Abstract

Sentences are easier to memorize than ungrammatical word strings, a phenomenon known as the sentence superiority effect. Yet, it is unclear how higher-order linguistic information facilitates verbal working memory and how this is implemented in the neural system. The goal of the current fMRI study was to specify the brain mechanisms underlying the sentence superiority effect during encoding and during maintenance in working memory by manipulating syntactic structure and working memory load. The encoding of sentence material, as compared with the encoding of ungrammatical word strings, recruited not only inferior frontal (BA 47) and anterior temporal language-related areas but also the medial-temporal lobe, which is not classically reported for language tasks. During maintenance, it was sentence structure as contrasted with ungrammatical word strings that led to activation decrease in Broca's area, SMA, and parietal regions. Furthermore, in Broca's area, an interaction effect revealed a load effect for ungrammatical word strings but not for sentences. The sentence superiority effect, thus, is neurally reflected in a twofold pattern, consisting of increased activation in classical language as well as memory areas during the encoding phase and decreased maintenance-related activation. This pattern reflects how chunking, based on sentential syntactic and semantic information, alleviates rehearsal demands and thus leads to improved working memory performance.

INTRODUCTION

Language and working memory are deeply intertwined. Language comprehension requires linguistic processing as well as working memory resources as soon as it gets beyond the level of single words (Rogalsky & Hickok, 2011; Grodzinsky & Santi, 2008; Fiebach, Schlesewsky, Lohmann, von Cramon, & Friederici, 2005; Caplan & Waters, 1999). In turn, structured language is known to aid working memory, which supports the assumption of a tight interaction between these two fundamental cognitive systems. An important empirical demonstration of the close relationship between language and working memory is the sentence superiority effect (SSE), that is, the observation that sentences are remembered better than ungrammatical word strings (Perham, Marsh, & Jones, 2009; Jefferies, Lambon Ralph, & Baddeley, 2004; Brener, 1940). Interestingly, the SSE is not restricted to full sentences but also holds true for small fragments, for example, word pairs. Perham et al. (2009) used adjective–noun pairs either in the correct or reversed grammatical order, observing a superior recall for items with a grammatically correct word order. The present article tackles the question how the SSE unfolds over the first consecutive stages of the memorizing process (i.e., encoding and maintenance) and how it is neurally implemented.

In addressing the question why memorizing sentences in verbal working memory is facilitated, Baddeley, Hitch, and Allen (2009) argue that the—phonologically coded—linguistic material interacts with knowledge about the sequential redundancy in language to bind chunks of multiple items. The importance of sequential redundancy was already highlighted decades earlier when Miller and Selfridge (1950) conducted a study where they gradually increased the dependent probability of words in word strings (i.e., the probability of certain words to occur after each other in natural language, which is constrained more and more over the course of a sentence). Here, it was shown that the more similar the word string was to natural language (i.e., high dependent probability), the more successful participants were in recall.

In contrast to this proposal, other authors have argued that recalling sentences is fundamentally different from recalling word lists, based on their observation that STM for word lists is carried mainly by phonological information, whereas STM for sentences relies on semantic information (Potter & Lombardi, 1990). Following up this observation, Potter and Lombardi put forward the conceptual regeneration hypothesis (CRH), stating that sentences are recalled based on their meaning, using previously utilized and thus primed words and syntactic structures (Potter & Lombardi, 1990, 1998). Note that sentence-level semantics emerges from the combination of single words and phrases via syntactic structure (the principle of compositionality, e.g., Hagoort, 2009; Pylkkanen, 2008), leading to a semantically enriched overall representation. A comprehensive account of the SSE, however, should take into account that different levels of linguistic representation (phonology, syntax, and semantics) can contribute to short-term or working memory for sentences, because recent studies showed that both phonological (Schweppe, Rummer, Bormann, & Martin, 2011) and syntactic (Schweppe & Rummer, 2007) information are retained during immediate sentence recall. In turn, this result might be taken to suggest that syntactic structure conveys the sequential redundancy in language. Specifically, in sentences, the order of words belonging to specific grammatical categories is highly predictable (e.g., an adjective after a determiner is usually followed by a noun), leading to the sequential redundancy that Baddeley et al. (2009) regarded as being helpful for successful memorizing.

A more general account of the SSE may arise from process models of working memory such as the active memory model by Zhou, Ardestani, & Fuster (2007) or the embedded processing model of working memory by Cowan (1999). On the basis of these models, one would predict that verbal working memory (VWM) is a state of sustained activation in relevant representations of the language system. This implies that VWM does not rely on a specific phonological component alone (i.e., the phonological loop, a subvocal articulatory rehearsal mechanism; Baddeley & Hitch, 1974). Instead, according to these models, all available representational properties of linguistic materials such as phonology, semantics, or syntax can be kept in an activated state, which serves as the basis of the maintenance of information in VWM. In support of this, it was shown, for example, that both lexical (Fiebach, Rissman, & D'Esposito, 2006) and word-level semantic information (Fiebach, Friederici, Smith, & Swinney, 2007) improve working memory and modulate WM-related maintenance activity by recruiting language-related areas.

With respect to the underlying neural resources, it appears that the temporary maintenance or manipulation of current relevant information relies on a fronto-parietal working memory network (e.g., Rottschy et al., 2012; Curtis & D'Esposito, 2003; Salmon et al., 1996), which, however, is modulated depending on working memory contents. In a recent meta-analysis, Rottschy et al. (2012) report that both verbal VWM and non-VWM rely on the same core network including bilateral (dorso-) lateral pFC, inferior frontal gyrus (IFG; BA 44), anterior insula, (pre-)SMA, and inferior parietal sulcus (extending into the left inferior parietal cortex). Moreover, the authors show that working memory processing of verbal material, when compared with nonverbal material, leads to increased activation within the left IFG (i.e., Broca's area; Rottschy et al., 2012). Although not all working memory studies are able to differentiate between different phases of the working memory process, there is a wide agreement in the literature that the process of memorizing can be subdivided into three stages: encoding, maintenance, and retrieval of information (e.g., Chein & Fiez, 2001). However, the neural mechanisms supporting better WM performance for sentences than for unstructured linguistic material during the different stages of the working memory process are not yet understood. Specifying these bears important implications for understanding the interaction of language and working memory in general.

Most neuroimaging studies so far have investigated the reverse question, namely, how WM supports language processing, most specifically sentence processing. During natural sentence processing, the different component processes of working memory, that is, encoding and maintenance, cannot be independently specified, because it is inherent to natural communication that maintenance of linguistic material happens simultaneously with ongoing encoding of new linguistic input. Studies investigating sentence comprehension frequently report activation of parts of Broca's area for both enhanced working memory costs (Rogalsky & Hickok, 2011; Kaan & Swaab, 2002) as well as genuine syntactic processes (Makuuchi, Bahlmann, Anwander, & Friederici, 2009; Grodzinsky & Santi, 2008), with some of the authors arguing for anatomical (but not functional) separation of the two processes (Makuuchi et al., 2009). Aside from the activation of Broca's area, these sentence processing studies frequently report anterior/superior temporal and left IFG activation. Although these studies approach the question of how WM might assist language processing, they do however leave unanswered as to which brain areas support encoding and maintenance of sentences in WM.

This study is the first to investigate syntactic contributions to VWM in an event-related fMRI paradigm that disentangles encoding processes from maintenance processes. We studied the short-term maintenance of sentence fragments versus ungrammatical strings containing identical words to isolate the effect of sentence structure on working memory performance and brain activation patterns. We additionally manipulated working memory load (WML; items of four vs. six words in length) as well as rehearsal capacities independently. In half of the trials, phonological rehearsal was blocked using articulatory suppression (AS; i.e., participants' ongoing articulation of nonwords during the maintenance phase; Murray, Rowan, & Smith, 1988; Hanley & Thomas, 1984), thereby testing whether the SSE is affected by the availability of the phonological rehearsal system (i.e., the phonological loop as described by Baddeley multicomponent model; cf. Baddeley et al., 2009; Baddeley & Hitch, 1974).

Previous studies in the nonlinguistic domain evidenced increased encoding-related activations for structured material (e.g., auditorily presented number sequences “8 6 4 2 3 5 7 9” vs. a random number sequence) in lateral pFC (Bor, Cumming, Scott, & Owen, 2004; Bor, Duncan, Wiseman, & Owen, 2003). The latter study reported subsequently reduced maintenance-related activity in parietal and premotor cortices, leading the authors to argue that structured material allows more efficient encoding, also described as “chunking,” which in turn might reduce demands in the maintenance phase (Bor et al., 2003). Given these findings, we assumed that language processing for working memory purposes leads to similar effects. As syntactic structure has a consequence for both the structural as well as the semantic representation of a sentence, we will use (in analogy to previous working memory studies including various types of memoranda), the neutral term “chunking” to describe the building of both semantic and syntactic relations during sentence encoding. Linguistically based chunking can also be described as an enriched encoding process because it entails, in addition to the simple sequence of items, semantic and syntactic relations between items. Thus, in line with Baddeley et al. (2009) who stressed the importance of binding processes during encoding of sentences and Potter and Lombardi (1998) who postulate that superior WM for sentences is strongly influenced by the generation of sentence meaning during encoding, we hypothesize that the WM benefit of sentence structure is to a large part because of enriched encoding. This enriched encoding in turn is hypothesized to result in reduced WM demands during the subsequent maintenance phase, which is compatible with the fact that Potter and Lombardi (1998)—unlike for example, process models of WM (e.g., Cowan, 1999)—assume that no specific mechanisms are additionally activated during WM for sentences.

