Abstract

Sentence comprehension requires the retrieval of single word information from long-term memory, and the integration of this information into multiword representations. The current functional magnetic resonance imaging study explored the hypothesis that the left posterior temporal gyrus supports the retrieval of lexical-syntactic information, whereas left inferior frontal gyrus (LIFG) contributes to syntactic unification. Twenty-eight subjects read sentences and word sequences containing word-category (noun–verb) ambiguous words at critical positions. Regions contributing to the syntactic unification process should show enhanced activation for sentences compared to words, and only within sentences display a larger signal for ambiguous than unambiguous conditions. The posterior LIFG showed exactly this predicted pattern, confirming our hypothesis that LIFG contributes to syntactic unification. The left posterior middle temporal gyrus was activated more for ambiguous than unambiguous conditions (main effect over both sentences and word sequences), as predicted for regions subserving the retrieval of lexical-syntactic information from memory. We conclude that understanding language involves the dynamic interplay between left inferior frontal and left posterior temporal regions.

Introduction

We hear and see thousands of words each day, and effortlessly interpret them in their context. To achieve this, several intricate processes are engaged by the brain. Whatever model of language comprehension one adheres to, all make the general distinction between retrieval and integration processes. Retrieval entails selecting the lexical representation of a word from memory. Information we have learned about single words has been laid down in long-term memory, in what psycholinguists usually call the “mental lexicon” (e.g., Levelt 1992). This information includes a word's form, its syntactic properties (e.g., word class, gender), and the meaning of a lexical item. In order to understand single words we have to map the input signal onto word form representations in the mental lexicon (access) and select the corresponding lexical representation (Marslen-Wilson 1987). In this way the information associated with the word form is retrieved. However, 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 (von Humboldt 1836). This process of combining the retrieved single word information into higher-level representations, has been called “integration” or “unification” (Marslen-Wilson 1987; Hagoort et al. 1999; Sag and Wasow 1999; Kempen and Harbusch 2002). Both memory and unification processes occur in parallel at the semantic (conceptual) and at the syntactic (structural) level (Jackendoff 2002). The current functional magnetic resonance imaging (fMRI) study focuses on the syntactic level. We aim to disentangle the syntactic retrieval and unification processes and identify their respective neural correlates.

Computational Model

Recently, in linguistic theories the separation between lexical items and traditional rules of grammar is fading. Increasing support has been found for lexicalist parsing models (Macdonald et al. 1994; Joshi and Schabes 1997; Jackendoff 2002). In lexicalist models syntactic properties of words needed for integration are retrieved from the lexicon (from memory), and the only remaining “rule” is unification (Jackendoff 2002).

A lexicalist model that is computationally explicit is the Unification Space Model of parsing by Vosse and Kempen (2000). This model accounts for a large series of empirical findings in the parsing literature and in the neuropsychological literature on aphasia. According to the Unification Space Model, every incoming word retrieves one or more lexical frames from the “mental lexicon” These lexical frames are elementary syntactic trees, specifying the possible structural environment of the particular input word. See Figure 1 for examples of lexical frames for a noun (“flights”) and a verb (“flee”). This parsing account is “lexicalist” in the sense that all syntactic information associated with a lexical item is retrieved from the mental lexicon rather than being supplied by grammatical rules (Vosse and Kempen 2000; for a similar account, see Culicover and Jackendoff 2005).

Figure 1.

Lexical frames for the noun vluchten “flights” and the verb vluchten “flee.” When encountering the word vluchten, the lexical frame of both the noun and the verb will be retrieved. In the Vosse and Kempen model, lexical frames consist of 3-tiered treelets. The top-layer of a frame consists of a single phrasal node (e.g., NP). This “root”-node is connected to one or more functional nodes in the second layer (e.g., subject, head, direct object, modifier). Every functional node is linked to again a phrasal node in the third layer. Each lexical frame is attached to one lexical item (situated under the “head” node).

Figure 1.

Lexical frames for the noun vluchten “flights” and the verb vluchten “flee.” When encountering the word vluchten, the lexical frame of both the noun and the verb will be retrieved. In the Vosse and Kempen model, lexical frames consist of 3-tiered treelets. The top-layer of a frame consists of a single phrasal node (e.g., NP). This “root”-node is connected to one or more functional nodes in the second layer (e.g., subject, head, direct object, modifier). Every functional node is linked to again a phrasal node in the third layer. Each lexical frame is attached to one lexical item (situated under the “head” node).

Lexical frames that are retrieved from memory will enter the Unification Space in a sequential fashion, as new words arrive as input. Then, in Unification Space binding operations between lexical frames are performed, resulting in an incremental structural interpretation of the sentence. During the unification process lexical frames are linked, and agreement features (number, gender, person, etc.) and word order constraints are applied. Unification links are dynamic, that is, the strength of the unification links varies over time until one stable phrasal configuration results. As language is intrinsically ambiguous, often several different unification possibilities exist. In the Unification Space model, selection among alternative unification links occurs via lateral inhibition (Vosse and Kempen 2000; see Thompson-Schill 2005 on selection).

Brain Regions

With the Unification Space Model as our theoretical framework, can we identify distinct neural systems subserving lexical-syntactic retrieval and syntactic unification? Large areas within left frontal and temporoparietal cortex are involved in language processing (e.g., Ojemann 1991; Vigneau et al. 2006). Within the perisylvian cortex there is some evidence for a distribution of labor between lexical retrieval and unification operations over temporal and inferior frontal regions respectively (Hagoort 2005b). The left temporal cortex plays a central role in the storage and retrieval of information about single words, information that has been encoded during language acquisition and stored in long-term memory. Information about the meaning of words is probably distributed over a number of brain regions, but there are indications that different parts of the left middle and inferior temporal gyri are most important for lexical-semantic processing (see e.g., Damasio et al. 1996; Saffran and Sholl 1999; Hickok and Poeppel 2000, 2004; Keller et al. 2001; Indefrey and Cutler 2005). Apart from extensive literature investigating the differential processing of nouns and verbs (see e.g., Caramazza and Hillis 1991; Damasio and Tranel 1993; Pulvermuller et al. 1999; Shapiro et al. 2006; Longe et al. 2007; Vigliocco et al. 2008), hardly anything is known about the brain regions involved in the lexical retrieval of a word's syntactic properties (grammatical gender, syntactic frames, etc.). Based on a meta-analysis of syntactic processing (Indefrey 2004), Hagoort (2003, 2005) hypothesized that left posterior temporal cortex might be involved in the retrieval of lexical-syntactic information, including the lexical-syntactic frames that form the building blocks for syntactic unification.

The left inferior frontal gyrus (LIFG) of the human brain might be particularly suitable for performing unification operations. The lateral prefrontal cortex—of which the LIFG is a part—plays a fundamental role in integration, in particular the integration of information in the temporal domain (Fuster et al. 2000; Fuster 2001, 2002). The prefrontal cortex is capable of actively maintaining representations of various forms of information, through recurrent circuits between prefrontal cortex and posterior cortex (Durstewitz et al. 2000; Fuster 2001; Miller and Cohen 2001). Furthermore, in lateral prefrontal cortex these representations can be manipulated and flexibly updated. For instance, the left inferior frontal cortex is involved in selection among competing sources of information (Thompson-Schill et al. 1999, 2005). The properties of LIFG (the ability to maintain information online, manipulate, select and temporally integrate this information) make it an ideal candidate for a brain region contributing to combinatorial (unification) operations in language (Hagoort 2005a, 2005b).

Thus, the hypothesis is that the left posterior temporal cortex and the LIFG are involved in the retrieval of lexical-syntactic information and the unification of this information, respectively (Hagoort 2003, 2005b). However, the evidence for this hypothesis is limited and indirect, coming from studies with widely varying design and stimulus materials. No study systematically manipulated both lexical-syntactic retrieval and syntactic unification. Here we set out to disentangle the retrieval and unification processes in sentence structure comprehension within one study, using the computationally explicit model by Vosse and Kempen (2000) as our starting point.

Ambiguity Processing in Sentences

In order to achieve this goal we exploited word-category (noun–verb) ambiguous words. These words can be interpreted as a noun or as a verb, such as the words “bike” or “trains.” Behavioral and electroencephalogram studies have shown that multiple meanings of ambiguous words are transiently activated, even in a strongly disambiguating context (see e.g., Swinney 1979; Seidenberg et al. 1982; Duffy et al. 1988; Federmeier et al. 2000; Van Petten 2002; Swaab et al. 2003). Relatively few neuroimaging studies have addressed the processing of lexical ambiguities within sentence comprehension (Stowe et al. 1994; Stowe et al. 1998; Rodd et al. 2005; Zempleni et al. 2007).

In an early positron emission tomography study, Stowe and colleagues presented subjects with sentences containing a noun–verb ambiguous word in a neutral context. The sentence remained ambiguous for at least 3 words, and then was disambiguated into the least frequent (subordinate) meaning of the ambiguous word. The ambiguous sentences elicited more activity in the LIFG than the unambiguous sentences (Stowe et al. 1994, 1998).

More recently, Rodd and colleagues compared “high-ambiguity” sentences that included several ambiguous words (e.g., “the shell was fired towards the tank”) with “low-ambiguity” sentences (“her secrets were written in her diary”). Besides noun-noun ambiguities their materials included noun–verb ambiguities as well (e.g., lock, beam). Rodd and colleagues found increased activation for high-ambiguity relative to low-ambiguity sentences in left posterior inferior temporal cortex and bilateral inferior frontal gyri (Rodd et al. 2005).

