, 2008, p.161). Thus, it seems reasonable to infer that Broca’s area is not the sole area responsible for processing syntax and that other areas are implicated in such a task. The advent of functional neuroimaging tools (as well as ERPs) for language studies has enabled researchers to see the healthy brain at work. These tools have allowed research to go beyond the simple dissociations offered by lesion studies to study the organization of the brain. ERPs have proved to be useful in characterizing the time course of various aspects of language processing while fMRI and PET provide the necessary spatial resolution to determine the neural loci of language processing. With these studies, the large-module theories and serial models started to be challenged, since it seems that “the language system is organized into a large number of relatively small, but tightly clustered and interconnected modules with unique contributions to language processing” (BOOKHEIMER, 2002, p.152). These studies are all consistent with the role Broca’s area (left inferior frontal gyrus) plays in syntactic processing, but, according to Newman, Just, Keller, Roth and Carpenter (2003, p.297), “they fail to converge on a single region within the inferior frontal cortex”. The time course of the neural activity related to syntactic processes can be measured by EEG (electroencephalogram), that registers the total postsynaptic electrical activity, and by MEG (magnetoencephalogram), that registers its magnetic field. 11 PUBLISHED MANUSCRIPT According to Friederici (2004, p.794), “different stimulus types have been shown to elicit ERP waveforms that are distinct with respect to their latency, amplitude and topography”. Four components are recurrent in language research: (1) the N400, a negativity peaking around 400 milliseconds (ms) after stimulus onset, has been implicated in processes of semantic integration; (2) the LAN, a left anterior negativity peaking around 200-500 ms after a syntactically anomalous element, appears in violations of phrase structure, subject-verb agreement and verb argument structure, being considered a marker of morphosyntactic processes; (3) the P600, also known as the syntactic positive shift: a bilaterally distributed positivity peaking around 600 ms, appears as a function of ambiguous syntactic structures, syntactic violations and syntactic complexity, being considered a marker of syntactic integration difficulty; and (4) the sustained left frontal negativity is “assumed to reflect processes of syntactic working memory that are presumably necessary for the processing of sentences in noncanonical word order” (FRIEDERICI, 2004, p.797). This author argues that syntax and semantics take place at different times, since “syntactic and semantic processes are independent during early processing stages, and … interaction takes place during a stage of late integration” (p.798). Friederici, Steinhauer, Mecklinger and Meyer (1998), for instance, investigated individual differences in syntactic ambiguity resolution of German sentences with the aid of ERPs. They found, in opposition to MacDonald, Just and Carpenter (1992), that “high span readers are more efficient parsers than low span readers because they commit themselves to a single preferred structure when confronted with structural ambiguities” (FRIEDERICI et al., 1998, p.219). Results may differ due to the languages researched and stimulus materials as well as different methods (behavioral, ERPs), that in turn, influence the interpretation of the data. 12 PUBLISHED MANUSCRIPT PET studies conducted by Stromswold, Caplan, Alpert and Rauch (1996) and Caplan, Alpert and Waters (1999) found increased rCBF (regional cerebral blood flow) in the inferior frontal gyrus (IFG) in syntactic processing of complex sentences. Stromswold et al. (1996) asked eight healthy monolingual native English-speaking college students (ages 19-28) to make acceptability judgments about sentences in three conditions: (1) center-embedded relative clauses, such as The juice that the child spilled stained the rug (plausible) and The child that the juice spilled stained the rug (implausible); (2) right-branching relative clauses, such as The child spilled the juice that stained the rug (plausible) and The juice stained the rug that spilled the child (implausible); and (3) the plausible center-embedded and right-branching relative clause sentences used in conditions 1 and 2 and unacceptable sentences where one of the verbs or nouns were replaced by an orthographically and phonetically possible pseudoword, such as The juice that the child chorried stained the rug and The child spilled the juice that mulved the rug6. These sentences were used since psycholinguistic research has shown that “normal subjects reliably make more errors and take longer to process sentences that contain center-embedded relative clauses sentences than sentences that contain right-branching relative clauses” (STROMSWOLD et al., 1996, p.457). The sentences were presented visually and increased rCBF was found in Broca’s area, particularly in the pars opercularis, when the subjects made judgments about the plausibility of syntactically more complex sentences (center-embedded relative clauses) as compared to syntactically less complex sentences (right-branching relative clauses). When the subjects had to decide the plausibility of sentences in condition 3 (plausible vs. implausible sentences with pseudowords), increased rCBF was found in the left perisylvian language areas, including Broca’s and Wernicke’s area and the adjacent 6 These sentences cited in the paragraph were extracted from pages 456-457 of the original article (STROMSWOLD et al., 1996). 