With respect to functional neuroanatomy, we predict that enriched encoding should go along with increased activity in the fronto-temporal language network for semantic and syntactic sentence processing. Moreover, in addition to the typical language- and WM-related areas (e.g., inferior/middle frontal gyrus, inferior parietal lobule), a plausible candidate for supporting the SSE is the hippocampus. Recently, the hippocampal formation has been discussed to support binding of multiple items in working memory, also in the domain of VWM (Axmacher et al., 2010; Baddeley, Allen, & Vargha-Khadem, 2010; Sederberg et al., 2007) and, more generally, sequential pattern prediction (Buckner, 2010). It is a critical open question whether the assumed linguistic chunking during the encoding of sentence material in WM is supported exclusively by peri-sylvian language-related brain regions like Broca's area or by more domain-general memory systems like the hippocampus or whether both systems are involved in the SSE.

During the subsequent maintenance period, we expected to see reduced activity for sentence material in VWM systems responsible for phonological rehearsal because of the encoding-induced load reduction. Furthermore, the present design allows one to test the independence of the SSE from rehearsal processes during the maintenance phase. If the SSE were independent of the availability of phonological rehearsal, AS should leave working memory processing of sentence material unaffected, although at the same time, it should impair WM processing of ungrammatical strings, because the latter critically depends on the availability of the phonological loop.

METHODS

Participants

Eighteen participants (nine women, mean age = 25.0 years, range = 20–31 years) participated after giving written informed consent in accordance with the Declaration of Helsinki. Participants were compensated with €8 per hour. All participants were right-handed according to their scores on the Edinburgh handedness inventory (adapted German version of Oldfield, 1971), and all were native German speakers without a known history of dyslexia, other psychiatric disorders, or neurological diseases.

Experimental Design

As illustrated in Figure 1A, the present event-related fMRI working memory paradigm included the within-subject factors sentence structure (SST+ vs. SST−) and working memory load (hiWML vs. loWML), plus a third within-subject factor, articulatory suppression (AS) during maintenance (AS+ vs. AS−, separate sessions), resulting in eight conditions.

Figure 1. 

(A) Example items and design. To illustrate the construction principle for original sentence fragments, the example sentence fragment is completed to a sentence (in this case, a question) and supplemented by a translation. The colored words form the sentence fragment used in the stimulus set; the exact colors of the words indicate equivalent parts in English and German. Below this example stimulus, the design is illustrated. WML was manipulated by presenting memoranda consisting of either four words (low) or six words (high). Sentence structure was either correct (SST+) or absent (SST−), and syntactically correct sentence fragments were half-sentences with auxiliary verbs only (to assure low word-level semantics). Word-by-word English translation is given below the original items in italics. Note that the example SST+ is correct in German, although not in the word-by-word translation. Importantly, ungrammatical word strings contained the same words as sentence fragments, but in a different, ungrammatical order. All items were presented twice, once in a session with AS and once in a session without AS. (B) Modeling of BOLD responses during fMRI analysis, for encoding (light gray), maintenance (black), and retrieval (dark gray). To increase statistical independence between trial phases, a jitter before and after the maintenance phase was introduced. Jitter lengths were counterbalanced across conditions. Pauses between trials (jittered, mean = 4.75 sec) were not explicitly modeled and therefore served as implicit baseline.

Figure 1. 

(A) Example items and design. To illustrate the construction principle for original sentence fragments, the example sentence fragment is completed to a sentence (in this case, a question) and supplemented by a translation. The colored words form the sentence fragment used in the stimulus set; the exact colors of the words indicate equivalent parts in English and German. Below this example stimulus, the design is illustrated. WML was manipulated by presenting memoranda consisting of either four words (low) or six words (high). Sentence structure was either correct (SST+) or absent (SST−), and syntactically correct sentence fragments were half-sentences with auxiliary verbs only (to assure low word-level semantics). Word-by-word English translation is given below the original items in italics. Note that the example SST+ is correct in German, although not in the word-by-word translation. Importantly, ungrammatical word strings contained the same words as sentence fragments, but in a different, ungrammatical order. All items were presented twice, once in a session with AS and once in a session without AS. (B) Modeling of BOLD responses during fMRI analysis, for encoding (light gray), maintenance (black), and retrieval (dark gray). To increase statistical independence between trial phases, a jitter before and after the maintenance phase was introduced. Jitter lengths were counterbalanced across conditions. Pauses between trials (jittered, mean = 4.75 sec) were not explicitly modeled and therefore served as implicit baseline.

Stimuli

Our stimulus set comprised a basic set of 30 sentence fragments. As depicted in Figure 1A, each set was presented as a short (four word) and a long (six word) version, and all word strings were presented with grammatical (SST+) and ungrammatical (SST−) word order, resulting in 120 items. These 120 items were presented in one session with (AS+) and one session without AS (AS−), giving a total of 240 trials. To rule out influences of word familiarity, function words were repeated across items, again limiting the overall set of words used in the present stimulus set. To keep the semantic content and associations between words as low as possible, we used sentence fragments ending within a phrase and containing mainly function words (pronouns, auxiliary verbs, and prepositions) except for temporal nouns and adverbs such as “yesterday/in the morning.” Ungrammatical word strings were generated from the sentence fragments such that each word string represented words of a specific sentence fragment in an ungrammatical word order. Thus, lexical properties, word length, and word frequency were matched between SST+ and SST−. The use of fragmentary as well as relatively content-free open-class words was intended to reduce semantic memory strategies in the ungrammatical (SST−) condition as well as possible automatic word reordering tendencies, which might be more likely if content words with their respective semantic associations were available. The material was tested in a pilot study (n = 19), resulting in matched stimulus sets with high acceptance of grammatical stimuli (SST+: mean = 99.47%, SD = 1.92%) and low acceptability ratings for ungrammatical stimuli (SST−: mean = 3.28%, SD = 9.64%). Each of the eight conditions contained 30 items, adding up to 240 items.

Experimental Procedures

During the training session and the fMRI experiment, each trial (cp. Figure 1B) consisted of four (i.e., loWML) or six (hiWML) visually presented words appearing one by one in the center of the computer screen with a rate of 0.5 sec per word (encoding). To keep visual input equal between low and hiWML conditions, four word items were preceded by strings of “+” and “−” symbols, as depicted in Figure 1A. Therefore, the total duration of the encoding period totaled 3 sec in all conditions. Participants were asked to keep the word strings in mind (maintenance period; jittered duration of 6.5, 8, 9.5, 11, 12.5, or 14 sec, jitter counterbalanced across experimental conditions). To ensure that participants would not try to “reorder” ungrammatical word strings into correct sentence structure during this delay, the subsequent task required them to remember not only word identity but also the serial order of appearance (retrieval period, duration of 6 sec). Specifically, the task was to decide whether a certain word A appeared before a certain word B (target item positions were randomized and counterbalanced across conditions). The reordering of words in ungrammatical word strings was therefore disadvantageous; none of the participants reported such a strategy. Responses were recorded using two response boxes for the left and right thumb with buttons “yes” and “no,” respectively. Intertrial intervals were jittered (3.75–4.75 sec), resulting in an average trial duration of 23.5 sec (range = 19.25–27.75 sec).

(a) Because the stimulus set was too extensive to be captured by one session and (b) to avoid task-switching-related brain activation, each participant was invited for two sessions, one session with and one without AS, respectively. In the AS session, participants were instructed to lie in the scanner with the lips slightly open. During the maintenance phase of each stimulus, participants articulated a simple and meaningless German consonant–vowel syllable sequence “nenadana nenadana …” starting with a nasal consonant and ensuring minimal jaw movement during the maintenance phase of the task. To match the rate of articulation between participants and trials, each articulation of “nenadana” was initiated by a recurring fixation cross (frequency = 1 sec) that appeared on the screen during the maintenance period. For comparability reasons, this recurring fixation cross was also shown during the maintenance in the sessions without AS. This way, in one session, it was possible to subvocally rehearse the items; the other session included AS. Order of sessions (with/without AS) was counterbalanced across participant gender and response hand assignment (yes/no button). All participants completed preexperimental questionnaires regarding their sleep patterns and their coffee/nicotine consumption and underwent a training procedure (30 example trials) outside the scanner before each fMRI session.

Data Acquisition

Behavioral data were recorded using Presentation software (Version 14.1, www.neurobs.com). fMRI data were assessed using a Siemens TIM Trio 3-T scanner with continuous scanning (1,500 scans per session). A gradient-echo EPI sequence was used with an echo time of 30 msec, flip angle of 90°, repetition time (TR) of 2000 msec, and an acquisition bandwidth of 116 kHz. The matrix acquired was 64 × 64 with a field of view of 19.2 cm, resulting in an in-plane resolution of 3 × 3 mm. Thirty slices were acquired with a slice thickness of 3 mm and an interslice gap of 1 mm, covering the whole brain.