Zempleni and colleagues showed participants sentences that contained noun-noun ambiguities of which one meaning was more frequent than the other (i.e., they were unbalanced). The final words of the sentence disambiguated the meaning to the dominant (most frequent), or the subordinate (less frequent) interpretation of the homograph. The sentences to a subordinate interpretation showed enhanced blood oxygenation level–dependent (BOLD) activation compared to sentences with a dominant interpretation in the left posterior and right anterior inferior frontal gyri and left posterior and right mid inferior/middle temporal gyri. Only the LIFG showed, in addition, more activity for the sentences with a dominant interpretation than for unambiguous sentences (Zempleni et al. 2007).

The above-mentioned studies cannot yet unravel the relative contributions of the lexical retrieval and unification processes as part of ambiguity resolution. Sentences containing lexical ambiguities tax both retrieval and unification processes stronger than unambiguous sentences (see below). Thus, although these previous studies do indicate that LIFG and left temporal gyrus (LTG) are involved in retrieval and unification, they do not disentangle the 2 processes, nor do they target specifically the retrieval and unification of syntactic information. This is what we set out to do in the current study.

Design and Predictions

To disentangle the lexical retrieval and unification processes related to syntactic information we presented subjects with (Dutch) sentences and with matched scrambled word sequences. In both sentences and word sequences the critical word was either word-class (noun/verb) ambiguous or unambiguous (see Table 1). The noun–verb ambiguous words were “balanced” in the sense that the noun and verb meaning had similar frequencies. The context preceding the ambiguous word was always neutral. For example, in the Dutch sentence beginning with Beide vluchten…, the word vluchten can be either a noun (flights) or a verb (flee), resulting in continuations such as for example: Beide vluchten werden geannuleerd (“both flights were cancelled”; disambiguation toward the noun reading for vluchten) or Beide vluchten het behekste huis uit (“both flee the bewitched house”; disambiguation toward the verb reading). The logic of this design, formalized in terms of the computational model of Vosse and Kempen, is as follows.

Table 1

Example of the experimental materials, with the critical word bewijzen (evidence/to prove)

SAn: sentence ambiguous (noun context) 
    Zodra jullie bewijzen(n/v) leveren kunnen we beginnen. 
    As-soon-as you evidence(n/v) provide can we start. 
    (As soon as you provide evidence(n/v) we can start.) 
SUn: sentence unambiguous (noun context) 
    Zodra jullie kopij(n) leveren kunnen we beginnen. 
    As-soon-as you copy(n) provide can we start. 
    (As soon as you provide copy(n) we can start.) 
SAv: sentence ambiguous (verb context) 
    Zodra jullie bewijzen(n/v) dat hij erbij betrokken is arresteren we hem. 
    As-soon-as you prove(n/v) that he in-it involved is arrest we him. 
    (As soon as you prove(n/v) that he is involved we will arrest him.) 
SUv: sentence unambiguous (verb context) 
    Zodra jullie beweren(v) dat hij erbij betrokken is arresteren we hem. 
    As-soon-as you claim(v) that he in-it involved is arrest we him. 
    (As soon as you claim(v) that he is involved we will arrest him.) 
WAn: words ambiguous (derived from SAn) 
    Genoemd tegen bewijzen(n/v) uit helaas gezeten jullie 
    Named against proof/prove(n/v) from alas seated you 
WUn: words unambiguous (derived from SUn) 
    Genoemd tegen kopij(n) uit helaas gezeten jullie 
    Named against copy(n) from alas seated you 
WAv: words ambiguous (derived from SAv) 
    In nogal bewijzen(n/v) meestal maar dit in struikelen hem verschil opeens 
    In quite proof/prove(n/v) mostly but this in stumble him difference suddenly 
WUv: words unambiguous (derived from SUv) 
    In nogal beweren(v) meestal maar dit in struikelen hem verschil opeens 
    In quite claim(v) mostly but this in stumble him difference suddenly 
SAn: sentence ambiguous (noun context) 
    Zodra jullie bewijzen(n/v) leveren kunnen we beginnen. 
    As-soon-as you evidence(n/v) provide can we start. 
    (As soon as you provide evidence(n/v) we can start.) 
SUn: sentence unambiguous (noun context) 
    Zodra jullie kopij(n) leveren kunnen we beginnen. 
    As-soon-as you copy(n) provide can we start. 
    (As soon as you provide copy(n) we can start.) 
SAv: sentence ambiguous (verb context) 
    Zodra jullie bewijzen(n/v) dat hij erbij betrokken is arresteren we hem. 
    As-soon-as you prove(n/v) that he in-it involved is arrest we him. 
    (As soon as you prove(n/v) that he is involved we will arrest him.) 
SUv: sentence unambiguous (verb context) 
    Zodra jullie beweren(v) dat hij erbij betrokken is arresteren we hem. 
    As-soon-as you claim(v) that he in-it involved is arrest we him. 
    (As soon as you claim(v) that he is involved we will arrest him.) 
WAn: words ambiguous (derived from SAn) 
    Genoemd tegen bewijzen(n/v) uit helaas gezeten jullie 
    Named against proof/prove(n/v) from alas seated you 
WUn: words unambiguous (derived from SUn) 
    Genoemd tegen kopij(n) uit helaas gezeten jullie 
    Named against copy(n) from alas seated you 
WAv: words ambiguous (derived from SAv) 
    In nogal bewijzen(n/v) meestal maar dit in struikelen hem verschil opeens 
    In quite proof/prove(n/v) mostly but this in stumble him difference suddenly 
WUv: words unambiguous (derived from SUv) 
    In nogal beweren(v) meestal maar dit in struikelen hem verschil opeens 
    In quite claim(v) mostly but this in stumble him difference suddenly 

Ambiguity Effect: Retrieval from Mental Lexicon

In a neutral context both noun and verb reading of a noun–verb ambiguous word are activated (Seidenberg et al. 1982; Duffy et al. 1988). That is, presentation of a noun–verb ambiguous word triggers the retrieval of both the noun and the verb version of the ambiguous word. As 2 lexical frames are retrieved instead of one (see Fig. 1), ambiguous words tax the lexical-syntactic retrieval process more heavily than unambiguous words. This ambiguity effect should occur in sentences as well as word sequences.

Grammaticality Effect: Unification Process

Whereas the factor Ambiguity manipulates the retrieval process, the Grammaticality factor is hypothesized to affect the unification process. When we compare sentences and word sequences, the sentences will require unification of the words into an overall sentence structure, whereas the word sequences do not. Thus, sentences will induce a higher Unification load.

Interaction: Unification Process

Given that the Grammaticality effect is not specific to syntactic unification (as there will be general semantic and phonological differences between sentences and word sequences as well), the crucial effect in our design is the interaction between Grammaticality and Ambiguity. In the sentence condition the 2 retrieved lexical frames (noun and verb) will compete for unification via lateral inhibition (the selection mechanism in our account). In the word condition no unification occurs, and the ambiguous words do not impose a higher unification load than their unambiguous counterparts. Thus, crucially, unification load will only be affected by ambiguity in the sentence condition, and not in the “random” word condition.

Because we hypothesize LIFG to be contributing to the unification process, we expect that LIFG will be activated more for sentences than for word sequences, and within the sentences more for ambiguous than for unambiguous words. We predict the lexical-syntactic retrieval processes to occur in the left posterior temporal cortex. The ambiguous words will induce a higher lexical retrieval load than the unambiguous words in both word sequences and sentences. Hence, we expect the Ambiguity effect to identify the temporal area as subserving the retrieval of lexical frames.

Materials and Methods

Participants

Twenty-eight right-handed healthy volunteers (14 females, aged 18–35) participated in the experiment after having given written informed consent. Subjects were paid for their participation. All participants were native speakers of Dutch, without any history of neurological illness or head injury. Six additional subjects were scanned but excluded from analysis because of excessive movement in the MR scanner (2 subjects) or poor task performance (4 subjects, see below).

Stimulus Material

The stimulus material consisted of 68 (Dutch) sentences (S) and 68 matched scrambled sequences of Dutch words (W). Both the Sentences and the Word sequences contained a critical word that was either word-class (noun/verb) ambiguous (A) or unambiguous (U). The critical word in the sentences was disambiguated by the continuation of the sentence into either a noun (n) or a verb (v) reading. In total, this resulted in 8 possible conditions: SAn, SAv, SUn, SUv, WAn, WAv, WUn, WUv (see e.g., Table 1). For the full set of experimental materials, see the Supplementary Appendix.

Ambiguous Critical Words

Word-class ambiguous words were selected from the Dutch lexical databases CELEX (Baayen et al. 1993) and CLEF (Beek et al. 2001). Selected words had to be both a noun and a verb (and should not belong to any other word category). To ensure that both noun and verb meaning would be initially activated, all selected ambiguous words had a noun-ratio (noun frequency/summed frequency) between 0.25 and 0.75. We required that the 2 databases (CELEX and CLEF) agree on all criteria. In this way we got a highly reliable indication of frequency and noun-ratio. The average noun-ratio of the single ambiguous critical words was 0.51.

Ambiguous Sentences

The ambiguous sentences were constructed such that both categories of the critical word fitted syntactically as well as semantically with the initial part of the sentences (up to and including the critical word); the sentences were disambiguated by the subsequent part of the sentence (after the ambiguous word; see Table 1 for an example). A pretest was conducted in order to assess the noun-ratio of the words in the sentence context. In this pretest, subjects (n = 38) had to complete sentences (e.g., “Beide vluchten…”). For all items the “pretest noun-ratio” was defined as the percentage of subjects that completed the sentence in accordance with a noun interpretation of the ambiguous word. Only critical items with a “pretest noun-ratio” between 0.2 and 0.8 were selected. Based on the pretest selection procedure, 68 suitable ambiguous sentence-beginnings were selected, with an average noun-ratio of 0.46.

For every ambiguous item a noun and a verb sentence ending was constructed, with the same neutral sentence context preceding the critical word (SAn and SAv, see an example in Table 1). Noun and verb sentences were matched for average length. The sentences consisted of 6–12 words, with an average of 8.5 words per sentence. The critical word occurred on the second (34 sentences), third (27 sentences), or fourth (7 sentences) position of the sentence. None of the noncritical words in the sentences were balanced noun–verb ambiguous words.