13 PUBLISHED MANUSCRIPT portions of the superior temporal gyrus (STG). According to the authors (STROMSWOLD et al., 1996, p.471), “the results provide evidence supporting the role of a portion of Broca’s area in the assignment of syntactic structure in sentence comprehension, or in operations associated with this process”. Caplan et al. (1999) asked sixteen healthy monolingual native English-speaking college students (ages 22-34) to make acceptability judgments about sentences presented auditorily in two conditions: (1) cleft object sentences: It was the juice that the child enjoyed (plausible) and It was the child that the juice enjoyed (implausible); and (2) cleft subject sentences: It was the child that enjoyed the juice (plausible) and It was the juice that enjoyed the child (implausible). The authors made use of cleft object and subject sentences because, according to them (p.345), “preliminary psycholinguistic research indicated that normal subjects reliably made more errors and took longer to process cleft object sentences than cleft subject sentences when they were presented auditorily”. Results revealed, in consort with Stromswold et al. (1996), that processing syntactically more complex sentences is associated with an increase of rCBF in Broca’s area. Nonetheless, while Stromswold et al. (1996) reported greater rCBF centered in the pars opercularis (Brodmann’s area BA 44), Caplan et al. (1999) found increased activity in the pars triangularis (BA 45). Although these studies implicate the IFG in syntactic processing, there is no convergence on a particular location within this area, “perhaps in part because of the differences in particular tasks and image subtractions” (NEWMAN et al., 2003, p.297). Mazoyer, Tzourio, Syrota, Murayama, Levrier, Salamon, Dehaene, Cohen and Mehler (1993), in their pioneer PET study about the neural basis of speech, investigated sixteen healthy French college-level students (mean age 23.6) to see whether there are, in the human brain, specialized regions responsible for the acoustic, phonological, 14 PUBLISHED MANUSCRIPT lexical, prosodic, syntactic and conceptual levels of linguistic analysis. They compared rCBF while subjects listened to “continuous speech in an unknown language, to lists of French words, or to meaningful and distorted stories in French” (p.467). Results indicated that speech processing recruits a network of areas, each of which may be specialized in one aspect but requires support from the others to achieve comprehension. Mazoyer et al. (1993, p.473) advocated that left superior and middle temporal areas “may be truly devoted to sentence-level processing, including syntactic parsing”, though these results do not necessarily isolate syntactic processing. According to Newman et al. (2003), neuroimaging studies on single word processing have implicated the pars triangularis of the IFG in semantic processing, the processing of meaning. Sirigu et al. (1998) compared syntax and script processing in ten patients with lesions in the pars triangularis and anterior extensions. These patients showed a great difficulty to produce a logical story narrative from a list of actions (script task) while their performance on a syntactic task (producing a grammatically correct sentence by assembling a list of phrases into a sensible order) was comparatively unimpaired. These results suggest that the pars triangularis is involved in semantic processing at the word level, but it is also involved in processing actions and their arguments, thus, it is involved in thematic processing, recognizing agents and patients. Additionally, Newman et al. (2003) argue that there is also evidence suggesting the involvement of the pars opercularis in syntactic processing (JUST et al., 1996). In Sirigu et al.’s study (1998) cited above, patients with lesions in the frontal operculum and extending posteriorly had difficulty in the syntax task and not in the script one. These results suggest “the portion of the left frontal operculum including area 44 and part of area 16 is an important region for processing word sequences at the syntactic level” (SIRIGU et al., 1998, p.776). 