Before functional data, a T1-weighted 3-D magnetization prepared rapid gradient echo sequence (as described in Mugler & Brookeman, 1990) was collected for coregistration (inversion time = 650 msec, TR of the total sequence cycle = 1300 msec, TR of the gradient-echo kernel (snapshot FLASH: Haase, 1990) = 10 msec, echo time = 3.93 msec, alpha = 10°, bandwidth = 130 Hz/pixel (i.e., 67 kHz in total), image matrix = 256 × 240, field of view = 256 mm × 240 mm, slab thickness = 192 mm, 128 partitions, 95% slice resolution, sagittal orientation, spatial resolution = 1 mm × 1 mm × 1.5 mm, two acquisitions).

Data Analysis

Behavioral data were analyzed using PASW Statistics 19 (SPSS, Inc., Chicago, www.spss.com), performing a repeated-measures ANOVA with three factors, that is, SST, WML, and AS, for error rates and response times (RTs) separately. All fMRI data analysis was done using SPM8 software provided by the Wellcome Department of Cognitive Neurology, London, United Kingdom. Preprocessing steps included (1) realignment to correct for head motion during scanning and coregistration of EPI images of both sessions separately to the participant's T1-weighted structural image, (2) slice timing correction with reslicing, and (3) spatial normalization into the standard stereotactic space (provided by Montreal Neurological Institute [MNI], implemented in SPM8) for group analysis. Additionally, to increase the signal-to-noise ratio, data were smoothed with an 8-mm FWHM isotropic Gaussian kernel. For first-level analysis, we modeled encoding for each condition (3 sec), an initial jitter between encoding and maintenance (0.75–4.5 sec) independent of condition, maintenance per condition (middle 5 sec of each trial, cp. Figure 1B), a jitter between maintenance and retrieval (0.75–4.5 sec) again independent of condition, retrieval per condition (full 6 sec), and button presses independent of condition, separately for both sessions. The implicit baseline included jittered pauses between trials (3.75–4.75 sec). To partial out the processes underlying error trials as well as instruction phases and half-time break, each phase was assigned to a separate regressor. Additionally, to account for head movement during scanning, realignment parameters were included in the model as well. Afterwards, to identify BOLD signal changes for each condition, second-level analysis was performed inserting first-level baseline contrast images for each condition in a flexible factorial ANOVA (2 × 2 × 2 design including the three factors SST, WML, and AS). Thereby, we directly compared effects of different conditions (e.g., with > without sentence structure, hiWML > loWML, without > with AS) and their interactions within the same model. To correct for multiple comparisons, the software package AlphaSim (Ward, 2000) performing a Monte Carlo simulation with 1,000 iterations was used. We estimated a minimal cluster size of 50 voxels, which, combined with a single-voxel threshold of p < .001, results in a cluster-level significance level of p < .05. This threshold was applied to all contrasts. For significant interaction clusters, a percent signal change analysis was performed; peak voxel values of prominent clusters were extracted per condition using the SPM toolbox MarsBar (http://marsbar.sourceforge.net), the subsequent statistical analysis (ANOVA) was performed in PASW.

RESULTS

Behavioral Results

Accuracy

Overall, accuracy was well above chance level in all conditions (mean values are depicted in Figure 2C). Participants' accuracy was higher for sentence fragments (SST+) than for ungrammatical word strings (SST−) (F(1, 17) = 23.74, p < .001), for loWML compared with hiWML (F(1, 17) = 22.89, p < .001), and when rehearsal was possible (AS−; F(1, 17) = 6.98, p = .017) compared with AS+. Additionally, sentence structure interacted with WML (F(1, 17) = 6.51; p = .021; Figure 2A, bottom) and AS (F(1, 17) = 13.25, p = .002; Figure 2A, top). Specifically, when sentence structure was available, no effects of WML or AS were evident in accuracy (both ts > −1.24). It was only in ungrammatical word strings that accuracy decreased with both AS (t = −3.63; p = .002) and hiWML (t = −5.04; p = .000). All ANOVA effects are listed in Table 1.

Figure 2. 

Behavioral results. Facilitation effects of sentence structure were evident in both accuracies (A) and RTs (B). Accuracy: AS (A, top) and WML (A, bottom) dependent effects are visible for ungrammatical word strings (SST−) but not for sentence fragments (SST+). All main effects (AS, WM, and SST) also reached significance (not included in Figure 2). RTs: Independent of AS and WM, responses were faster for SST+ compared with responses for SST− (B, top). Differences between loWML and hiWML are significantly larger during the session without AS, compared with the session with AS (B, bottom). Both the main effects of AS and WM (not included in Figure 2) reached significance as well. (C) Mean values of RTs (blue) and accuracy (red) for all experimental conditions. Error bars represent the SEM (during maintenance phase; *p < .05, **p < .01, ***p < .001).

Figure 2. 

Behavioral results. Facilitation effects of sentence structure were evident in both accuracies (A) and RTs (B). Accuracy: AS (A, top) and WML (A, bottom) dependent effects are visible for ungrammatical word strings (SST−) but not for sentence fragments (SST+). All main effects (AS, WM, and SST) also reached significance (not included in Figure 2). RTs: Independent of AS and WM, responses were faster for SST+ compared with responses for SST− (B, top). Differences between loWML and hiWML are significantly larger during the session without AS, compared with the session with AS (B, bottom). Both the main effects of AS and WM (not included in Figure 2) reached significance as well. (C) Mean values of RTs (blue) and accuracy (red) for all experimental conditions. Error bars represent the SEM (during maintenance phase; *p < .05, **p < .01, ***p < .001).

Table 1. 

ANOVA Results for Behavioral Measures (Accuracy and RTs)

Accuracy
RTs
Effect
F
Sig.
Partial η2
Effect
F
Sig.
Partial η2
AS 6.983 .017 .291 AS 4.968 .040 .226 
SST 23.744 .000 .583 SST 36.099 .000 .680 
WML 22.892 .000 .574 WML 61.399 .000 .783 
AS × SST 13.247 .002 .438 AS × SST 2.843 .110 .143 
AS × WML .962 .340 .054 AS × WML 9.037 .008 .347 
SST × WML 6.511 .021 .277 SST × WML 0.148 .705 .009 
AS × SST × WML .000 1.000 .000 AS × SST × WML 3.412 .082 .167 
Accuracy
RTs
Effect
F
Sig.
Partial η2
Effect
F
Sig.
Partial η2
AS 6.983 .017 .291 AS 4.968 .040 .226 
SST 23.744 .000 .583 SST 36.099 .000 .680 
WML 22.892 .000 .574 WML 61.399 .000 .783 
AS × SST 13.247 .002 .438 AS × SST 2.843 .110 .143 
AS × WML .962 .340 .054 AS × WML 9.037 .008 .347 
SST × WML 6.511 .021 .277 SST × WML 0.148 .705 .009 
AS × SST × WML .000 1.000 .000 AS × SST × WML 3.412 .082 .167 

RTs

Mean RT values of all conditions are provided in Figure 2C. As shown in Table 1, participants responded faster to sentence fragments (SST+) than to ungrammatical word strings (SST−; F(1, 17) = 36.10, p < .001), for loWML compared with hiWML (F(1, 17) = 61.40, p < .001) and when rehearsal was possible (AS−; F(1, 17) = 4.97, p = .04) compared with AS+. In addition, WML interacted with AS (F(1, 17) = 9.04, p = .008; Figure 3A, top), indicating that the RT differences because of AS are only significant in the loWML condition (t = 2.61, p = .018), not in the hiWML condition (t = −1.747, p = .099). Figure 2B illustrates the main effect of SST as well as the interaction between AS and WML.

Figure 3. 

fMRI activation results. (A) Main effect of sentence structure during encoding (top row) shows significantly increased activations for structured, compared with unstructured memoranda in yellow-marked regions, and for unstructured, compared with structured memoranda in blue-marked regions. During the maintenance period, structured memoranda produce less activity (blue) than unstructured memoranda (bottom row). (B) Top: Significant activations for the main effect of WML (WML+ > WML−, cyan), the interaction of WML and sentence structure (red), and the overlap between effects (white). Diagrams below the activation maps display the results of ROI analyses (see Methods section) for the interaction clusters. Bottom: Significant activations for the main effect of WML (WML+ > WML−, cyan; identical to B) and the interaction of WML and AS (purple) as well as the results of ROI analyses for significant interaction clusters. All activations are rendered onto an inflated representation of the brain template provided by SPM8, with a threshold of p < .001 (corrected for cluster size, p < .001). Error bars represent SEM (during maintenance phase). LH = left hemisphere; RH = right hemisphere; dmPFC = dorsomedial prefrontal cortex; HIPP = hippocampus; IFG = inferior frontal gyrus; IPL = inferior parietal lobule; IPS = intraparietal sulcus; MFG = middle frontal gyrus; MTL = medial temporal lobe; SMA = Supplementary motor area.