Unambiguous Sentences

For every ambiguous sentence item we constructed 2 suitable, unambiguous alternatives for the ambiguous critical word (a noun and a verb; see example SUn and SUv in Table 1) that fitted the remaining part of the sentence equally well as the ambiguous word. We selected part of the alternative words from the CLEF-corpus based on distributional similarity (Plas and Bouma 2004). Words that are distributionally similar are words that share a large number of lexical-syntactic contexts, that is, they form grammatical dependency relations with the same words (e.g., lemon and orange can both be the direct object of squeeze). If for a critical word no such alternative could be found in the corpus, it was constructed by hand. Ambiguous and unambiguous words were matched for average length and (summed) word form frequency.

Word Sequences

Word sequences were constructed from the sentences (WAn, WUn, WAv, WUv). For each sentence, every word (except the critical word) was substituted by a different corpus word belonging to the same syntactic category (noun, verb, adjective, rest), and with a similar length and frequency (the resulting average length for both sentences and word sequences was 42 characters, and the average CLEF/CELEX log frequency was 3.7 for sentences compared with 4.2 for word sequences). Subsequently the order of all words in the sequence (except the critical word) was randomized, with the constraint that 2 words could not be succeeding each other in the same way as in the original sentence. Sequences were checked on local grammaticality and were scrambled again if 3 or more consecutive words formed a coherent sentence fragment. We again made sure that none of the noncritical words in the sequence were balanced noun–verb ambiguous words.

Experimental Lists

The stimuli were distributed over 4 lists. For every item, one of the following combinations occurred in each list: SAn+WUn; SAv+WUv; SUn+WAn; SUv+WAv; hence, no subject encountered the same critical word twice. This resulted in 34 items per subject per condition (SA, SU, WA, WU). The pretest noun-ratios, sentence lengths, and position and frequency of the critical words were all matched across the 4 lists. In addition, each list contained the same 28 sentence and 28 word sequence fillers, 36 of which (18 for each condition) contained a consonant string (e.g., grpsd) at various positions in the sentence/sequence (see Procedure).

Procedure

Stimuli were presented visually in serial presentation mode (word by word in the middle of the screen) using the Presentation software (Version 9.13, www.neuro-bs.com). Every word remained on the screen for 300 ms, with a 200-ms interword interval. Between sentences a visual fixation cross was presented for 5–8 s (low-level baseline). The participants were instructed to read each sentence/sequence carefully and attentively, and were told that after the experiment some questions concerning the experiment would have to be answered. The participants’ task was spotting the consonant-strings (e.g., cdsnl), that were presented in 36 of the fillers. This simple control task was added to check whether subjects were paying attention. Subjects were defined as poor task performers if they made more than 5 errors (missing hits and false alarms) on the task, suggesting that these subjects did not pay enough attention to the stimuli.

Every subject saw 68 sentences and 68 word sequences (ambiguous/unambiguous; in noun/verb version), intermingled with 56 fillers (28 sentences and 28 sequences). Stimuli were presented in mini-blocks of 3 to 4 sentences or word sequences. All mini-blocks were shorter than 40 s. Before each block the label “Zinnen:”(“Sentences:”) or “Woorden:” (“Words:”) appeared on the screen (for 1.5 s) to indicate the condition of the following mini-block, which started after a fixation cross of 1–3 s. We expected the labels to encourage (“Sentences”) or discourage (“Words”) attempts to syntactically/semantically integrate the stimulus items in the upcoming mini-block. The ambiguous/unambiguous and verb/noun conditions were intertwined within the mini-blocks in a pseudo-randomized presentation order.

fMRI Data Acquisition

During the sentence/sequence presentation we acquired T2*-weighted EPI-BOLD fMRI data with a SIEMENS Trio 3T MR scanner using an ascending slice acquisition sequence (volume time repetition = 2 s, time echo [TE] = 35 ms, 90° flip-angle, 29 slices, slice-matrix size = 64 × 64, slice thickness = 3 mm, slice gap = 0.5 mm, field of view = 224 mm, isotropic voxel size = 3.5 × 3.5 × 3.5 mm3). At the end of the scanning session, a structural MR image volume was acquired for which a high-resolution T1-weighted 3D MP–RAGE sequence was used (TE = 3.93 ms, 8° flip-angle, 192 sagittal slices, slice thickness = 1.0 mm, voxel size = 1 × 1 × 1 mm3).

Data Analysis

Image preprocessing and statistical analysis were performed using Statistical Parametric Mapping (SPM2; www.fil.ion.ucl.ac.uk/spm). The first 5 image volumes were discarded in order to avoid transient non-saturation effects. The functional EPI-BOLD images were realigned, slice-time corrected, and the subject-mean functional MR images were coregistered with the corresponding structural MR images using mutual information optimization. Subsequently, images were normalized onto a Montreal Neurological Institute (MNI)–aligned echo planar imaging template (based on 28 male brains acquired on the Siemens Trio at the F. C. Donders Centre) and resampled to an isotropic voxel size of 2 mm. Finally, the normalized images were spatially filtered by convolving the functional images with an isotropic 3D Gaussian kernel (10 mm full width at half maximum).

The fMRI data were proportionally scaled to account for various global effects, and analyzed statistically using the general linear model and statistical parametric mapping (Friston et al. 1995) in a 2-step mixed design procedure. At the first-level, single-subject fixed effect analyses were conducted. The linear model included mini-block regressors to model the sentence/sequence presentation from the onset of the critical word to the offset of the sentence/sequence-final word. The beginnings of sentences/sequences and filler items were modeled together as a regressor of no interest (other words, OW), and the presentation of the fixation cross (FIX) was modeled as explicit baseline. We temporally convolved the explanatory variables with the canonical hemodynamic response function provided by SPM2. We included the realignment parameters for movement artifact correction and a temporal high-pass filter (cut-off 128 s) to account for various low-frequency effects as effects of no interest. Temporal autocorrelation was modeled as a first-order plus white noise autoregressive process.

ROI Analysis

A meta-analysis (Bookheimer 2002) suggests that activations related to syntactic processing in LIFG are centered in a 13-mm sphere around MNI coordinates [−44, 19, 14] (Petersson et al. 2004). This sphere was taken as the region of interest (ROI) of the relevant subpart of the LIFG (which henceforth we will simply refer to as LIFG). An average time course was calculated for LIFG (for every participant separately) using Marsbar (http://marsbar.sourceforge.net/). Although we had a strong a priori hypothesis regarding the part of LIFG involved in syntactic unification, we did not have such a specifically defined region for the part of left posterior temporal cortex involved in lexical-syntactic retrieval. Thus, we specified a region of interest only for LIFG, and did not use an ROI for LTG. For the ROI analysis at the second level a repeated measure ANOVA with the factors Grammaticality (S,W), Ambiguity (A,U), and Word class (n,v) was carried out on the subject contrast values using the SPSS software (SPSS Inc., Chicago, IL).

Whole-Brain Analysis

For the second-level whole-brain analysis, we generated single-subject contrast images for the SAn, WAn, SUn, WUn, SAv, WAv, SUv, and WUv items relative to the baseline FIX, and used these in a one-way random effects repeated measures ANOVA (including the factors: condition [8] and subject [28]). To correct for multiple comparisons, statistical inference was based on the cluster-size statistics from the relevant second-level SPM[T] volumes (Forman et al. 1995; Friston et al. 1996). SPMs were thresholded at P < 0.002 (uncorrected at the voxel level). To protect against false-positive results, only clusters of a size of 50 voxels or more are reported (unless otherwise specified). In this way SPM[T] volumes were generated to investigate the effect of grammaticality (i.e., sentences > words; words > sentences; Fig. 3A; Tables 2 and 3) and the effects of ambiguity (ambiguous > unambiguous; unambiguous > ambiguous, Fig. 3B, Table 4). Additionally, an SPM[T] volume was created for the effect of ambiguity within sentences only (SA > SU, Fig. 3C, Table 5).

Table 2

Sentences versus word sequences

Region  BA Cluster size Voxel T189 value x y z 
Sentences > words 
    L temporal gyrus, L IFG 13 262     
    L temporal pole 38  24.21 −54 18 −30 
    L anterior MTG 21  22.01 −56 −6 −16 
    L posterior MTG 21  18.77 −62 −44 −2 
    L IFG 47  17.13 −52 34 −8 
    L posterior STG 22  15.12 −58 −56 12 
    L IFG 45  15.04 −58 22 12 
    L posterior STG 39  14.22 −44 −58 18 
    L amygdala   10.66 −26 −6 −20 
    L ITG 20  9.56 −44 −16 −30 
    L culmen (cerebellum)   8.61 −26 −36 −26 
    L putamen   8.28 −18 
    R temporal gyrus  3138     
    R temporal pole 38  16.22 54 20 −32 
    R anterior MTG 21  13.39 56 −26 
    R middle MTG 21  10.24 52 −14 −16 
    R posterior MTG 21  6.96 62 −42 
    R IFG  370     
    R anterior IFG 47  8.85 56 36 −10 
    R IFG 45  6.5 60 34 
    R posterior IFG 45  5.94 62 28 10 
    L SFG/mOrbG 11 328 10.85 −4 54 −20 
    L anterior SFG 227 13.01 −8 60 28 
    L MFG 208 7.55 −38 −2 −50 
    L Lingual gyrus  79 5.35 −6 −62 
    L posterior cingulate gyrus  78 6.88 −12 −46 34 
    R Culmen (cerebellum)  62 6.01 30 −32 −32 
        