15 PUBLISHED MANUSCRIPT Syntactic and semantic information are so intertwined that it turns out to be difficult to study them separately. When the syntactic structure is disrupted, a thematic anomaly can be produced, as in the examples above with object- and subject-relative clauses. As well, in the sentence all the eaten have chickens snakes (NEWMAN et al., 2003, p.298) the anomalous syntactic structure disturbs the thematic interpretation. Consequently, one possibility for the inconsistency in the location of activation within IFG (pars triangularis vs. pars opercularis) during complex syntactic processing is this interaction between syntax and semantics: “complex syntactic processing is likely to result in complex semantic processing” (NEWMAN et al., 2003, p.298). Figure 2 presents the summary of activations in the IFG across the studies reviewed by Bookheimer (2002). Figure 2: Summary of IFG activations across the studies reviewed by Bookheimer (2002). “Semantic areas (shown in red) cluster around the anterior, inferior IFG (pars orbitalis); phonological regions center around the posterior superior IFG at the border of Brodmanns areas 44 and 6; syntax regions fall in the center near middle IFG in pars triangularis, area 44/45” (BOOKHEIMER, 2002, p.190). Just et al. (1996) made use of fMRI and increasingly more complex syntactic sentences to examine the brain activation modulated by sentence comprehension in fifteen college-age students: (1) conjoined active clauses as The reporter attacked the 16 PUBLISHED MANUSCRIPT senator and admitted the error; and the embedded relative clauses used by King and Just (1991): (2) subject-relative clauses as The reporter that attacked the senator admitted the error; and (3) object-relative clauses as The reporter that the senator attacked admitted the error. They observed “an increase in the amount of brain activation as the demand on the language processing system increased from the simplest to the most complex sentence structures” (JUST et al., 1996, p.114) in four areas: the classical left-hemisphere language areas (Broca’s and Wernicke’s areas) and their right homologues. Just, Carpenter and Varma (1999) used this study (JUST et al., 1996) as a test bed for their 4CAPS (Cortical Capacity-Constrained Concurrent Activation-based Production System) model, a computational modeling architecture that accounts for the word-by-word processing times and error rates on these types of sentences. In the authors’ words, “The model processes the successive words of a sentence, one at a time, attempting to interpret each word as fully as possible in the context of the preceding words while incrementally constructing a representation of the sentence, just as human readers do” (JUST et al., 1999, p.132-133). Keller, Carpenter and Just (2001) manipulated syntactic complexity and word frequency to investigate with fMRI “how the demands made by syntactic and lexical processes are manifested in patterns of cortical activation” (p.223) in thirty collegestudent level subjects. The stimuli, based on Just et al.’s study (1996), consisted of: (1) conjoined active sentences as The writer attacked the king and admitted the mistake at the meeting; (2) object-relative clauses as The writer that the king attacked admitted the mistake at the meeting; and (3) sentences with word frequency manipulated as The pundit that the regent attacked admitted the gaffe at the conclave. Results revealed that the left inferior frontal gyrus (Broca’s), the left superior and middle temporal region (Wernicke’s), the left inferior parietal cortex, the left dorsolateral prefrontal cortex and 17 PUBLISHED MANUSCRIPT the ventral extrastriate cortex work in collaboration to achieve sentence comprehension. These results demonstrate that different linguistic processes influence each other as they unfold and that different areas collaborate as demands increase. For Keller, Carpenter and Just (2001, p.235), frontal areas may be relatively specialized for the generation of linguistic representations, while parietal and temporal areas may be relatively specialized for the interpretation, elaboration and storage of such representations. Distinctions among types of representations at the linguistic level (syntactic, semantic, phonological, orthographic), however, do not neatly correspond to anatomical locations. Syntactic processing and maintenance appears to require coordinated communication between at least Broca’s and Wernicke’s areas, and may involve right-hemisphere areas as well. Mason, Just, Keller and Carpenter (2003), in their two fMRI experiments, studied the time course and amplitude of brain activity during the processing of syntactically ambiguous sentences. The stimuli consisted of the main verb (MV) and reduced relative (RR) ambiguous sentences used by MacDonald et al. (1992), being the former considered preferred and the latter, unpreferred sentences. As results, the authors found that for both types of ambiguous sentences there was a higher level of brain activity, even when it was resolved in favor of the preferred syntactic structure. They argue that “the processing of ambiguous sentences need not consume more time than unambiguous sentences, it requires additional processing” (p.1332). Results showed that syntactic ambiguity poses extra brain workload that is supported by at least two areas, the left IFG and the left STG. Both areas are recruited to handle the ambiguity, but only the left IFG kept increasingly activated longer, during the processing of the probe question and even with the activation decay to baseline. The authors agree with Keller, Carpenter and Just (2001) that possibly the “left IFG (Broca’s area) is involved in the internal generation of abstract syntactic representations that are reiteratively communicated to left STG (Wernicke’s area) for interpretation and elaboration through the activation of semantic representations” (MASON et al., 2003, p.1334). 