Figure 3. 

fMRI activation results. (A) Main effect of sentence structure during encoding (top row) shows significantly increased activations for structured, compared with unstructured memoranda in yellow-marked regions, and for unstructured, compared with structured memoranda in blue-marked regions. During the maintenance period, structured memoranda produce less activity (blue) than unstructured memoranda (bottom row). (B) Top: Significant activations for the main effect of WML (WML+ > WML−, cyan), the interaction of WML and sentence structure (red), and the overlap between effects (white). Diagrams below the activation maps display the results of ROI analyses (see Methods section) for the interaction clusters. Bottom: Significant activations for the main effect of WML (WML+ > WML−, cyan; identical to B) and the interaction of WML and AS (purple) as well as the results of ROI analyses for significant interaction clusters. All activations are rendered onto an inflated representation of the brain template provided by SPM8, with a threshold of p < .001 (corrected for cluster size, p < .001). Error bars represent SEM (during maintenance phase). LH = left hemisphere; RH = right hemisphere; dmPFC = dorsomedial prefrontal cortex; HIPP = hippocampus; IFG = inferior frontal gyrus; IPL = inferior parietal lobule; IPS = intraparietal sulcus; MFG = middle frontal gyrus; MTL = medial temporal lobe; SMA = Supplementary motor area.

fMRI Results

Encoding of Information

WML

hiWML, as compared with loWML, led to increased activation in bilateral IFG (BA 44), SMA, insulae, precentral gyri, middle temporal gyri, and cerebella. Additionally, left IFG (BA 45), right middle frontal gyrus, left inferior parietal lobule, right supramarginal gyrus, and left middle occipital gyrus were activated stronger for hiWML during encoding (cf. Table 2). This is in line with previous WM studies (for a recent review, see Rottschy et al., 2012) documenting the data reliability of this study.

Table 2. 

Activation Clusters and Increased Activation Peaks for fMRI Contrasts during Encoding (WML, Sentence Structure) and Maintenance (Sentence Structure)

Brain Region
Hemisphere
BA
Cluster Size (Voxel)
Zmax
MNI Coordinates
x
y
z
Encoding: WML High > Low 
Middle frontal gyrus 46 160 4.74 39 41 25 
 189 4.6 18 10 
  4.09 12 −1 −2 
IFG (p. opercularis) 44 4,129 6.98 −51 14 16 
 SMA  6.83 −3 14 52 
 SMA  6.83 17 46 
 IFG (p. opercularis) 44  6.74 51 14 
 Insula   6.53 −30 23 
 Precentral gyrus  6.29 −39 −1 61 
 Insula   6.28 33 23 
 Precentral gyrus  6.26 −42 37 
 Putamen   6.14 −18 10 
 Superior frontal gyrus   5.64 27 52 
 IFG 47  5.59 −51 17 −5 
 Superior frontal gyrus  4.92 −24 67 
 IFG (p. triangularis) 45  4.36 −51 35 
 Middle frontal gyrus 46  3.92 39 43 
 IFG (p. opercularis) 44  3.85 39 25 
 Precentral gyrus  3.82 54 46 
 Middle temporal gyrus 21/38  3.59 −48 −20 
Middle temporal gyrus 21 506 6.45 −54 −46 10 
Middle temporal gyrus  54 4.22 48 −31 −2 
Supramarginal gyrus (IPC) 40 150 4.52 51 −31 46 
Inferior parietal lobule (hlP3) 7/39 390 5.22 −30 −58 43 
 Middle occipital gyrus 19  5.06 −27 −70 37 
 Inferior parietal lobule (hlP2)   4.06 −48 −43 46 
Cerebellum  61 4.48 −30 −64 −32 
Cerebellum  326 5.65 30 −61 −29 
 
Encoding: Sentence Fragments > Ungrammatical Word Strings 
Medial superior frontal gyrus 2,486 6.85 −9 50 34 
 Mid orbital gyrus 10  6.04 −6 59 −11 
 Middle frontal gyrus  5.94 30 26 49 
 Superior medial gyrus  5.14 53 37 
 Rectal gyrus 11  4.99 −3 44 −20 
 ACC 32  4.66 35 −8 
Angular gyrus, IPC (PFm) 39/40 6,038 7.15 −48 −58 25 
 Posterior cingulate cortex 23  6.58 −6 −52 25 
 Middle temporal gyrus 21  6.47 −54 −13 −17 
 IFG (p. orbitalis) 47  6.31 −39 32 −14 
 Fusiform gyrus 37  5.62 −27 −37 −17 
 Hippocampus (SUB)   5.25 −21 −25 −20 
 Inferior temporal gyrus 20  5.04 54 −16 −17 
 Angular gyrus 39  4.98 57 −64 34 
 Hippocampus (CA)   4.84 27 −16 −17 
 Middle temporal gyrus 21  4.69 54 −1 −23 
 Supramarginal gyrus 40  4.66 66 −46 31 
 Insula lobe (Ig2)   4.59 36 −19 10 
 Hippocampus, amygdala (LB)   4.53 −21 −7 −20 
 Rolandic operculum 4/43  4.48 54 −22 22 
 Fusiform gyrus 37  4.43 24 −40 −14 
 Superior temporal gyrus 38?  4.36 −39 −16 −2 
Middle occipital gyrus 18/19 330 4.13 33 −82 13 
 Fusiform gyrus 37  3.88 24 −79 −11 
 Cuneus 18  3.22 15 −97 13 
 Middle temporal gyrus 22/19  3.11 48 −76 10 
 Precentral gyrus  3.69 45 −13 49 
Cerebellum  66 4.99 27 −79 −32 
 
Encoding: Ungrammatical Word Strings > Sentence Fragments 
Middle frontal gyrus 46 58 4.05 42 41 22 
Superior frontal gyrus 4/6 139 4.74 21 52 
Inferior parietal lobule 40 185 4.62 −39 −40 37 
 Middle occipital gyrus 19  3.43 −27 −70 31 
 
Encoding: SST × AS 
SMA 107 3.46 −3 61 
 Superior medial gyrus  3.41 23 43 
 SMA  3.19 11 52 
Inferior parietal lobule (hlP1) 40 141 3.71 39 −49 43 
 Supramarginal gyrus 40  3.66 42 −40 40 
 IPL (PFt) 40  3.57 51 −34 52 
Superior parietal lobule 149 3.67 −27 −64 49 
 Inferior parietal lobule 40  3.64 −45 −40 40 
 
Maintenance: Ungrammatical Word Strings > Sentence Fragments 
Middle orbital gyrus 10/46 134 3.62 −39 44 −2 
IFG (p. triangularis) 45 274 3.82 −42 29 22 
 IFG (p. opercularis) 44  3.82 −48 11 28 
Medial superior frontal gyrus (SMA) 6/8 214 4.86 −3 26 43 
Insula lobe  55 4.20 39 23 −5 
Superior frontal gyrus 702 4.79 21 14 49 
 Middle frontal gyrus 46  4.74 42 38 28 
Middle frontal gyrus 4/6 64 4.43 −42 55 
 Precentral gyrus  3.15 −45 11 40 
Inferior parietal lobule 40 382 4.59 48 −52 52 
 Superior parietal lobule  4.15 33 −70 52 
Inferior parietal lobule 40 241 4.74 −48 −55 49 
 Superior parietal lobule  3.54 −33 −67 55 
Brain Region
Hemisphere
BA
Cluster Size (Voxel)
Zmax
MNI Coordinates
x
y
z
Encoding: WML High > Low 
Middle frontal gyrus 46 160 4.74 39 41 25 
 189 4.6 18 10 
  4.09 12 −1 −2 
IFG (p. opercularis) 44 4,129 6.98 −51 14 16 
 SMA  6.83 −3 14 52 
 SMA  6.83 17 46 
 IFG (p. opercularis) 44  6.74 51 14 
 Insula   6.53 −30 23 
 Precentral gyrus  6.29 −39 −1 61 
 Insula   6.28 33 23 
 Precentral gyrus  6.26 −42 37 
 Putamen   6.14 −18 10 
 Superior frontal gyrus   5.64 27 52 
 IFG 47  5.59 −51 17 −5 
 Superior frontal gyrus  4.92 −24 67 
 IFG (p. triangularis) 45  4.36 −51 35 
 Middle frontal gyrus 46  3.92 39 43 
 IFG (p. opercularis) 44  3.85 39 25 
 Precentral gyrus  3.82 54 46 
 Middle temporal gyrus 21/38  3.59 −48 −20 
Middle temporal gyrus 21 506 6.45 −54 −46 10 
Middle temporal gyrus  54 4.22 48 −31 −2 
Supramarginal gyrus (IPC) 40 150 4.52 51 −31 46 
Inferior parietal lobule (hlP3) 7/39 390 5.22 −30 −58 43 
 Middle occipital gyrus 19  5.06 −27 −70 37 
 Inferior parietal lobule (hlP2)   4.06 −48 −43 46 
Cerebellum  61 4.48 −30 −64 −32 
Cerebellum  326 5.65 30 −61 −29 
 