Region  BA Cluster size Voxel T189 value x y z 
Sentences > words 
    L temporal gyrus, L IFG 13 262     
    L temporal pole 38  24.21 −54 18 −30 
    L anterior MTG 21  22.01 −56 −6 −16 
    L posterior MTG 21  18.77 −62 −44 −2 
    L IFG 47  17.13 −52 34 −8 
    L posterior STG 22  15.12 −58 −56 12 
    L IFG 45  15.04 −58 22 12 
    L posterior STG 39  14.22 −44 −58 18 
    L amygdala   10.66 −26 −6 −20 
    L ITG 20  9.56 −44 −16 −30 
    L culmen (cerebellum)   8.61 −26 −36 −26 
    L putamen   8.28 −18 
    R temporal gyrus  3138     
    R temporal pole 38  16.22 54 20 −32 
    R anterior MTG 21  13.39 56 −26 
    R middle MTG 21  10.24 52 −14 −16 
    R posterior MTG 21  6.96 62 −42 
    R IFG  370     
    R anterior IFG 47  8.85 56 36 −10 
    R IFG 45  6.5 60 34 
    R posterior IFG 45  5.94 62 28 10 
    L SFG/mOrbG 11 328 10.85 −4 54 −20 
    L anterior SFG 227 13.01 −8 60 28 
    L MFG 208 7.55 −38 −2 −50 
    L Lingual gyrus  79 5.35 −6 −62 
    L posterior cingulate gyrus  78 6.88 −12 −46 34 
    R Culmen (cerebellum)  62 6.01 30 −32 −32 
        

Note: Significant activation peaks >8 mm apart (P < 0.05 FWE corrected, cluster extent threshold 20 voxels). Multiple peaks within a single activation cluster are shown indented. BA = Brodmann's area; T189 value = T value for 189 degrees of freedom; x,y,z = the original SPM x,y,z coordinates in millimeters of the MNI space; STG = superior temporal gyrus; MTG = middle temporal gyrus; ITG = inferior temporal gyrus; MFG = middle frontal gyrus; IFG = inferior frontal gyrus; mOrbG = medial Orbital Gyrus.

Table 3

Word sequences versus sentences

Region BA Cluster size Voxel T189 value x y z 
Words > sentences 
    Bilateral M/SFG, CG  12 144     
    R anterior MFG 11  11.21 36 58 −12 
    R CG/med SFG 32  11.15 36 30 
    R posterior MFG  10.88 40 26 42 
    L anterior MFG 10  10.61 −38 58 
    R anterior I/MFG 46  10.31 48 44 16 
    L MFG  10.25 −40 34 32 
    R frontopolar G 10  10.25 24 64 
    L anterior M/SFG 10  9.24 −30 56 −4 
    R AC/rostral G 32  8.43 46 −4 
    R anterior SFG  8.20 28 44 36 
    L AC/rostral G 32  6.94 −8 40 
    R middle CG 23 1704 13.64 −24 30 
    R SMG 40 1280 12.25 58 −50 36 
    L sup parietal lob 772 9.76 −10 −72 30 
    R precuneus  9.07 −12 −62 34 
    L angular G /SMG  125     
    L angular gyrus 40  7.19 −36 −50 36 
    L SMG 40  6.06 −44 −44 38 
    L SMG 40  5.24 −60 −38 40 
    Planum polare/orbital G  82 6.69 32 22 −20 
    L posterior ITG  31 6.23 −58 −54 −20 
Region BA Cluster size Voxel T189 value x y z 
Words > sentences 
    Bilateral M/SFG, CG  12 144     
    R anterior MFG 11  11.21 36 58 −12 
    R CG/med SFG 32  11.15 36 30 
    R posterior MFG  10.88 40 26 42 
    L anterior MFG 10  10.61 −38 58 
    R anterior I/MFG 46  10.31 48 44 16 
    L MFG  10.25 −40 34 32 
    R frontopolar G 10  10.25 24 64 
    L anterior M/SFG 10  9.24 −30 56 −4 
    R AC/rostral G 32  8.43 46 −4 
    R anterior SFG  8.20 28 44 36 
    L AC/rostral G 32  6.94 −8 40 
    R middle CG 23 1704 13.64 −24 30 
    R SMG 40 1280 12.25 58 −50 36 
    L sup parietal lob 772 9.76 −10 −72 30 
    R precuneus  9.07 −12 −62 34 
    L angular G /SMG  125     
    L angular gyrus 40  7.19 −36 −50 36 
    L SMG 40  6.06 −44 −44 38 
    L SMG 40  5.24 −60 −38 40 
    Planum polare/orbital G  82 6.69 32 22 −20 
    L posterior ITG  31 6.23 −58 −54 −20 

Note: Significant activation peaks > 8 mm apart (P < 0.05 FWE corrected, cluster extent threshold 20 voxels). Multiple peaks within a single activation cluster are shown indented. AC = anterior cingulate; SMG = supramarginal gyrus; CG = cingulate gyrus; Sup Parietal Lob = superior parietal lobule; for additional abbreviations see Table 2.

Table 4

Ambiguous versus unambiguous conditions

Region BA Cluster size Voxel T189 value x y z 
Ambiguous > unambiguous 
    R striatum  502     
    R putamen   3.96 24 
    R caudate body   3.94 14 16 
    R putamen   3.83 26 −10 
    L posterior I/MTG  109     
    L post-MTG / ITS 37  3.82 −52 −50 −8 
    L posterior ITG 37  3.27 −46 −46 −14 
    L post-MTG / ITS 37  3.11 −46 −54 −4 
    R MTG 37 72 3.57 48 −44 −6 
    WM/PCL/CG  56 3.61 −22 −38 34 
    R parahippocampal gyrus  50     
    R para-HCG 36  3.28 22 −42 −6 
    R para-HCG/occipital G   3.19 28 −48 −2 
Unambiguous > ambiguous: 
     no significant activations 
Region BA Cluster size Voxel T189 value x y z 
Ambiguous > unambiguous 
    R striatum  502     
    R putamen   3.96 24 
    R caudate body   3.94 14 16 
    R putamen   3.83 26 −10 
    L posterior I/MTG  109     
    L post-MTG / ITS 37  3.82 −52 −50 −8 
    L posterior ITG 37  3.27 −46 −46 −14 
    L post-MTG / ITS 37  3.11 −46 −54 −4 
    R MTG 37 72 3.57 48 −44 −6 
    WM/PCL/CG  56 3.61 −22 −38 34 
    R parahippocampal gyrus  50     
    R para-HCG 36  3.28 22 −42 −6 
    R para-HCG/occipital G   3.19 28 −48 −2 
Unambiguous > ambiguous: 
     no significant activations 

Note: Significant activation peaks > 8 mm apart (P < 0.002 uncorrected, cluster extent threshold 50 voxels). Multiple peaks within a single activation cluster are shown indented. ITS = inferior temporal sulcus; WM = white matter; PCL = paracentral lobule; CG = cingulate gyrus; paraHCG = parahippocampal gyrus; for additional abbreviations see Table 2.

Table 5

Ambiguous versus unambiguous, separately for sentence and word sequence conditions

Region BA Cluster size Voxel T189 value x y z 
SA > SU 
    L posterior I/MTG  290     
    L post-MTG/ITS 37  4.48 −52 −50 −8 
    L post-MTG 21  3.42 −60 −44 −4 
    L post-MTG/ITS 37  3.37 −46 −54 −4 
    L mid-ITG  73     
    L mid-ITG 20  4.02 −50 −20 −28 
    L fusiform gyrus 20  3.80 −42 −24 −28 
    L mid-ITG 20  3.18 −46 −30 −22 
R MTG  65     
    R ITS   3.54 48 −34 −14 
    R posterior MTG 37  3.31 50 −44 −4 
    L IFG/PrG 62 3.49 −44 22 
    R IFG  51     
    R IFG 45  3.35 46 28 
    R IFG 44/45  3.24 44 18 14 
    R IFG 44/45  2.99 54 18 12 
SU > SA; WA > WU; WU > WA: no significant activations 
Region BA Cluster size Voxel T189 value x y z 
SA > SU 
    L posterior I/MTG  290     
    L post-MTG/ITS 37  4.48 −52 −50 −8 
    L post-MTG 21  3.42 −60 −44 −4 
    L post-MTG/ITS 37  3.37 −46 −54 −4 
    L mid-ITG  73     
    L mid-ITG 20  4.02 −50 −20 −28 
    L fusiform gyrus 20  3.80 −42 −24 −28 
    L mid-ITG 20  3.18 −46 −30 −22 
R MTG  65     
    R ITS   3.54 48 −34 −14 
    R posterior MTG 37  3.31 50 −44 −4 
    L IFG/PrG 62 3.49 −44 22 
    R IFG  51     
    R IFG 45  3.35 46 28 
    R IFG 44/45  3.24 44 18 14 
    R IFG 44/45  2.99 54 18 12 
SU > SA; WA > WU; WU > WA: no significant activations 

Note: Significant activation peaks >8 mm apart (P < 0.002 uncorrected, cluster extent threshold 50 voxels). Multiple peaks within a single activation cluster are shown indented. SA = sentence ambiguous; SU = sentence unambiguous; WA = word sequence ambiguous; WU = word-sequence unambiguous; ITS = inferior temporal sulcus; PrG = precentral gyrus; for additional abbreviations see Table 2.

Anatomical Inference

All local maxima are reported as MNI coordinates (Evans et al. 1993). Relevant anatomical landmarks were identified and Brodmann areas were defined using the Atlas of the Human Brain (Mai et al. 2004) and the Talairach Daemon (Lancaster et al. 2000).

Results

Control Task

All 28 included participants made ≤ 3 errors (mean: 0.18 misses, 0.5 false alarms). Subjects that made more than 5 errors were excluded from analysis (4 subjects, see participants).