18 PUBLISHED MANUSCRIPT More recently, Prat, Mason and Just (2011) have proposed the dynamic spillover hypothesis to explain the increased activation of the right hemisphere as syntactic complexity increased in Just et al.’s study (1996). In this line, according to Keller, Carpenter and Just (2001, p.223), “The general suggestion is that not just one area but several participate in sentence processing and are affected by the increased processing demands imposed by the manipulation of syntactic complexity”. Since language processing is such a complex task, it is not surprising that several brain areas subserve syntactic processing. Snijders et al. (2009) explored word-category (noun-verb) ambiguous words, like bike or trains. According to the authors, “sentences containing lexical ambiguities tax both retrieval and unification processes stronger than unambiguous sentences” (SNIJDERS et al., 2009, p.1495). Based on the literature, the authors hypothesized that the left posterior temporal gyrus (LpMTG) would support the retrieval of lexicalsyntactic information (selecting the representation of the word from memory) and the left inferior frontal gyrus (LIFG) would contribute to syntactic unification (combining the retrieved single word form information into higher-level representations). Twentyeight Dutch participants read sentences and word sequences containing noun-verb ambiguous words at critical positions. Results revealed a similar pattern to the one found by Keller, Carpenter and Just (2001) that the syntactic unification process requires the dynamic interplay between LIFG and LpMTG, being the posterior LIFG implicated in the unification of words into a sentence structure and the LpMTG involved in the retrieval of lexical-syntactic information from memory. Newman et al. (2003) investigated the brain activation of thirteen college participants in their responses to two types of grammatical violations. Conjoined active and object-relative sentences were used with two types of ungrammaticalities: (1) noun- 19 PUBLISHED MANUSCRIPT verb agreement as The lady praises the sister and meet the artist in the night and The waitress that the lawyer interrupts notice the commander on the stairs; and (2) extra verb as The coach watched the poet and told the visitor took in the evening and The duke that seamstress forgave walked the baby took down the hall (p.298). Results showed that the pars triangularis is involved in semantic/thematic aspects of comprehension (extra-verb condition), while the pars opercularis is more involved in manipulating the syntactic structure (subject-verb agreement). Activation in the temporal region significantly increased with sentence complexity corroborating Keller, Carpenter and Just’s claim (2001) that this area is involved in the coactivation of distributed semantic representations required in lexical access, in the mapping of thematic roles and in syntactic parsing. The activation of the intraparietal sulcus, long associated with visuo-spatial processing WM, suggests that this area is “involved in generating a spatial structure that encodes the thematic roles of a sentence” (NEWMAN et al., 2003, p.306). Lee and Newman (2010) used conjoined-active and object-relative clauses adapted from Keller, Carpenter and Just (2001) to investigate with fMRI the effects of two different presentation paradigms on syntactic processing. Twenty college-level participants were presented with the two types of sentences in the rapid serial visual presentation (RSVP) format and the whole sentence format and were required to answer a comprehension probe after each sentence. The RSVP format entails presenting each word serially in the center of the screen and it is commonly employed in the attempt to minimize head motion. Findings indicated that the RSVP format is more computationally, resource demanding and that it interferes with syntactic processing when compared to the whole sentence format. The left posterior MTG was commonly activated for both presentation formats, indicating that it is essential for sentence 20 PUBLISHED MANUSCRIPT processing. The LIFG showed a more widespread syntactic complexity effect during the processing of the probes, indicating its strong association with syntactic processing. The study is the first fMRI study to compare both presentation methods and it points out a critical difference between them in terms of memory processing and how they interact with syntactic complexity. The authors acknowledge that there are other presentation methods, such as the moving window paradigm, used