Encoding: Sentence Fragments > Ungrammatical Word Strings 
Medial superior frontal gyrus 2,486 6.85 −9 50 34 
 Mid orbital gyrus 10  6.04 −6 59 −11 
 Middle frontal gyrus  5.94 30 26 49 
 Superior medial gyrus  5.14 53 37 
 Rectal gyrus 11  4.99 −3 44 −20 
 ACC 32  4.66 35 −8 
Angular gyrus, IPC (PFm) 39/40 6,038 7.15 −48 −58 25 
 Posterior cingulate cortex 23  6.58 −6 −52 25 
 Middle temporal gyrus 21  6.47 −54 −13 −17 
 IFG (p. orbitalis) 47  6.31 −39 32 −14 
 Fusiform gyrus 37  5.62 −27 −37 −17 
 Hippocampus (SUB)   5.25 −21 −25 −20 
 Inferior temporal gyrus 20  5.04 54 −16 −17 
 Angular gyrus 39  4.98 57 −64 34 
 Hippocampus (CA)   4.84 27 −16 −17 
 Middle temporal gyrus 21  4.69 54 −1 −23 
 Supramarginal gyrus 40  4.66 66 −46 31 
 Insula lobe (Ig2)   4.59 36 −19 10 
 Hippocampus, amygdala (LB)   4.53 −21 −7 −20 
 Rolandic operculum 4/43  4.48 54 −22 22 
 Fusiform gyrus 37  4.43 24 −40 −14 
 Superior temporal gyrus 38?  4.36 −39 −16 −2 
Middle occipital gyrus 18/19 330 4.13 33 −82 13 
 Fusiform gyrus 37  3.88 24 −79 −11 
 Cuneus 18  3.22 15 −97 13 
 Middle temporal gyrus 22/19  3.11 48 −76 10 
 Precentral gyrus  3.69 45 −13 49 
Cerebellum  66 4.99 27 −79 −32 
 
Encoding: Ungrammatical Word Strings > Sentence Fragments 
Middle frontal gyrus 46 58 4.05 42 41 22 
Superior frontal gyrus 4/6 139 4.74 21 52 
Inferior parietal lobule 40 185 4.62 −39 −40 37 
 Middle occipital gyrus 19  3.43 −27 −70 31 
 
Encoding: SST × AS 
SMA 107 3.46 −3 61 
 Superior medial gyrus  3.41 23 43 
 SMA  3.19 11 52 
Inferior parietal lobule (hlP1) 40 141 3.71 39 −49 43 
 Supramarginal gyrus 40  3.66 42 −40 40 
 IPL (PFt) 40  3.57 51 −34 52 
Superior parietal lobule 149 3.67 −27 −64 49 
 Inferior parietal lobule 40  3.64 −45 −40 40 
 
Maintenance: Ungrammatical Word Strings > Sentence Fragments 
Middle orbital gyrus 10/46 134 3.62 −39 44 −2 
IFG (p. triangularis) 45 274 3.82 −42 29 22 
 IFG (p. opercularis) 44  3.82 −48 11 28 
Medial superior frontal gyrus (SMA) 6/8 214 4.86 −3 26 43 
Insula lobe  55 4.20 39 23 −5 
Superior frontal gyrus 702 4.79 21 14 49 
 Middle frontal gyrus 46  4.74 42 38 28 
Middle frontal gyrus 4/6 64 4.43 −42 55 
 Precentral gyrus  3.15 −45 11 40 
Inferior parietal lobule 40 382 4.59 48 −52 52 
 Superior parietal lobule  4.15 33 −70 52 
Inferior parietal lobule 40 241 4.74 −48 −55 49 
 Superior parietal lobule  3.54 −33 −67 55 

Activation peaks are defined as local maxima separated by at least 8 mm.

R = right; L = left.

SST

The encoding of sentence fragments (SST+) elicited broadly increased activations as compared with the encoding of ungrammatical word strings (SST−), in a network including prefrontal areas (left BA 47, extending into BA 45 and dorsomedial pFC), bilateral middle temporal gyri, bilateral angular gyri and fusiform gyri, precuneus as well as bilateral hippocampi and adjacent parahippocampal gyri, right middle occipital gyrus, and right cerebellum (cf. Figure 3A, top, yellow; peak activations are depicted in Table 2). Ungrammatical word strings (SST−) led to stronger activations in the left intraparietal sulcus (IPS), right SMA (BA 6), and right superior frontal sulcus during encoding (cf. Figure 3A, top, blue; peak activations in Table 2).

Because we were interested in the interplay between memory and sentence structure during the encoding period, interactions between SST and both WML and AS were calculated. The fMRI data did not reveal significant interaction effects between SST and WML. However, we found an interaction between SST and AS in bilateral SMA and intraparietal sulci, extending into superior as well as inferior parietal lobes (right supramarginal gyrus, cf. Table 2). Notably, participants did not yet start articulating during the encoding of words. The three-way interaction (SST × WML × AS) failed to show any significant results.

Maintenance of Information

WML

Higher WML led to increased activation in the right superior, middle, and inferior (BA 44) frontal gyrus as well as the left middle orbital gyrus, left inferior parietal lobule, and right superior parietal lobule extending into supramarginal gyrus during maintenance of items (cp. Table 3 and Figure 3B, activation increases of hiWML compared with loWML are depicted in cyan). These results are in line with previous literature on WM-related activations (for a recent meta-analysis, see Rottschy et al., 2012).

Table 3. 

Activation Clusters and Increased Activation Peaks for fMRI Contrasts During Maintenance (WML, WML × Sentence Structure Interaction, and WML × AS Interaction)

Brain Region
Hemisphere
BA
Cluster Size (Voxel)
Zmax
MNI Coordinates
x
y
z
Maintenance: WML High > Low 
Middle frontal gyrus 46 205 4.35 42 38 31 
 Middle orbital gyrus  10  3.49 36 59 −2 
Superior frontal gyrus 88 3.79 24 49 
 Middle frontal gyrus  46  3.27 36 14 37 
 IFG (p. opercularis)  44  3.22 30 34 
Inferior parietal lobule/IPS 40/7 367 3.92 39 −43 37 
 Superior parietal lobule   3.87 45 −49 58 
 Supramarginal gyrus  40  3.80 45 −37 43 
Inferior parietal lobule/IPS 40/7 130 3.98 −39 −49 40 
 Angular gyrus  39  3.37 −30 −58 34 
 
Maintenance: WML × Sentence Structure Interaction 
Middle orbital gyrus 11 60 3.80 24 50 −11 
Middle frontal gyrus 46 258 4.26 33 32 22 
 IFG (p. triangularis)  45  3.19 51 26 25 
Middle cingulate cortex 32 274 4.19 26 40 
 Left superior medial gyrus (SMA)  4.11 −9 23 43 
 Right superior medial gyrus (SMA)  3.91 35 40 
Left IFG (p. Opercularis) 44 297 4.46 −60 17 16 
 Left IFG (p. triangularis)  45  4.21 −45 26 25 
Inferior parietal lobule 40 75 3.45 −39 −49 40 
 
Maintenance: WML × AS Interaction 
Middle frontal gyrus 46 114 4.41 48 31 36 
IFG (p. triangularis) 45 93 3.82 −57 20 25 
 IFG (p. opercularis) 44  3.45 −51 14 
SMA 59 4.67 −3 61 
Brain Region
Hemisphere
BA
Cluster Size (Voxel)
Zmax
MNI Coordinates
x
y
z
Maintenance: WML High > Low 
Middle frontal gyrus 46 205 4.35 42 38 31 
 Middle orbital gyrus  10  3.49 36 59 −2 
Superior frontal gyrus 88 3.79 24 49 
 Middle frontal gyrus  46  3.27 36 14 37 
 IFG (p. opercularis)  44  3.22 30 34 
Inferior parietal lobule/IPS 40/7 367 3.92 39 −43 37 
 Superior parietal lobule   3.87 45 −49 58 
 Supramarginal gyrus  40  3.80 45 −37 43 
Inferior parietal lobule/IPS 40/7 130 3.98 −39 −49 40 
 Angular gyrus  39  3.37 −30 −58 34 
 
Maintenance: WML × Sentence Structure Interaction 
Middle orbital gyrus 11 60 3.80 24 50 −11 
Middle frontal gyrus 46 258 4.26 33 32 22 
 IFG (p. triangularis)  45  3.19 51 26 25 
Middle cingulate cortex 32 274 4.19 26 40 
 Left superior medial gyrus (SMA)  4.11 −9 23 43 
 Right superior medial gyrus (SMA)  3.91 35 40 
Left IFG (p. Opercularis) 44 297 4.46 −60 17 16 
 Left IFG (p. triangularis)  45  4.21 −45 26 25 
Inferior parietal lobule 40 75 3.45 −39 −49 40 
 
Maintenance: WML × AS Interaction 
Middle frontal gyrus 46 114 4.41 48 31 36 
IFG (p. triangularis) 45 93 3.82 −57 20 25 
 IFG (p. opercularis) 44  3.45 −51 14 
SMA 59 4.67 −3 61 
SST

In contrast to the encoding period, the presence of sentence structure (SST+) did not lead to additional activations during maintenance but to a relative reduction in brain activation in bilateral middle frontal (BA 46), left inferior frontal (BA 44/45), precentral (BA 6) as well as superior medial frontal and parietal regions (cp. Figure 3A, bottom, decreases in activation are depicted in blue color; all activation clusters and respective peaks are listed in Table 2).