ROI Analysis

For the syntactic processing in the LIFG we had a very specific region of interest (see Data Analysis). Figure 2 shows the ROI, and the mean contrast estimates of LIFG for SA, SU, WA, and WU. LIFG was activated more strongly for sentences than for words (F1,27 = 31.2, P < 0.001), whereas only within sentences there was an effect of ambiguity (ambiguity × grammaticality: F1,27 = 5.6, P = 0.025; SA > SU: T(27) = 2.75, P = 0.005). This activation pattern corresponds to the activation predicted for the Unification Space, supporting the hypothesis that the LIFG plays a role in syntactic unification operations during language comprehension.

Figure 2.

ROI results. Mean contrast estimates for LIFG for SA, SU, WA, and WU. The ROI used is shown on top (13 mm sphere around coordinates [−44,19,14]).

Figure 2.

ROI results. Mean contrast estimates for LIFG for SA, SU, WA, and WU. The ROI used is shown on top (13 mm sphere around coordinates [−44,19,14]).

Figure 3.

Whole-brain analysis. Significant activations projected onto a rendered template brain surface in MNI stereotactic space. (A) Effect of grammaticality. Enhanced activity within sentences is shown in red, increased signal for words is shown in green. (Activations shown at voxel-level PFWE < 0.05, cluster-size threshold 20 voxels.) (B) Effect of ambiguity: enhanced activity for ambiguous as compared to unambiguous conditions. The right panel shows a coronal view of the brain at y = 0, displaying the ambiguity effect in the right striatum. (Activations shown at voxel-level Puncorr < 0.002, cluster-size threshold 50 voxels.) (C) Effect of ambiguity within sentences: enhanced activity for ambiguous sentences as compared to unambiguous sentences. (Activations shown at voxel-level Puncorr < 0.002, cluster-size threshold 50 voxels.)

Figure 3.

Whole-brain analysis. Significant activations projected onto a rendered template brain surface in MNI stereotactic space. (A) Effect of grammaticality. Enhanced activity within sentences is shown in red, increased signal for words is shown in green. (Activations shown at voxel-level PFWE < 0.05, cluster-size threshold 20 voxels.) (B) Effect of ambiguity: enhanced activity for ambiguous as compared to unambiguous conditions. The right panel shows a coronal view of the brain at y = 0, displaying the ambiguity effect in the right striatum. (Activations shown at voxel-level Puncorr < 0.002, cluster-size threshold 50 voxels.) (C) Effect of ambiguity within sentences: enhanced activity for ambiguous sentences as compared to unambiguous sentences. (Activations shown at voxel-level Puncorr < 0.002, cluster-size threshold 50 voxels.)

As mentioned in the Data Analysis section, we did not have an a priori specifically defined region of interest within the left posterior temporal cortex, so we used the whole-brain analysis (ambiguity effect) to identify the part of the LTG involved in the retrieval of lexical-syntactic information.

Whole-Brain Analysis

Grammaticality Effect

For the whole-brain comparison between sentences and word sequences a voxel-level threshold of 0.05 corrected for multiple comparisons based on random field theory (Worsley et al. 1996), and a cluster-size threshold of 20 voxels were used (Fig. 3A, Tables 2 and 3). As shown in Figure 3A, sentences caused increased activation in comparison with word sequences in a large area in the perisylvian network, including LIFG, left middle temporal gyrus (LMTG) and both temporal poles (see Table 2 for a complete list). This is the network that we would indeed expect to see activated in sentence comprehension (e.g., Bavelier et al. 1997; Friederici 2002; Indefrey and Cutler 2005; Vigneau et al. 2006). For word sequences there was enhanced activation in the bilateral middle/superior frontal gyrus (M/SFG), the cingulate cortex, and supramarginal/angular gyrus (see Table 3, Fig. 3A). This network of areas is reminiscent of the “default mode” network described by Raichle et al. (2001). Indeed, inspection of the beta weights showed that both word sequences and sentences deactivate these areas compared to low-level baseline, but sentences gave rise to a larger deactivation than word sequences.

Ambiguity Effect

Ambiguous words are expected to induce a larger load on the mental lexicon for both sentences and word sequences, as 2 lexical frames have to be retrieved from memory. We hypothesized these retrieval operations to take place in the left posterior temporal cortex. Indeed, we see an increased signal for the ambiguous compared to the unambiguous items in the left posterior middle temporal gyrus (LpMTG, see Fig. 3B). Furthermore, enhanced activation for ambiguous items was seen in the striatum (Fig. 3B), right posterior middle temporal gyrus and the right parahippocampal gyrus (see Table 4). There were no areas that showed larger activation for the unambiguous than for the ambiguous items. Figure 4 shows the mean contrast estimates for the activated cluster in LpMTG and the striatum (computed with Marsbar, http://marsbar.sourceforge.net/). It can be seen that the ambiguity effect in LpMTG was larger in the sentences than in the word sequences. An ANOVA on the mean contrast estimates in LpMTG showed a significant interaction between grammaticality and ambiguity, with the WA > WU comparison being marginally significant (Fgram x amb(1,27) = 6.98, P = 0.014; SA > SU: T(27) = 4.72, P < 0.001; WA > WU: T(27) = 1.40, P = 0.086, see Fig. 4). The effect of ambiguity in the striatum (the largest effect for this comparison, see Table 4) did not differ for sentences and word sequences (F < 1; see Fig. 4).

Figure 4.

Mean contrast estimates for LpMTG and the right striatum (clusters identified by the whole-brain ambiguity effect) for SA, SU, WA, and WU.

Figure 4.

Mean contrast estimates for LpMTG and the right striatum (clusters identified by the whole-brain ambiguity effect) for SA, SU, WA, and WU.

Ambiguity Effect in Sentences

Additionally, we explored the effect of ambiguity within sentences only (SA > SU). The ambiguous sentences showed enhanced activation in the bilateral posterior MTG and the bilateral posterior IFG compared to the unambiguous sentences (see Table 5, Fig. 3C). This is the same network of areas that was found previously in studies of lexical ambiguity in sentence processing (Rodd et al. 2005; Zempleni et al. 2007). The left inferior frontal region we identified in this contrast (SA > SU) is overlapping with, but slightly posterior to, our region of interest. To explore whether other areas within LIFG (besides our ROI) might be involved, we show the whole-brain sentence ambiguity effect at lower thresholds in Figure 5 (frontal cortex). In Figure 5B this SA > SU activation is masked by the Grammaticality effect (Sentences > Words), as this is the pattern we were looking for in our ROI (unification). Although there is some activation in the anterior LIFG at a low threshold, most of the activation for SA > SU is situated posterior of [−44 19 14] (the center of our ROI). For the masked activation, we see activation only in posterior LIFG (Fig. 5B).

Figure 5.

LIFG effect: comparison of ROI and whole-brain analysis (exploratory thresholds). Activations from the whole-brain analysis are displayed on a template (frontal cortex, sagittal view). Voxel-level Puncorr: yellow P < 0.01; pink P < 0.005; red P < 0.001. Crosshair at [−44 19 14] (ROI). (A) Effect of ambiguity within sentences (SA > SU). (B) Effect of ambiguity within sentences (SA > SU), masked (inclusively) with the effect of grammaticality (sentences > words) at PFWE < 0.05. Regions contributing to the syntactic unification process should show this pattern of activation.

Figure 5.

LIFG effect: comparison of ROI and whole-brain analysis (exploratory thresholds). Activations from the whole-brain analysis are displayed on a template (frontal cortex, sagittal view). Voxel-level Puncorr: yellow P < 0.01; pink P < 0.005; red P < 0.001. Crosshair at [−44 19 14] (ROI). (A) Effect of ambiguity within sentences (SA > SU). (B) Effect of ambiguity within sentences (SA > SU), masked (inclusively) with the effect of grammaticality (sentences > words) at PFWE < 0.05. Regions contributing to the syntactic unification process should show this pattern of activation.

Post Hoc Analysis on Context-Irrelevant Associates

It is the case that word-category ambiguous words are semantically ambiguous as well. Therefore, we conducted a post-test to investigate the semantic consequences of the ambiguity. The issue at hand is whether our results can be explained by a different amount of selection/inhibition of semantic information evoked by ambiguous and unambiguous words. A full description of the methods and results can be found in the supplementary materials. First, in a normative study we determined the semantic associates of the sentence onsets (e.g., Beide vluchten). Of these associates we identified the ones that were context-irrelevant (i.e., inhibited in the remainder of the sentence). Then we compared the number of context-irrelevant semantic associates for ambiguous and unambiguous conditions. Statistical analysis indicated that there were more context-irrelevant associates for ambiguous than unambiguous items (F1,43 = 46.16, P < 0.001). To investigate whether this difference in context-irrelevance could explain our fMRI results, we included the context-irrelevance score for each sentence item as a covariate in the fMRI analysis (see Supplementary Materials). Context-irrelevance showed no effect in LIFG or LpMTG and the effect of word-category ambiguity was very similar, irrespective of whether the context-irrelevance regressor was included as a covariate in the statistical analysis or not. This shows that the difference in the amount of context-irrelevant information evoked by ambiguous and unambiguous items cannot explain our fMRI results.

Discussion

Starting point of this study was the general distinction between retrieval and unification (integration) processes in language (Hagoort 2005b; Vosse and Kempen 2000). We explored the hypothesis that LIFG contributes to syntactic unification operations, whereas the left posterior temporal gyrus subserves the retrieval of lexical-syntactic information. To do this, we exploited word-category ambiguities in sentence and word sequence contexts. First, we predicted that the syntactic unification load should be larger for the sentence than for the word sequence condition, and larger for ambiguous than unambiguous items only within the sentences. This is exactly the activation pattern we found in the LIFG, supporting the hypothesis that LIFG is involved in the unification process. Second, we expected retrieval of lexical-syntactic frames to be more demanding in the ambiguous than in the unambiguous condition in both sentences and word sequences. Indeed, LpMTG showed a main effect of ambiguity, suggesting that LpMTG subserves the retrieval of lexical-syntactic information from the mental lexicon. Thus, our study could confirm the role of posterior LIFG and LpMTG in syntactic unification and lexical-syntactic retrieval processes, respectively. (The effect of ambiguity in sentences that we find could be an effect occurring at the ambiguous word [2 lexical frames enter Unification Space and are competing] as well as an effect at the disambiguating word [one lexical frame wins and is selected]. The low temporal resolution of fMRI does not permit us to distinguish between these 2 possibilities. We are currently running an MEG study using the same materials to look into this issue.)