by Keller, Carpenter and Just (2001), in which “stimuli are presented serially but in the space to the right of the previous word instead of the same location as the previous word”, which would be expected to require less WM load (LEE & NEWMAN, 2010, p.76). When comparing these fMRI results to those PET findings aforementioned (STROMSWOLD et al, 1996; CAPLAN et al., 1999; MAZOYER et al., 1993), it is possible to observe that fMRI is more sensitive than PET in terms of spatial resolution, since fMRI studies show activation in regions that were missed by PET scanning. Neither of the technological tools measure directly the activity of the nerve cells in the brain, instead, they take advantage of the fact that the neural activity in a particular brain region induces an increased local blood flow to the region. It is well known that PET is considered invasive, since it is necessary to inject a radioactive tracer into the participant’s bloodstream, while fMRI is considered a non-invasive technique. Likewise, the differences in results may have arisen from differences in the demands of the tasks, methodological decisions and statistical methods used in each experiment. Based on the literature, Hagoort (2013) has proposed the Memory Unification Control model (MUC). According to this model, regions in the temporal cortex and the angular gyrus in the parietal cortex store information about morphology, phonology and syntax. Frontal regions, including Broca’s area, are essential for unification operations: semantic unification recruits BA 47 and 45, syntactic unification BA 45 and 44 and 21 PUBLISHED MANUSCRIPT phonological unification, BA 44 and ventral parts of BA 6. The dorsolateral prefrontal cortex, anterior cingulate cortex and parts of the parietal cortex are involved in attention processes. In the attempt to trace the anatomy of syntactic processing, Indefrey (2012) reviewed 79 sentence comprehension studies and 6 production studies that either compared sentences to a below-sentence-level control condition or syntactically more demanding sentences to less demanding ones. He concluded that the pars opercularis (BA 44) and the pars triangularis (BA 45) of the left posterior IFG and the posterior parts of the superior and MTG were more strongly activated when participants read or listened to syntactically demanding sentences, if compared to when they read or listened to simpler sentences. As expected, for reading, there was additional activation of the right homologue of Broca’s area and the left occipital cortex, whereas for listening, additional activation was found in the lower motor cortex and the more anterior parts of the temporal lobe. In the same vein, Price (2010) reviewed 100 online published studies in 2009 using fMRI to unveil the functional anatomy of speech comprehension and production in the healthy adult brain. She acknowledged the fact that Syntactic processing has been investigated by comparing sentences with grammatical errors to sentences without grammatical errors; and for sentences with more versus less syntactically complex structures. In both cases, the demands on syntactic processing are confounded by the differing demands on semantics because both grammatical errors and complex sentences make it more difficult to extract the meaning of the sentence (p.72). By reviewing those studies, Price found (1) left pars opercularis activation for sentences with syntactic errors and violations in the verb-agreement structure; (2) left ventral pars opercularis activation associated with verbal WM and predicting the sequence of events; (3) left dorsal pars opercularis activation as syntactic complexity increased in auditory sentence comprehension; (4) the region on the border between the 22 PUBLISHED MANUSCRIPT planum temporale and the ventral supramarginal gyrus has been associated with syntactic errors and complexity, but also associated with semantic difficulties, perhaps because the activation reflects a WM rehearsal strategy or subvocal articulation. Figure 3: Areas involved in language comprehension and production as reviewed by Price (2010, p.64). As it can be seen, for syntactic processing, the regions pPT, vSMG and vpOP/dpOp seem to be involved. 23 PUBLISHED MANUSCRIPT As expected, activation in these regions is not specific to syntactic processing. According to Price (2010, p.75), “The cortical networks supporting language comprehension are dynamically determined by the task and context”. Still, future research will have to elucidate which aspects of syntactic processing are subserved by which areas of the brain and under what circumstances, taking into account individual differences and different languages, as well as studying sentence processing within a much broader perspective, reconciling previous results and leading to more detailed models and theories. 