Most important for our present research question, we observed a reliable interaction between sentence structure and WML in left IFG (BA 44, extending into BA 45), right middle frontal gyrus (BA 46, extending into BA 45; largely colocalized with the main effect of WML), ACC (BA 32, extending into bilateral superior medial gyrus, BA 6), and left inferior parietal lobule (BA 40). A percent signal change analysis revealed that this effect was driven by the behaviorally most difficult condition, that is, hiWML in SST—ungrammatical word strings. This condition specifically elicited increased activations compared with all other conditions (cf. Table 3 and Figure 3B, top, red). Notably, maintenance-related activity in these areas did not differ between loWML and hiWML for SST+ items.

AS

AS led to significantly enhanced activations of motor cortex, auditory cortex, and cerebellum. Additionally, we found an interaction between AS and WML, in a rehearsal network containing activations in bilateral PFC, that is, right middle frontal cortex (BA 46), left inferior frontal cortex (BA 44/45), and SMA (cf. Figure 3B, bottom, purple). A percent signal change analysis revealed that hiWML induced larger activations in these areas during maintenance only in the session without AS. With AS, no load effects were found in the specified regions. The three-way interaction (SST, WML, and AS) did not yield any significant results.

DISCUSSION

This study explored the brain processes underlying the SSE. In line with previous work (Baddeley et al., 2009; Jefferies et al., 2004; Brener, 1940), the availability of sentence structure (SST) indeed improved working memory performance and response speed. Our results further showed that the SST-related performance improvement abolishes the behavioral effects of increased WML and AS, suggesting that the facilitatory effect of SST on memory formation reduces WML to a degree that allows participants to cope with substantially increased extents of input. These behavioral results are in line with the findings of Baddeley et al. (2009), who demonstrated that the SSE survives AS. Our fMRI data revealed that a complex pattern of enhanced and reduced brain activations across the encoding and maintenance phases goes along with the behavioral SSE. In the following sections, we will discuss these brain activation effects for the two phases consecutively.

Sentence Structure Leads to Enriched Encoding

During encoding of sentence material into working memory (compared with ungrammatical word strings), sentence fragments led to less activity in the left inferior parietal sulcus. The left IPS has been discussed as a task-related attentional modulator in WM, which functionally connects to the right IPS in serial ordering tasks (Majerus et al., 2006). Following this line of interpretation, our data indicate that encoding a sentence fragment might require relatively fewer attentional resources than encoding and maintaining the exact sequence of an ungrammatical word string. Interestingly, in addition to this BOLD decrease, we found increased activity in a distributed system including prefrontal areas (left BA 47/45, dorsomedial pFC) as well as temporal and parietal regions and, bilaterally, the hippocampus and adjacent parahippocampal gyrus. Some of these regions have been associated earlier with semantic processing, whereas others have been linked to chunking during working memory encoding. The activation increases support the notion of enriched encoding processes that we hypothesized to support the SSE. Before discussing the nature of the engagement of these areas in detail, note that enhanced activation for the sentence material (the easier condition as indexed by behavioral performance data), although counterintuitive at first glance, is fully consistent with two earlier studies that investigated the processes involved in the chunking of working memory contents, that is, in combining individual items into larger bits of information to reduce WML (Bor et al., 2003, 2004). Bor and colleagues (2003, 2004) report, similar to our study, increased activation during encoding of structured material versus unstructured material. More specifically, across visually presented structured versus unstructured spatial sequences and auditorily presented structured versus unstructured number sequences, they showed encoding engagement of the right anterior pFC (BA 10), bilateral IFG (BAs 45 and 47, right BA 44), lateral temporal cortex (BAs 21 and 22, left BA 37), inferior parietal areas (BA 40) as well as left precuneus, right superior parietal gyrus, and the caudate nucleus. Some of these areas associated with chunking are also found in our study, such as inferior frontal areas, inferior parietal cortex, and precuneus. However, our data do not allow an unambiguous decision as to which areas in this study are activated for the SSE because of a domain-general contribution to chunking processes or because of language-specific processes. On the one hand, results from Bor et al. (2004) point toward modality-independent chunking processes in the dorsolateral pFC. On the other hand, they also found specific (e.g., left inferior frontal, BA 47) activation for auditory–verbal material, suggesting that the engagement of frontal regions in our study may reflect language-specific processes during WM encoding.

Indeed, most of the brain regions showing SSE-related activation increase in our study are known from language studies, such as the left pars orbitalis (BA 47) for semantic processing (De Carli et al., 2007; Chou et al., 2006; Demb et al., 1995), dorsomedial pFC for semantic processing (Binder & Desai, 2011) and text comprehension (Yarkoni, Speer, & Zacks, 2008), middle temporal gyrus for semantic relatedness effects (e.g., bed, rest; McDermott, Petersen, Watson, & Ojemann, 2003), and sentence generation (Brown, Martinez, & Parsons, 2006), or the IPL, specifically the supramarginal gyrus for semantic processing and integration (Chou et al., 2006) and the angular gyrus for semantic memory retrieval (Binder, Desai, Graves, & Conant, 2009). This is supported by our observation that the overall activation pattern strongly resembles the networks reported for semantic processing (Binder & Desai, 2011) as well as the one proposed for imagination and sequential pattern prediction (Buckner, 2010).

In this context, two questions arise: First, how might a stronger semantic network engagement (elicited by a comparison between sentence fragments and ungrammatical word strings, which only differed with respect to correct grammatical structure) contribute to working memory? Second, why are classical syntax-related areas like BA 44/45 (e.g., Santi & Grodzinsky, 2010; Makuuchi et al., 2009; Friederici, Fiebach, Schlesewsky, Bornkessel, & von Cramon, 2006) not involved?

With regard to the latter question, the aforementioned syntax studies compared sentences with more versus less complex syntactic structures, rather than simple sentence structures to ungrammatical word strings as in this study. Activations in BA 44/45 have been specifically attributed to the processing of complex syntactic structures—which was not required in this study. However, a simpler syntactic process, the formation of syntactic constituents, has also been related to activation in Broca's area (Pallier, Devauchelle, & Dehaene, 2011). In principle, it is conceivable that the use of function words in our ungrammatical word strings, which was necessary to keep the to-be-remembered items constant, led to the formation of short (two-word) constituents, attenuating a possible syntactic effect in Broca's area. The fact that constituents that can be built are much larger in the sentence condition speaks against this explanation. Syntactic features of simple sentences compared with word lists (e.g., noun lists) or ungrammatical word strings are thought to be processed in anterior temporal regions (e.g., Humphries, Binder, Medler, & Liebenthal, 2006; Kaan & Swaab, 2002; Vandenberghe, Nobre, & Price, 2002; Friederici, Meyer, & von Cramon, 2000; Stowe et al., 1998). This is in line with the activation observed in this study during the encoding of well-structured sentence fragments. Vandenberghe et al. (2002) compared syntactically correct items with their respective scrambled (thus ungrammatical) counterparts and found both effects of syntactic and semantic violations as well as their interaction in the left anterior temporal gyrus, leading the authors to suggest that this region is involved in deriving sentence-level meaning. However, others see anterior temporal regions to support combinatorial processes both in the semantic and syntactic domain (for a review, see Friederici, 2011).

Returning to the question of how additional semantic network activation in the present data might benefit memorizing, we find it particularly noteworthy that this prominent semantic activity pattern is observed in sentence fragments containing words that are relatively content-free when considered in isolation but that gain semantic value in the grammatical combination with other words (i.e., at the phrasal or sentence level). Take, for example, the high-load sample stimulus in Figure 1: “will er ihn heute Abend von” or, literally, “wants he him today evening from.” This sentence fragment clearly establishes semantic relations, that is, the intention of one male agent to do something with or for another male patient at a specified point in time. Thus, although the details of a full semantic interpretation of the message are lacking, it is possible to build up a partial representation of meaning.

But how does the involvement of the semantic system contribute to a performance advantage in the present working memory task? One possible account is chunking of information. From cognitive psychology, we know that chunking requires the encoding of at least two hierarchical levels: item level and chunk level (Feigenson & Halberda, 2004). The grammatical information contained in the word list makes it possible to integrate the words (i.e., items) into a larger unit (i.e., chunk) that is specified by grammatical relationships and a basic meaning representation, as outlined above. This constitutes not only a syntactically but also a semantically enriched unit that contains agents and patients characterized, for example, by specific semantic roles. Additional encoding of sentence-level meaning of this kind, triggered by syntactic structure, might facilitate the following stages (i.e., maintenance and retrieval) of the working memory process. Whereas on the one hand, one might have expected greater effects of sentence structure in brain regions more directly related to syntactic processing, such as Broca's area, on the other hand, it is important to keep in mind that the linguistic chunking processes in our task design entail the integration of both syntactic and semantic sentential information. In addition to the semantic network, we find increased activation during encoding of sentence fragments in a brain structure that is not typically discussed in sentence comprehension—the hippocampus. Generally, the hippocampus is regarded as essential for long-term memory (LTM, Squire, 1992) and, more recently, also for working memory processes (Axmacher et al., 2010). Jefferies and colleagues (2004) proposed that immediate sentence recall might be better for sentences, because meaningful sentences receive additional support from LTM. Combining this insight with the observed activation of the semantic network, one possible explanation for our results may be that the information in a sentence might be enriched by semantic information stored in LTM. For example, sentence fragments such as “er hat sie gestern Abend von”/“he has her yesterday evening from” could be associated with LTM contents such as, for example, scripts of actions, emotions, or autobiographical memories such as “yesterday evening he took her to the train station,” which might be an event recovered from memory that is not necessarily verbal. Indeed, the hippocampus has been linked to relational binding (i.e., storing items and their associations; for a review, see Moses & Ryan, 2006), specifically to multi-item maintenance in the visual domain (Axmacher et al., 2010), but also to semantic, associative processing (Henke, Weber, Kneifel, Wieser, & Buck, 1999). This assumed contribution of hippocampus to semantic-level binding is supported by the work of MacKay, Stewart, and Burke (1998). The authors report that patient H. M., who lost most of his hippocampi bilaterally undergoing an operation that was supposed to heal his epilepsy, was unable to recognize ambiguities in sentences, which the authors interpreted as reflecting a specific deficit regarding semantic-level binding.