LIFG as the Unification Space for Language

Evidence on the involvement of LIFG in unification/integration processes in language comprehension is accumulating (Hagoort 2005b; Willems et al. 2007). Our study confirms the contribution of the posterior LIFG to the syntactic part of the unification process. However, we do not claim that the processing role of LIFG is restricted to syntax or even to language in general (see e.g., Decety et al. 1997; Hamzei et al. 2003; Kaan and Swaab 2002; Patel 2003). Nor do we claim that unification is the only function of the LIFG. A different but related proposal is that LIFG is associated with selection of information among competing alternatives (Thompson-Schill et al. 2005). This is compatible with our results, as the outcome of the unification process is dependent on a selection mechanism that chooses between different unification options (Hagoort 2005a; Vosse and Kempen 2000). We prefer the term “unification” over “selection”, as we think “unification” is both a more general account of LIFG functioning and a computationally explicit process. For example, unification in LIFG occurs also for information that has no stable representation in long-term memory, and thus cannot be selected (such as co-speech gestures, see Willems and others 2007). Specific brain regions (such as LIFG) probably participate in a wide range of tasks, with specialized function emerging from the unique cooperation of a network of brain areas subserving domain-general mechanisms (Mesulam 1998; Fuster 2001; Marcus et al. 2003). Even if the processing role of LIFG as a whole is domain-general, due to the interplay with content specific areas there might emerge some specialization for unification “content,” with anterior LIFG being involved with semantic operations, and posterior LIFG with syntactic operations (see Hagoort 2005b). In our case, syntactic unification might be the result of posterior LIFG working together with representational areas in the posterior temporal lobe.

Retrieval in LpMTG: Sustained Activation in Dynamic Interplay with LIFG

We found the LpMTG (on the border of the inferior temporal sulcus) to be activated more strongly for ambiguous than unambiguous conditions. This ambiguity effect is what we predicted for retrieval of lexical-syntactic information from the mental lexicon However, in LpMTG there was more activation for sentences than for words, and the ambiguity effect in LpMTG was larger in sentences than in word sequences (see Fig. 4). Moreover, the ambiguity effect in the word sequences was only marginally significant. One possible interpretation of this pattern of results is that LpMTG plays a role in unification that is similar to that of LIFG. However, we favor an alternative interpretation: Sentence processing requires sustained activation of lexical frame information. During sentence comprehension, the lexical information has to be available for longer time intervals than during the processing of random word sequences. The lexical-syntactic information is most likely not “copied” from the area necessary for its retrieval (mental lexicon) to the area necessary for unification (Unification Space). Instead, the sustained activation of lexical frame information could be triggered by feedback from the Unification Space to the mental lexicon (as implemented—for independent theoretical modeling reasons—in the recently revised Unification Space model, see Vosse and Kempen 2008). The amount and/or duration of lexical frame activation is a function of the unification load imposed by the combinatorial operations necessary for unification. This explains why the lexical activation of the noun and verb frames has to be maintained longer in sentences as the unification load increases due to a word-class ambiguity.

How could the above-mentioned sustained activation and feedback be implemented neurally? LIFG has the neural machinery to provide feedback signals to other areas in the brain (Miller and Cohen 2001). Research has shown that the sustained activation of representations in posterior cortices is under dynamic frontal top-down control (Tomita et al. 1999; Fuster 2001; Curtis and D'Esposito 2003; Miller and D'Esposito 2005; Fiebach et al. 2006). For the present experiment, our interpretation is that the syntactic unification process requires the dynamic interplay between posterior LIFG and LpMTG. A similar reasoning can be found in Keller et al. (2001), who manipulated lexical and syntactic factors (word frequency and syntactic complexity) and found a similar interaction pattern for both temporal and frontal regions. This interpretation is also in line with a recent fMRI study suggesting context-dependent interactions or feedback mechanism between LIFG and LpMTG (Gennari et al. 2007).

Our findings with regard to ambiguity resolution are consistent with earlier neuroimaging studies. The activation in LIFG is roughly comparable, but slightly more posterior, to the region that was identified in the studies by Rodd and Zempleni (Rodd et al. 2005; Zempleni et al. 2007). Striking is the fact that in these studies the coordinates for the activation peak in LpMTG for sentence ambiguous compared to sentence unambiguous conditions are very similar (this study: [−52 −50 −8], Rodd et al: [−52 −50 −10], Zempleni et al. [−50 −48 −12]; Rodd et al. 2005; Zempleni et al. 2007). Although Rodd also included noun–verb ambiguities in her study, the focus of both these other studies was on noun-noun ambiguities. Thus, LpMTG might be involved in the retrieval of both semantic and syntactic properties of words. The more posterior locus of our LIFG activation fits well with the idea that there might be some specialization within subregions of LIFG for unification content, with the posterior LIFG contributing to syntactic unification, whereas semantic unification is subserved by a more anterior portion of LIFG (Hagoort 2005b).

Other Perspectives

An alternative interpretation of the interaction effects in LIFG and LpMTG might be a general attentional difference between the processing of sentences and word sequences, resulting in larger ambiguity effects for sentences. Sentences evoked more activation in perisylvian areas than word sequences. However, it is not the case that the ambiguity effect is larger in the sentences for all brain areas involved in ambiguity processing (see e.g., the activity in the striatum in Fig. 4). Moreover, the fact that our participants were highly accurate in the control task (for both sentences and word sequences) argues against an attention interpretation.

A second alternative interpretation relates to semantic ambiguity. Word-category ambiguous words are intrinsically also ambiguous semantically. Thus, besides syntactic retrieval and unification differences, our ambiguity manipulation will also lead to semantic retrieval and unification differences. With our current experiment we cannot entirely rule out the possibility that our fMRI results reflect semantic rather than syntactic processes. In order to address this issue we conducted a post-test to investigate the semantic consequences of the ambiguity. Each meaning of the ambiguous word can potentially activate a distinct set of semantic associates. The post-test showed that for the word-category ambiguous condition more associates were “context-irrelevant” in that they had to be inhibited as a consequence of the ambiguity resolution. The left inferior prefrontal cortex is sensitive to difficulty in the selection of semantic information (e.g., Thompson-Schill et al. 2005). Thus, selection/inhibition of this semantic information during sentence processing might potentially drive the observed LIFG ambiguity effects, and possibly the observed LpMTG ambiguity effects as well. However, when we included the context-irrelevance score for each sentence item as a covariate in the fMRI analysis, this did not alter the results, suggesting that the difference in the amount of context-irrelevant information evoked by ambiguous and unambiguous items cannot explain our fMRI results. More importantly, when looking at the effect of context-irrelevance itself, there was absolutely no activation in LIFG or LpMTG, not even at very low threshold levels. In other words, the post-test strongly suggests that our fMRI results are not due to the diffuse semantic consequences (more context-irrelevant associates for ambiguous words) of the word-category ambiguity. Thus, the sentence ambiguity effect in LIFG is most likely due to the increased unification demands (competition between alternatives and/or selection of the correct syntactic structure, see note 1) resulting from our syntactic manipulation (word-category ambiguity).

Additionally Involved Brain Regions

A striking effect that we did not expect beforehand is that, in addition to LpMTG, also the right striatum showed a large effect of ambiguity. The striatum is known to be involved in selection of behaviorally relevant stimuli and inhibition of competing alternatives (e.g., Cools et al. 2004, 2006). The left striatum has been found to be modulated by syntactic variables (e.g., Moro et al. 2001; Friederici and Kotz 2003). Copland and colleagues have demonstrated the importance of the striatum in ambiguity processing, using semantic priming paradigms in patients with damage to the basal ganglia. Typically striatal dysfunction spares automatic activation of multiple meanings of ambiguous words, but disrupts later language processes that require inhibition of competing alternatives (Copland et al. 2000, 2001; Copland 2003, 2006). Our study again stresses the importance of the striatum in ambiguity processing. The ambiguity effect we find in the striatum in both sentences and word sequences might reflect the selection of one meaning/lemma of the ambiguous word and/or the inhibition of the competing alternative.

Interestingly, when processing ambiguities in sentences, not only LIFG and LpMTG are activated, but also their right-hemisphere homologue areas (see Table 5 and Fig. 3C). The involvement of the right hemisphere in language processing is more and more acknowledged, especially for the processing of complex, natural language (Faust and Chiarello 1998; Kircher et al. 2001; Jung-Beeman 2005). Our study again emphasizes the involvement of right perisylvian areas in language comprehension in ambiguous contexts. However, our results do not clarify whether the right-hemisphere areas really contribute functionally to unification processes in language, or whether the activations we find simply reflect interhemispheric connections between homologous areas (see also Rodd et al. 2005). The precise role of the striatum, RIFG, and RpMTG in language processing remains an issue for future research.

Conclusion

In conclusion, using the computational model of syntactic parsing by Vosse and Kempen (2000) as a starting point, this study disentangled syntactic unification and lexical-syntactic retrieval processes. Posterior LIFG plays a role in the unification of words into a sentence structure, whereas LpMTG is involved in the retrieval of lexical-syntactic information from memory. Although their specific contributions to the process may differ, syntactic unification in language requires the dynamic interplay between both brain regions.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

Netherlands Organization for Scientific Research (grant 051.04.030) to the Cognitive Psychology Unit of Leiden University and the F.C. Donders Centre for Cognitive Neuroimaging.