3. Final remarks As Price (2010, p.62) states, “a review is limited to the reviewer’s own understanding of the topic and how the conclusions of each paper fit together”, this paper attempted to provide a brief review of behavioral and neuroimaging studies of syntactic processing, focusing on the whole that emerges from the data. The overall body of literature on the neurobiology of language, more specifically, syntax, although sometimes controversial, demonstrates that (a) the language network is more extended than the classical language regions, including right hemisphere regions; (b) the division of labor between Broca’s and Wernicke’s areas is not language production vs. language comprehension, since both regions seem to be involved in syntactic processing; (c) the language relevant regions are not language specific; and (d) “the function-to-structure mapping as one-area-one-function is almost certainly incorrect. More likely, any cortical region is a node that participates in the function of more than one functional network” (PETERSSON, FOLIA & HAGOORT, 2012, p.84). In Hagoort’s own words (2013, p.10-11), “the basic principle of brain organization for higher cognitive functions 24 PUBLISHED MANUSCRIPT is that these are based on the interaction between a number of neuronal circuits and brain regions that support the different contributing functional components”. The networks reviewed here are not necessarily specialized for language, but they need to be recruited to enable successful language comprehension and production. Neuroimaging studies (as well as ERP studies) have added anatomical precision and a level of complexity unavailable to lesion studies, in addition to the possibility of conducting studies with healthy participants. In spite of the limitations (number of participants, different tasks, different methods), the main findings highlight the complexity of the brain organization, “reflecting not only a high degree of specialization …, but also a high degree of interactivity and interdependence” (BOOKHEIMER, 2002, p.183). In Prat et al.’s terms (2011, p.1), “higher cognitive abilities are indexed by less (more focal distribution or lower intensity) brain activation”, what is known as neural efficiency, the ‘doing more with less’. As well, they coined the term neural synchronization, known earlier as functional connectivity, which refers to the collaboration of the areas responsible for executing component processes of the different tasks so that the goals are achieved in our daily lives, for instance, in relation to text comprehension and speech production. All in all, the attempt to describe the neural basis of language and human behavior is a clear example of interdisciplinarity, of how areas can stimulate each other. As stated by Friederici (2004, p.798), Neuroimaging needs input from cognitive models, in this case, psycholinguistics, in order to be able to pose the right questions (and to construct the crucial material). Modeling psycholinguistic functions in turn can fruitfully use the input from neuroscience to constrain crucial aspects of the cognitive architecture of the language system. 25 PUBLISHED MANUSCRIPT Abstract: This review chapter aims at presenting an overview of the seminal and the most recent behavioral and neuroimaging studies on the implementation of syntactic processes in the brain. More specifically, the present study seeks to provide an understanding of the most recent findings in relation to how syntactic processing takes place behaviorally and also concerning its cortical representation. Traditionally, syntactic processing has been investigated by comparing sentences with ambiguities and sentences without ambiguities; sentences with grammatical errors and sentences without grammatical errors; and sentences with more versus less complex syntactic structures. Different models have sought to explain how parsing occurs: serial models have posited that syntax is computed before word meanings while interactive models have suggested that syntax and semantics are computed simultaneously. Behavioral research has found evidence for both views. Lesion studies have led researchers to suggest that Broca’s area (left inferior frontal gyrus) would be responsible for syntax. With the advent of functional neuroimaging tools (as well as event-related potentials, ERPs) for language studies, researchers could examine the healthy brain at work. These tools have allowed research to go beyond the simple dissociations offered by lesion studies to study the organization of the brain. ERPs characterize the time course of a language process whereas fMRI (functional magnetic resonance imaging) and PET (positron-emission tomography) provide the necessary spatial resolution to determine the neural loci of language processing. These technological tools have challenged what we knew about syntactic processing, since they have been showing “the language system is organized into a large number of relatively small, but tightly clustered and interconnected modules with unique contributions to language processing” (BOOKHEIMER, 2002, p.152). The studies reviewed in the chapter are all consistent with the role that Broca’s area plays in syntactic processing, although “they fail to converge on a single region within the inferior frontal cortex” (NEWMAN et al., 2003, p.297). Differences