All of the aspects mentioned above are plausible candidate mechanisms for building a memory chunk. Thus, a general explanation for the hippocampal activation found in our study might be relational binding of multiple items during the process of chunking. Because the sentence fragments used in this study primarily relied on function words that carry little content and that were frequently repeated within the experiment, the demand on explicit binding of the current item may have been particularly strong.

Apart from LTM enrichment or relational binding, there is empirical support for the idea that the hippocampus is involved in syntax processing. In an intracranial EEG recording study, Meyer et al. (2005) found a syntax effect in the hippocampus by comparing conditions where syntactic expectations were or were not met (i.e., syntactic violations), which the authors took as evidence that the hippocampus supports syntactic integration processes. The hippocampal response to syntactic violations seems to be in line with data showing that the hippocampus is also involved in processing prediction error signals (Schiffer, Ahlheim, Wurm, & Schubotz, 2012; Kumaran & Maguire, 2006). Most specifically, it has been shown that most activity occurs in the hippocampus when predictions are violated within familiar sequences and not when the sequence is novel and unpredictable. Ungrammatical word strings can be regarded as unpredictable sequences, and it thus seems plausible that the hippocampus is only engaged when predictions are possible, that is, when processing stimuli with familiar syntactic structures.

However, because this study did not test these different explanations explicitly, we cannot evaluate them on the basis of our present results. Nonetheless, the results demonstrate that the hippocampus plays a role when language and working memory processes interact. A possible account for the SSE is thus that relational or predictive processing of linguistic memoranda—based on the rules of syntax and mediated by the hippocampus—facilitates short-term maintenance through combining the memoranda into a linguistically enriched chunk. Whether those processes fundamentally differ from sentence processing in the absence of a working memory task cannot be concluded from the present data. However, as the hippocampus is usually not reported for ordinary sentence comprehension, it can be speculated that some of the observed effects are specific to sentence encoding in a working memory task context.

Sentence Structure Leads to Facilitated Maintenance

We hypothesized that, because of enriched encoding of sentence fragments, as compared with ungrammatical word strings, less phonological rehearsal would be required for sentence fragments during maintenance. Indeed, our results reveal that, in contrast to encoding, maintenance of sentence fragments is accompanied by reduced brain activation compared with ungrammatical word strings in bilateral premotor (BA 6), SMA, right prefrontal (BA 46), and left inferior parietal cortex. Decreases in these areas have been considered to reflect facilitation of cognitive processes (e.g., the left inferior parietal lobule is linked to phonological storage, Awh et al., 1996; BA 46 to general maintenance of items during delay phases in WM tasks, Curtis & D'Esposito, 2003; and more generally, all of these regions are engaged in cognitive control, Kubler, Dixon, & Garavan, 2006). As already mentioned, although activity in the left IPS has been linked to the modulation of attention, it was proposed that the left and right intraparietal areas connect with each other when serial order is relevant during STM tasks (Majerus et al., 2006). Thus, based on the bilateral IPS activity during the maintenance of ungrammatical word lists (SST−), we would argue not only that attention demands seem to be higher but also that order memory (as indicated by right IPS activity, cf. Marshuetz, Smith, Jonides, DeGutis, & Chenevert, 2000) is especially important when sentence structure is missing. In contrast, less effort might be necessary to keep the order of words in mind when sentence structure is available.

In line with these findings, facilitation of cognitive demands indeed might be one of the main differences between processing sentence fragments versus ungrammatical word strings in this study, because (a) the sentence fragments used in the stimulus set are highly frequent in everyday language and (b) maintenance of sentence fragments is, as concluded above, supported by a semantic representation generated during encoding. As we used ungrammatical strings including mainly function words rather than classical word lists (e.g., lists consisting of nouns or adjectives exclusively), the possibility for building semantic associations is extremely limited in the SST− condition in our case. Moreover, the ungrammatical word strings we used are not only infrequent in everyday language, but they violate phrase structure rules. Both of these properties may have further contributed to the facilitation for sentence fragments. It seems unlikely that the direction of the facilitatory effect would have changed if classical word lists had been used, but further investigations could specify how the facilitatory effect would be modified by using ungrammatical word order versus unrelated word lists.

Additional to the decreased brain activation induced by sentence structure and to prototypical rehearsal-related activations (BA 46, SMA, BA 40; e.g., Zarahn, Rakitin, Abela, Flynn, & Stern, 2005) for high load versus low load (cf. Paulesu, Frith, & Frackowiak, 1993), we found a significant interaction of sentence structure and WML in the left Broca's area, bilateral BA 46, and SMA during maintenance. This interaction was driven by the ungrammatical word strings: A load effect in these areas was observed exclusively for ungrammatical strings, not for sentence fragments, which, interestingly, resembles the pattern of results for behavioral accuracy (Figure 2A). Thus, our findings suggest that, if sentence structure is available, sufficient resources are available to cope with increased demands on the working memory system—for example, when the load on working memory is increased. Thus, we conclude that the presence of syntactic structure in verbal memoranda reduces the demands on neural systems involved in working memory maintenance—presumably because of the chunking processes discussed in the previous section.

In addition, we aimed to discover whether the SSE is affected by the availability of the phonological rehearsal mechanism. Although we found an interaction between SST and AS in the behavioral data, that is, increased error rates in conditions with AS only for ungrammatical word strings but not for sentence fragments, we did not find a neuronal counterpart of this interaction effect during the maintenance phase in our fMRI data. Independent of the syntactic manipulation, the main effect of AS during maintenance revealed significant activations of motor cortex, auditory cortex, and cerebellum. These results are expected given the literature regarding the role of auditory and motor/premotor cortex in articulation (Wildgruber, Ackermann, Klose, Kardatzki, & Grodd, 1996; cf. Yetkin et al., 1995) and motor timing for the cued production of syllable strings supported by the cerebellum, which are said to be involved while “computing the temporal parameters of incoming sensory stimuli” (Penhune, Zatorre, & Evans, 1998). Thus, although the prevention of rehearsal via concurrent articulation was successful, the fMRI data provide no evidence that the facilitation caused by sentence structure is necessarily affected by the availability of rehearsal mechanisms during the maintenance of information.

Combining the results from both encoding and maintenance, an integration of our results into current working memory models is still lacking. In the following paragraph, we will discuss our results in the light of two major models explicitly referring to sentence structure phenomena in memory tasks.

Sentence Structure Unburdens Working Memory

Neither general psychological models of working memory (such as Baddeley's multicomponent model (Baddeley, 2012) or process-based models (e.g., Zhou, Ardestani, & Fuster, 2007; Cowan, 1999)) nor more specific models of sentence recall (Jefferies et al., 2004; Potter & Lombardi, 1990, 1998) directly address the neuronal mechanisms underlying encoding and maintenance of sentences. Nevertheless, our results provide support for some of the assumptions of the latter class of models. First, the CRH states that a “sentence is regenerated in immediate recall from a representation of its meaning, using recently activated words” (Potter & Lombardi, 1990, p. 633) and later was expanded to syntactic priming, which means that the encoding of the syntactic surface structure facilitates their reproduction in immediate recall (Potter & Lombardi, 1998). The authors argue that, instead of maintaining a “surface” representation such as a phonological string, in sentences, the preactivation of words and the generation of a semantic interpretation during encoding may be sufficient for facilitating later recall. Indeed, we found semantic network activation for sentences already during the encoding of information, which provides support for the proposed importance of semantic processes during encoding. Furthermore, the neuronal deactivation pattern for sentence fragments as compared with ungrammatical word strings during maintenance is at least consistent with the CRH, which assumes no specific maintenance mechanisms for sentences. However, our results go beyond this assumption and suggest that more elaborate encoding processes in fact unburden neural systems supporting maintenance. Finally, because the current study focused on the encoding and maintenance of information, we cannot provide direct evidence for the third part of the CRH, namely, the recall of meaning and rebuilding of sentences using previously activated words.