We are indebted to Paul Gaalman for technical assistance during scanning, and the NCL PhD group for their comments on the manuscript. We also thank 2 anonymous reviewers for their thoughtful comments. Conflict of Interest: None declared.

References

Baayen
RH
Piepenbrock
R
Hv
Rijn
The CELEX lexical database [CD-ROM]
 , 
1993
Philadelphia
Linguistic Data Consortium, University of Pennsylvania
Bavelier
D
Corina
D
Jezzard
P
Padmanabhan
S
Clark
VP
Karni
A
Prinster
A
Braun
A
Lalwani
A
Rauschecker
JP
, et al.  . 
Sentence reading: a functional MRI study at 4 tesla
J Cogn Neurosci.
 , 
1997
, vol. 
9
 (pg. 
664
-
686
)
Beek
Lvd
Bouma
G
Malouf
R
Noord
Gv
Theune
M
Nijholt
A
Hondorp
H
The Alpino dependency treebank
Computational linguistics in the Netherlands 2001
 , 
2002
Amsterdam: Rodopi
(pg. 
8
-
22
)
Bookheimer
S
Functional MRI of language: new approaches to understanding the cortical organization of semantic processing
Annu Rev Neurosci.
 , 
2002
, vol. 
25
 (pg. 
151
-
188
)
Caramazza
A
Hillis
AE
Lexical organization of nouns and verbs in the brain
Nature.
 , 
1991
, vol. 
349
 (pg. 
788
-
790
)
Cools
R
Clark
L
Robbins
TW
Differential responses in human striatum and prefrontal cortex to changes in object and rule relevance
J Neurosci.
 , 
2004
, vol. 
24
 (pg. 
1129
-
1135
)
Cools
R
Ivry
RB
D'Esposito
M
The human striatum is necessary for responding to changes in stimulus relevance
J Cogn Neurosci.
 , 
2006
, vol. 
18
 (pg. 
1973
-
1983
)
Copland
DA
The basal ganglia and semantic engagement: Potential insights from semantic priming in individuals with subcortical vascular lesions, Parkinson's disease, and cortical lesions
J Int Neuropsychol Soc.
 , 
2003
, vol. 
9
 (pg. 
1041
-
1052
)
Copland
DA
Meaning selection and the subcortex: evidence of reduced lexical ambiguity repetition effects following subcortical lesions
J Psycholinguist Res.
 , 
2006
, vol. 
35
 (pg. 
51
-
66
)
Copland
DA
Chenery
HJ
Murdoch
BE
Understanding ambiguous words in biased sentences: evidence of transient contextual effects in individuals with nonthalamic subcortical lesions and Parkinson's disease
Cortex.
 , 
2000
, vol. 
36
 (pg. 
601
-
622
)
Copland
DA
Chenery
HJ
Murdoch
BE
Discourse priming of homophones in individuals with dominant nonthalamic subcortical lesions, cortical lesions and Parkinson's disease
J Clin Exp Neuropsychol.
 , 
2001
, vol. 
23
 (pg. 
538
-
556
)
Culicover
PW
Jackendoff
R
Simpler syntax
 , 
2005
Oxford
Oxford University Press
Curtis
CE
D'Esposito
M
Persistent activity in the prefrontal cortex during working memory
Trends Cogn Sci.
 , 
2003
, vol. 
7
 (pg. 
415
-
423
)
Damasio
AR
Tranel
D
Nouns and verbs are retrieved with differently distributed neural systems
Proc Natl Acad Sci USA.
 , 
1993
, vol. 
90
 (pg. 
4957
-
4960
)
Damasio
H
Grabowski
TJ
Tranel
D
Hichwa
RD
Damasio
AR
A neural basis for lexical retrieval
Nature.
 , 
1996
, vol. 
381
 (pg. 
810
-
810
)
Decety
J
Grezes
J
Costes
N
Perani
D
Jeannerod
M
Procyk
E
Grassi
F
Fazio
F
Brain activity during observation of actions—influence of action content and subject's strategy
Brain.
 , 
1997
, vol. 
120
 (pg. 
1763
-
1777
)
Duffy
SA
Morris
RK
Rayner
K
Lexical ambiguity and fixation times in reading
J Mem Lang.
 , 
1988
, vol. 
27
 (pg. 
429
-
446
)
Durstewitz
D
Seamans
JK
Sejnowski
TJ
Neurocomputational models of working memory
Nat Neurosci.
 , 
2000
, vol. 
3
 
Suppl.
(pg. 
1184
-
1191
)
Evans
AC
Collins
DL
Mills
SR
Brown
ED
Kelly
RL
Peters
TM
3D statistical neuroanatomical models from 305 MRI volumes
Proc IEEE Nuclear Science Symposium and Medical Imaging Conference. IEEE
 , 
1993
(pg. 
1813
-
1817
)
Faust
M
Chiarello
C
Sentence context and lexical ambiguity resolution by the two hemispheres
Neuropsychologia.
 , 
1998
, vol. 
36
 (pg. 
827
-
835
)
Federmeier
KD
Segal
JB
Lombrozo
T
Kutas
M
Brain responses to nouns, verbs and class-ambiguous words in context
Brain.
 , 
2000
, vol. 
123
 (pg. 
2552
-
2566
)
Fiebach
CJ
Rissman
J
D'Esposito
M
Modulation of inferotemporal cortex activation during verbal working memory maintenance
Neuron.
 , 
2006
, vol. 
51
 (pg. 
251
-
261
)
Forman
SD
Cohen
JD
Fitzgerald
M
Eddy
WF
Mintun
MA
Noll
DC
Improved assessment of significant activation in functional magnetic-resonance-imaging (fMRI)—use of a cluster-size threshold
Magn Reson Med.
 , 
1995
, vol. 
33
 (pg. 
636
-
647
)
Friederici
AD
Towards a neural basis of auditory sentence processing
Trends Cogn Sci.
 , 
2002
, vol. 
6
 (pg. 
78
-
84
)
Friederici
AD
Kotz
SA
The brain basis of syntactic processes: functional imaging and lesion studies
Neuroimage.
 , 
2003
, vol. 
20
 (pg. 
S8
-
S17
)
Friston
KJ
Holmes
A
Poline
JB
Price
CJ
Frith
CD
Detecting activations in PET and fMRI: levels of inference and power
Neuroimage.
 , 
1996
, vol. 
4
 (pg. 
223
-
235
)
Friston
KJ
Holmes
AP
Worsley
KJ
Poline
JB
Frith
C
Frackowiak
RSJ
Statistical parametric maps in functional imaging: a general linear approach
Hum Brain Mapp.
 , 
1995
, vol. 
2
 (pg. 
189
-
210
)
Fuster
JM
The prefrontal cortex—an update: time is of the essence
Neuron.
 , 
2001
, vol. 
30
 (pg. 
319
-
333
)
Fuster
JM
Frontal lobe and cognitive development
J Neurocytol.
 , 
2002
, vol. 
31
 (pg. 
373
-
385
)
Fuster
JM
Bodner
M
Kroger
JK
Cross-modal and cross-temporal association in neurons of frontal cortex
Nature.
 , 
2000
, vol. 
405
 (pg. 
347
-
351
)
Gennari
SP
Macdonald
MC
Postle
BR
Seidenberg
MS
Context-dependent interpretation of words: evidence for interactive neural processes
Neuroimage.
 , 
2007
, vol. 
35
 (pg. 
1278
-
1286
)
Hagoort
P
How the brain solves the binding problem for language: a neurocomputational model of syntactic processing
Neuroimage.
 , 
2003
, vol. 
20
 (pg. 
S18
-
S29
)
Hagoort
P
Cutler
A
Broca's complex as the unification space for language
Twenty-first century psycholinguistics: four cornerstones
 , 
2005
Mahwah (NJ)
Lawrence Erlbaum Press
Hagoort
P
On Broca, brain, and binding: a new framework
Trends Cogn Sci.
 , 
2005
, vol. 
9
 (pg. 
416
-
423
)
Hagoort
P
Brown
CM
Osterhout
L
Brown
CM
Hagoort
P
The neural architecture of syntactic processing
The neurocognition of language
 , 
1999
Oxford
Oxford University Press
(pg. 
273
-
316
)
Hamzei
F
Rijntjes
M
Dettmers
C
Glauche
V
Weiller
C
Buchel
C
The human action recognition system and its relationship to Broca's area: an fMRI study
Neuroimage.
 , 
2003
, vol. 
19
 (pg. 
637
-
644
)
Hickok
G
Poeppel
D
Towards a functional neuroanatomy of speech perception
Trends Cogn Sci.
 , 
2000
, vol. 
4
 (pg. 
131
-
138
)
Hickok
G
Poeppel
D
Dorsal and ventral streams: a framework for understanding aspects of the functional anatomy of language
Cognition.
 , 
2004
, vol. 
92
 (pg. 
67
-
99
)
Indefrey
P
Müller
HM
Rickheit
G
Hirnaktivierungen bei syntaktischer Sprachverarbeitung: eine Meta-Analyse
Neurokognition der Sprache, Band 1
 , 
2004
Tübingen
Stauffenburg Verlag
Indefrey
P
Cutler
A
Gazzaniga
MS
Prelexical and lexical processing in listening
The cognitive neurosciences
 , 
2005
3rd ed
Cambridge (MA)
MIT Press
(pg. 
759
-
774
)
Jackendoff
R
Foundations of language: brain, meaning, grammar, evolution
 , 
2002
Oxford
Oxford University Press
Joshi
AK
Schabes
Y
Rozenberg
G
Salomaa
A
Tree-adjoining grammars
Handbook of formal languages (Vol.3)
 , 
1997
Berlin
Springer-Verlag
(pg. 
69
-
124
)
Jung-Beeman
M
Bilateral brain processes for comprehending natural language
Trends Cogn Sci.
 , 
2005
, vol. 
9
 (pg. 
512
-
518
)
Kaan
E
Swaab
TY
The brain circuitry of syntactic comprehension
Trends Cogn Sci.
 , 
2002
, vol. 
6
 (pg. 
350
-
356
)
Keller
TA
Carpenter
PA
Just
MA
The neural bases of sentence comprehension: a fMRI examination of syntactic and lexical processing
Cereb Cortex.
 , 
2001
, vol. 
11
 (pg. 
223
-
237
)
Kempen
G
Harbusch
K
Theune
M
Nijholt
A
Hondorp
H
Performance grammar: a declarative definition
Computational linguistics in the Netherlands 2001
 , 
2002
Amsterdam
Rodopi
(pg. 
148
-
162
)
Kircher
TTJ
Brammer
M
Andreu
NT
Williams
SCR
McGuire
PK
Engagement of right temporal cortex during processing of linguistic context
Neuropsychologia.
 , 
2001
, vol. 
39
 (pg. 
798
-
809
)
Lancaster
JL
Woldorff
MG
Parsons
LM
Liotti
M
Freitas
ES
Rainey
L
Kochunov
PV
Nickerson
D
Mikiten
SA
Fox
PT
Automated Talairach Atlas labels for functional brain mapping
Hum Brain Mapp.
 , 
2000
, vol. 
10
 (pg. 
120
-
131
)
Levelt
WJM
Accessing words in speech production—stages, processes and representations
Cognition.
 , 
1992
, vol. 
42
 (pg. 
1
-
22
)
Longe
O
Randall
B
Stamatakis
EA
Tyler
LK
Grammatical categories in the brain: the role of morphological structure
Cereb Cortex.
 , 
2007
, vol. 
17
 (pg. 
1812
-
1820
)
Macdonald
MC
Pearlmutter
NJ
Seidenberg
MS
Lexical nature of syntactic ambiguity resolution
Psychol Rev.
 , 
1994
, vol. 
101
 (pg. 
676
-
703
)
Mai
JK
Assheuer
J
Paxinos
G
Atlas of the human brain
 , 
2004
London
Elsevier Academic Press
Marcus
GF
Vouloumanos
A
Sag
IA
Does Broca's play by the rules?
Nat Neurosci.
 , 
2003
, vol. 
6
 (pg. 
651
-
652
)
Marslen-Wilson
WD
Functional parallelism in spoken word-recognition
Cognition.
 , 
1987
, vol. 
25
 (pg. 
71
-
102
)
Mesulam
MM
From sensation to cognition
Brain.
 , 
1998
, vol. 
121
 (pg. 
1013
-
1052
)
Miller
BT
D'Esposito
M
Searching for “the top” in top-down control
Neuron.
 , 
2005
, vol. 
48
 (pg. 
535
-
538
)
Miller
EK
Cohen
JD
An integrative theory of prefrontal cortex function
Annu Rev Neurosci.
 , 
2001
, vol. 
24
 (pg. 
167
-
202
)
Moro
A
Tettamanti
M
Perani
D
Donati
C
Cappa
SF
Fazio
F
Syntax and the brain: disentangling grammar by selective anomalies
Neuroimage.
 , 
2001
, vol. 
13
 (pg. 
110
-
118
)
Ojemann
GA
Cortical organization of language
J Neurosci.
 , 
1991
, vol. 
11
 (pg. 
2281
-
2287
)
Patel
AD
Language, music, syntax and the brain
Nat Neurosci.
 , 
2003
, vol. 
6
 (pg. 
674
-
681
)
Petersson
KM
Forkstam
C
Ingvar
M
Artificial syntactic violations activate Broca's region
Cogn Sci.
 , 
2004
, vol. 
28
 (pg. 
383
-
407
)
Plas
Lvd
Bouma
G
Wouden
Tvd
Poss
M
Reckman
H
Cremers
C
Syntactic contexts for finding semantically similar words
Computational linguistics in the Netherlands 2004
 , 
2005
Utrecht: LOT
(pg. 
173
-
186
)
Pulvermuller
F
Mohr
B
Schleichert
H
Semantic or lexico-syntactic factors: what determines word-class specific activity in the human brain?
Neurosci Lett.
 , 
1999
, vol. 
275
 (pg. 
81
-
84
)
Raichle
ME
MacLeod
AM
Snyder
AZ
Powers
WJ
Gusnard
DA
Shulman
GL
A default mode of brain function
Proc Natl Acad Sci USA.
 , 
2001
, vol. 
98
 (pg. 
676
-
682
)
Rodd
JM
Davis
MH
Johnsrude
IS
The neural mechanisms of speech comprehension: fMRI studies of semantic ambiguity
Cereb Cortex.
 , 
2005
, vol. 
15
 (pg. 
1261
-
1269
)
Saffran
E
Sholl
A
Brown
CM
Hagoort
P
Clues to the functional and neural architecture of word meaning
The neurocognition of language
 , 
1999
Oxford
Oxford University Press
(pg. 
241
-
273
)
Sag
IA
Wasow
T
Syntactic theory: a formal introduction
1999
 