may have arisen from differences in the demands of the tasks, design of the studies as well as the statistical methods applied in each experiment. The studies, as a group, seem to implicate the classical left-hemisphere language areas (inferior frontal and superior temporal regions) and their right homologues for syntactic processing, but as expected, activation in these regions is not specific to syntactic processing. There is large venue for future research, such as to elucidate which aspects of syntactic processing are subserved by which areas of the brain and under what circumstances. Future studies should take into account individual differences and different languages, as well as studying sentence processing within a much broader perspective, reconciling previous results and leading to more detailed models and theories. Keywords: syntactic processing; reading comprehension; literature review Resumo: Este capítulo de revisão de literatura tem como objetivo apresentar um panorama geral dos estudos seminais comportamentais e mais recentes de neuroimagem sobre a implementação do processamento sintático no cérebro. Mais especificamente, o presente estudo visa proporcionar uma compreensão das mais recentes descobertas em relação à forma como o processamento sintático ocorre e também sobre a sua representação cortical. Tradicionalmente, investiga-se o processamento sintático comparando frases com ambiguidades e frases sem ambiguidades; frases com erros gramaticais e frases sem erros gramaticais; e frases com estruturas sintáticas mais e menos complexas. Diferentes modelos têm procurado explicar como ocorre o parseamento: modelos de série postulam que a sintaxe é computada antes do significado das palavras, enquanto os modelos interativos sugerem que a sintaxe e a semântica são computadas simultaneamente. A pesquisa comportamental encontrou evidências para os dois tipos de modelo. Estudos conduzidos com pacientes com lesões cerebrais levaram 26 PUBLISHED MANUSCRIPT os pesquisadores a sugerir que a área de Broca (giro frontal inferior esquerdo) seria responsável pelo processamento sintático. Com o advento das ferramentas de neuroimagem funcional (bem como os potenciais relacionados a eventos, ERPs) para estudos de linguagem, os pesquisadores passaram a examinar o cérebro saudável trabalhando. Estas ferramentas permitiram a pesquisa a ir além das dissociações simples oferecidas por estudos de lesões para estudar a organização do cérebro. ERPs caracterizam o curso temporal de um processo de linguagem enquanto fMRI (ressonância magnética funcional) e PET (tomografia por emissão de pósitrons) fornecem a resolução espacial necessária para determinar a localização neural do processamento da linguagem. Estas ferramentas tecnológicas desafiaram o que sabíamos sobre o processamento sintático, uma vez que estão revelando que "o sistema de linguagem é organizado em um grande número de módulos relativamente pequenos, mas bem agrupados e interconectados com contribuições únicas para o processamento da linguagem" (BOOKHEIMER, 2002, p.152, nossa tradução). Os estudos revisados no capítulo são todos consistentes com o papel que a área de Broca desempenha no processamento sintático, embora "eles não conseguem convergir em uma única região no córtex frontal inferior" (NEWMAN et al., 2003, p.297, nossa tradução). As diferenças podem ter surgido a partir das diferenças na demanda das tarefas, no desenho dos estudos, bem como os métodos estatísticos aplicados em cada experimento. Os estudos, como um grupo, parecem implicar as áreas clássicas da linguagem do hemisfério esquerdo (regiões frontal inferior e temporal superior) e suas áreas homólogas no hemisfério direito para o processamento sintático, mas como seria de se esperar, a ativação nessas regiões não é específica para o processamento sintático. Há um grande espaço para pesquisas futuras, tais como para elucidar quais aspectos do processamento sintático são subservidos por quais áreas do cérebro e sob quais circunstâncias. Futuros estudos devem levar em conta as diferenças individuais e línguas diferentes, bem como o estudo de processamento da frase dentro de uma perspectiva muito mais ampla, conciliando os resultados anteriores e levando a teorias e modelos mais detalhados. Palavras-chave: processamento sintático; compreensão de leitura; revisão de literatura References ANDERSON, J.R. Cognitive Psychology and Its Implications. 7th ed. New York: Worth Publishers, 2011. BAILER, C. Working memory capacity and attention to form and meaning in EFL reading. Dissertação (Mestrado em Letras) – Pós-Graduação em Letras/Inglês e Literatura Correspondente, Universidade Federal de Santa Catarina, Florianópolis, 2011. BERETTA, A. Disorders of Syntax. In: STEMMER, B.; WHITAKER, H.A. (Eds.). Handbook of the Neuroscience of Language (p.155-163). USA: Academic Press Elsevier, 2008. BOOKHEIMER, S. Functional MRI of Language: new approaches to understanding the cortical organization of semantic processing. Annual Review Neuroscience, 25, p.151191, 2002. 27 PUBLISHED MANUSCRIPT CAPLAN, D.; ALPERT, N.; WATERS, G. PET Studies of Syntactic Processing with Auditory Sentence Presentation. Neuroimage, 9, p.343-351, 1999. DANEMAN, M.; CARPENTER, P.A. Individual differences in integrating information between and within sentences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(4), p.561-584, 1983. FERREIRA, F.; CLIFTON, C.J. The Independence of syntactic processing. Journal of Memory and Language, 25, p.348-368, 1986. FERREIRA, F.; ENGELHARDT, P.E. Syntax and Production. In: Traxler, M.J.; Gernsbacher, M.A. (Eds.), Handbook of Psycholinguistics (p.61-91). 2nd ed. Amsterdam: Elsevier, 2006. FRIEDERICI, A.D.; STEINHAUER, K.; MECKLINGER, A.; MEYER, M. Working memory constraints on syntactic ambiguity resolution as revealed by electrical brain responses. Biological Psychology, 47, 193–221, 1998. FRIEDERICI, A.D. The Neural Basis of Syntactic Processes. In: Gazzaniga, M.S. (Ed.) The Cognitive Neurosciences III (p.789-801). Cambridge, MA: MIT Press, 2004. GAGNÉ, E.D.; YEKOVICH, C.W.; YEKOVICH, F.R. The cognitive psychology of school learning. New York: Harper Collins College Publishers, 1993. GAZZANIGA, M.S.; IVRY, R.B.; MANGUN, G.R. Cognitive Neuroscience: The Biology of the Mind. 3rd ed. New York: Norton, 2009. HAGOORT, P. MUC (Memory, Unification, Control) and beyond. Frontiers in Psychology, 4, article 416, 1-13, 2013. INDEFREY, P. Hemodynamic Studies of Syntactic Processing. In: FAUST, M. The Handbook of the Neuropsychology of Language (p.209-228). 1st ed. Oxford, UK: Wiley-Blackwell, 2012. JUST, M.A.; CARPENTER, P.A. A Capacity Theory of Comprehension: Individual Differences in Working Memory. Psychological Review, 99, p.122-149, 1992. JUST, M.A.; CARPENTER, P.A.; KELLER, T.A.; EDDY, W.F.; THULBORN, K.R. Brain activation modulated by sentence comprehension. Science, 274, p.114-116, 1996. JUST, M.A., CARPENTER, P.A., VARMA, S. Computational Modeling of High-Level Cognition and Brain Function. Human Brain Mapping, 8, 128-136, 1999. KELLER, T.A.; CARPENTER, P.A.; JUST, M.A. The Neural Bases of Sentence Comprehension: a fMRI examination of syntactic and lexical processing. Cerebral Cortex, 11, 223-237, 2001. KING, J.; JUST, M.A. Individual Differences in Syntactic Processing: The Role of Working Memory. Journal of Memory and Language, 30, p.580-602, 1991. 28 PUBLISHED MANUSCRIPT LEE, D., NEWMAN, S.D. The effect of presentation paradigm on syntactic processing: an event-related fMRI study. Human Brain Mapping, 31, 65-79, 2010. MACDONALD, M.E.; JUST, M.A.; CARPENTER, P.A. Working Memory Constraints on the Processing of Syntactic Ambiguity. Cognitive Psychology, 24, p.56-98, 1992. MASON, R.A., JUST, M.A., KELLER, T.A., CARPENTER, P.A. Ambiguity in the Brain: What Brain Imaging Reveals About the Processing of Syntactically Ambiguous Sentences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29(6), 1319-1338, 2003. MAZOYER, B.M.; TZOURIO, N.; FRAK, V.; SYROTA, A.; MURAYAMA, N.; LEVRIER, O.; SALAMON, G.; DEHAENE, S.; COHEN, L.; MEHLER, J. The cortical representation of speech. Journal of Cognitive Neuroscience, 5:4, p.467-479, 1993. MIYAKE, A.; JUST, M.A.; CARPENTER, P.A. Working memory constraints on the resolution of lexical ambiguity: maintaining multiple interpretations in neutral contexts. Journal of Memory and Language, 33, p.175-202, 1994. NEWMAN, S.D.; JUST, M.A.; KELLER, T.A.; ROTH, J.; CARPENTER, P.A. Differential effects of syntactic and semantic processing on the subregions of Broca’s area. Cognitive Brain Research, 16, p.297-307, 2003. PETERSSON, K.; FOLIA, V.; HAGOORT, P. What artificial grammar learning reveals about the neurobiology of syntax. Brain & Language, 120, p.83-95, 2012. PICKERING, M.J.; VAN GOMPEL, R.P.G. Syntactic Parsing. In: M.J. Traxler; M.A. Gernsbacher (Eds.), Handbook of Psycholinguistics (p.455-503). 2nd ed. Amsterdam: Elsevier, 2006. PRAT, C.S.; JUST, M.A. Exploring the neural dynamics underpinning individual differences in sentence comprehension. Cerebral Cortex, 21(8), p.1747-1760, 2010. PRAT, C.S.; MASON, R.A.; JUST, M.A. Individual differences in the neural basis of causal inferencing. Brain & Language, 116, p.1-13, 2011. PRICE, C.J. The anatomy of language: a review of 100 fMRI studies published in 2009. Annals of the New York Academy of Sciences, 1191, p.62-88, 2010. SIRIGU, A.; COHEN, L.; ZALLA, T.; PRADAT-DIEHL, P.; VAN EECKHOUT, P.; GRAFMAN, J.; AGID, Y. Distinct frontal regions for processing sentence syntax and story grammar. Cortex, 34, p.771-778, 1998. SNIJDERS, T.M., VOSSE, T., KEMPEN, G., VAN BERKUM, J.J.A., PETERSSON, K.M., HAGOORT, P. Retrieval and Unification of Syntactic Structure in Sentence Comprehension: an fMRI Study Using Word-Category Ambiguity. Cerebral Cortex, 19, 1493-1503, 2009. 29 PUBLISHED MANUSCRIPT STROMSWOLD, K; CAPLAN, D.; ALPERT, N.; RAUCH, S. Localization of Syntactic Comprehension by Positron Emission Tomography. Brain and Language, 52, p.452-473, 1996. SWETS, B.; DESMET, T.; HAMBRICK, D.Z.; FERREIRA, F. The role of working memory in syntactic ambiguity resolution: a psychometric approach. Journal of Experimental Psychology: General, 136(1), p.64-81, 2007. 30