Stanford (CA): CSLI Publications
Seidenberg
MS
Tanenhaus
MK
Leiman
JM
Bienkowski
M
Automatic access of the meanings of ambiguous words in context—some limitations of knowledge-based processing
Cogn Psychol.
 , 
1982
, vol. 
14
 (pg. 
489
-
537
)
Shapiro
KA
Moo
LR
Caramazza
A
Cortical signatures of noun and verb production
Proc Natl Acad Sci USA.
 , 
2006
, vol. 
103
 (pg. 
1644
-
1649
)
Stowe
LA
Broere
CAJ
Paans
AMJ
Wijers
AA
Mulder
G
Vaalburg
W
Zwarts
F
Localizing components of a complex task: sentence processing and working memory
Neuroreport.
 , 
1998
, vol. 
9
 (pg. 
2995
-
2999
)
Stowe
LA
Wijers
AA
Willemsen
A
Reuland
E
Paans
AMJ
Vaalburg
W
PET studies of language—an assessment of the reliability of the technique
J Psycholinguist Res.
 , 
1994
, vol. 
23
 (pg. 
499
-
527
)
Swaab
T
Brown
C
Hagoort
P
Understanding words in sentence contexts: the time course of ambiguity resolution
Brain Lang.
 , 
2003
, vol. 
86
 (pg. 
326
-
343
)
Swinney
DA
Lexical access during sentence comprehension—(re)consideration of context effects
J Verbal Learn Verbal Behav.
 , 
1979
, vol. 
18
 (pg. 
645
-
659
)
Thompson-Schill
SL
Cutler
A
Dissecting the language organ: a new look at the role of Broca's area in language processing
Twenty-first Century psycholinguistics: four cornerstones
 , 
2005
Mahwah (NJ)
Lawrence Erlbaum Press
Thompson-Schill
SL
Bedny
M
Goldberg
RF
The frontal lobes and the regulation of mental activity
Curr Opin Neurobiol.
 , 
2005
, vol. 
15
 (pg. 
219
-
224
)
Thompson-Schill
SL
D'Esposito
M
Kan
IP
Effects of repetition and competition on activity in left prefrontal cortex during word generation
Neuron.
 , 
1999
, vol. 
23
 (pg. 
513
-
522
)
Tomita
H
Ohbayashi
M
Nakahara
K
Hasegawa
I
Miyashita
Y
Top-down signal from prefrontal cortex in executive control of memory retrieval
Nature.
 , 
1999
, vol. 
401
 (pg. 
699
-
703
)
Van Petten
C
Nadel
L
Lexical ambiguity resolution
Encyclopedia of cognitive science
 , 
2002
London
Macmillan
(pg. 
867
-
872
)
Vigliocco
G
Vinson
DP
Arciuli
J
Barber
H
The role of grammatical class on word recognition
Brain Lang.
 , 
2008
, vol. 
105
 (pg. 
175
-
184
)
Vigneau
M
Beaucousin
V
Herve
PY
Duffau
H
Crivello
F
Houde
O
Mazoyer
B
Tzourio-Mazoyer
N
Meta-analyzing left hemisphere language areas: phonology, semantics, and sentence processing
Neuroimage.
 , 
2006
, vol. 
30
 (pg. 
1414
-
1432
)
von Humboldt
W
Über die Verschiedenheit des menslichen Sprachbaues und ihren Einfluss auf die geistige Entwicklung des Menschengeschlechts (on the structural variety of human language and its influence on the intellectual development of mankind)
 , 
1836
Berlin
F. Dummler
Vosse
T
Kempen
G
Syntactic structure assembly in human parsing: a computational model based on competitive inhibition and a lexicalist grammar
Cognition.
 , 
2000
, vol. 
75
 (pg. 
105
-
143
)
Vosse
T
Kempen
G
Love
BC
McRae
K
Sloutsky
VM
Parsing verb-final clauses in German: Garden-path and ERP effects modeled by a parallel dynamic parser
Proceedings of the 30th Annual Conference on Social Cognitive Science
 , 
2008
Austin (TX)
Cognitive Science Society
(pg. 
261
-
266
)
Willems
R
Ozyurek
A
Hagoort
P
When language meets action: the neural integration of gesture and speech
Cereb Cortex.
 , 
2007
, vol. 
17
 (pg. 
2322
-
2333
)
Worsley
KJ
Marret
S
Neelin
P
Vandal
AC
Friston
KJ
Evans
AC
A unified statistical approach for determining significant signals in images of cerebral activation
Hum Brain Mapp.
 , 
1996
, vol. 
4
 (pg. 
58
-
73
)
Zempleni
MZ
Renken
R
Hoeks
JCJ
Hoogduin
JM
Stowe
LA
Semantic ambiguity processing in sentence context: evidence from event-related fMRI
Neuroimage.
 , 
2007
, vol. 
34
 (pg. 
1270
-
1279
)