Whereas Potter and Lombardi focus mostly on the regeneration of sentences during recall, Baddeley and colleagues (Baddeley et al., 2009) attribute the advantage of sentential material to the automatic chunking of sentences during encoding. The authors state that sequential redundancy, that is, the familiarity with specific word combinations, makes it easier to combine them in the very same order again—which would thus also benefit working memory. But even if one acknowledges that “sequential redundancy” contributed to better working memory encoding of sentences in our experiment, still, the question remains unresolved how this facilitation effect is mirrored in the brain. In general, easier tasks evoke less brain activity. Following this line of argumentation, one would have expected less activation for sentences than for word lists. Instead, we found an interesting pattern of stronger activations during encoding and activation decreases during the maintenance of sentence fragments (compared with ungrammatical word strings) in our data. Thus, from our data, we conclude that the effect of sentence structure on working memory (possibly relying on sequential redundancy) is reflected in this temporal interplay of brain activity during encoding and maintenance. In short, the memory facilitation for SST+ is not accompanied by less overall brain activity but rather by enhanced activity during LTM-enriched encoding followed by activation decreases during the less effortful maintenance phase. Process models of working memory therefore do not explain the present findings well as no sustained sentence-specific activation was found during the maintenance phase.

Conclusion

This study investigates for the first time the neural mechanisms underlying the SSE. Sentence structure activates a network of prefrontal, middle temporal, hippocampal, and inferior parietal brain areas during encoding. Therefore, we propose that the brain mechanisms underlying the SSE most probably involve (a) chunking, because of easier, hippocampally supported, relational binding of items according to grammatical rules, and (b) the association of items with LTM contents during encoding. According to our data, this elaborated encoding process results in a semantically enriched memory representation, subsequently facilitating the maintenance of information in working memory: Sentence structure leads to less engagement of rehearsal-related areas, specifically the left IFG, SMA, and right middle frontal gyrus, in the maintenance phase of the working memory task. Complementing earlier behavioral evidence, this study delineates the interplay between enriched encoding and resulting reduced maintenance demands that underlies the SSE.

Acknowledgments

For contributions to the design and comments on this manuscript, we are grateful to Angela D. Friederici. C. J. F. is supported by an Emmy-Noether grant from the German Research Foundation (DFG FI 848/3) and by the Hessian Initiative for the development of scientific and economic excellence (LOEWE).

Reprint requests should be sent to Corinna E. Bonhage, Albrechtstr. 28, 49076 Osnabrück, Germany, or via e-mail: bonhage@cbs.mpg.de.

REFERENCES

REFERENCES
Awh
,
E.
,
Jonides
,
J.
,
Smith
,
E. E.
,
Schumacher
,
E. H.
,
Koeppe
,
R. A.
, &
Katz
,
S.
(
1996
).
Dissociation of storage and rehearsal in verbal working memory: Evidence from positron emission tomography.
Psychological Science
,
7
,
25
31
.
Axmacher
,
N.
,
Henseler
,
M. M.
,
Jensen
,
O.
,
Weinreich
,
I.
,
Elger
,
C. E.
, &
Fell
,
J.
(
2010
).
Cross-frequency coupling supports multi-item working memory in the human hippocampus.
Proceedings of the National Academy of Sciences, U.S.A.
,
107
,
3228
3233
.
Baddeley
,
A.
(
2012
).
Working memory: Theories, models, and controversies.
Annual Review of Psychology
,
63
,
1
29
.
Baddeley
,
A.
,
Allen
,
R.
, &
Vargha-Khadem
,
F.
(
2010
).
Is the hippocampus necessary for visual and verbal binding in working memory?
Neuropsychologia
,
48
,
1089
1095
.
Baddeley
,
A.
, &
Hitch
,
G.
(
1974
).
Working memory.
In G. H. Bower (Ed.)
,
The psychology of learning and motivation
(
Vol. 8
, pp.
47
89
).
New York, NY
:
Academic Press
.
Baddeley
,
A.
,
Hitch
,
G. J.
, &
Allen
,
R. J.
(
2009
).
Working memory and binding in sentence recall.
Journal of Memory and Language
,
61
,
438
456
.
Binder
,
J. R.
, &
Desai
,
R. H.
(
2011
).
The neurobiology of semantic memory.
Trends in Cognitive Sciences
,
15
,
527
536
.
Binder
,
J. R.
,
Desai
,
R. H.
,
Graves
,
W. W.
, &
Conant
,
L. L.
(
2009
).
Where is the semantic system? A critical review and meta-analysis of 120 functional neuroimaging studies.
Cerebral Cortex
,
19
,
2767
2796
.
Bor
,
D.
,
Cumming
,
N.
,
Scott
,
C. E.
, &
Owen
,
A. M.
(
2004
).
Prefrontal cortical involvement in verbal encoding strategies.
European Journal of Neuroscience
,
19
,
3365
3370
.
Bor
,
D.
,
Duncan
,
J.
,
Wiseman
,
R. J.
, &
Owen
,
A. M.
(
2003
).
Encoding strategies dissociate prefrontal activity from working memory demand.
Neuron
,
37
,
361
367
.
Brener
,
R.
(
1940
).
An experimental investigation of memory span.
Journal of Experimental Psychology
,
26
,
467
482
.
Brown
,
S.
,
Martinez
,
M. J.
, &
Parsons
,
L. M.
(
2006
).
Music and language side by side in the brain: A PET study of the generation of melodies and sentences.
European Journal of Neuroscience
,
23
,
2791
2803
.
Buckner
,
R. L.
(
2010
).
The role of the hippocampus in prediction and imagination.
Annual Review of Psychology
,
61
,
27
48
.
Caplan
,
D.
, &
Waters
,
G.
(
1999
).
Verbal working memory and sentence comprehension.
Behavioral and Brain Sciences
,
22
,
77
94
;
discussion 95–126
.
Chein
,
J. M.
, &
Fiez
,
J. A.
(
2001
).
Dissociation of verbal working memory system components using a delayed serial recall task.
Cerebral Cortex
,
11
,
1003
1014
.
Chou
,
T. L.
,
Booth
,
J. R.
,
Bitan
,
T.
,
Burman
,
D. D.
,
Bigio
,
J. D.
,
Cone
,
N. E.
,
et al
(
2006
).
Developmental and skill effects on the neural correlates of semantic processing to visually presented words.
Human Brain Mapping
,
27
,
915
924
.
Cowan
,
N.
(
1999
).
An embedded-processes model of working memory.
In A. Miyake & P. Shah (Eds.)
,
Models of working memory: Mechanisms of active maintenance and executive control
(pp.
62
101
).
Cambridge
:
Cambridge University Press
.
Curtis
,
C. E.
, &
D'Esposito
,
M.
(
2003
).
Persistent activity in the prefrontal cortex during working memory.
Trends in Cognitive Sciences
,
7
,
415
423
.
De Carli
,
D.
,
Garreffa
,
G.
,
Colonnese
,
C.
,
Giulietti
,
G.
,
Labruna
,
L.
,
Briselli
,
E.
,
et al
(
2007
).
Identification of activated regions during a language task.
Magnetic Resonance Imaging
,
25
,
933
938
.
Demb
,
J. B.
,
Desmond
,
J. E.
,
Wagner
,
A. D.
,
Vaidya
,
C. J.
,
Glover
,
G. H.
, &
Gabrieli
,
J. D. E.
(
1995
).
Semantic encoding and retrieval in the left inferior prefrontal cortex—A functional MRI study of task-difficulty and process specificity.
Journal of Neuroscience
,
15
,
5870
5878
.
Feigenson
,
L.
, &
Halberda
,
J.
(
2004
).
Infants chunk object arrays into sets of individuals.
Cognition
,
91
,
173
190
.
Fiebach
,
C. J.
,
Friederici
,
A. D.
,
Smith
,
E. E.
, &
Swinney
,
D.
(
2007
).
Lateral inferotemporal cortex maintains conceptual–semantic representations in verbal working memory.
Journal of Cognitive Neuroscience
,
19
,
2035
2049
.
Fiebach
,
C. J.
,
Rissman
,
J.
, &
D'Esposito
,
M.
(
2006
).
Modulation of inferotemporal cortex activation during verbal working memory maintenance.
Neuron
,
51
,
251
261
.
Fiebach
,
C. J.
,
Schlesewsky
,
M.
,
Lohmann
,
G.
,
von Cramon
,
D. Y.
, &
Friederici
,
A. D.
(
2005
).
Revisiting the role of Broca's area in sentence processing: Syntactic integration versus syntactic working memory.
Human Brain Mapping
,
24
,
79
91
.
Friederici
,
A. D.
(
2011
).
The brain basis of language processing: From structure to function.
Physiological Reviews
,
91
,
1357
1392
.
Friederici
,
A. D.
,
Fiebach
,
C. J.
,
Schlesewsky
,
M.
,
Bornkessel
,
I. D.
, &
von Cramon
,
D. Y.
(
2006
).
Processing linguistic complexity and grammaticality in the left frontal cortex.
Cerebral Cortex
,
16
,
1709
1717
.
Friederici
,
A. D.
,
Meyer
,
M.
, &
von Cramon
,
D. Y.
(
2000
).
Auditory language comprehension: An event-related fMRI study on the processing of syntactic and lexical information.
Brain and Language
,
74
,
289
300
.
Grodzinsky
,
Y.
, &
Santi
,
A.
(
2008
).
The battle for Broca's region.
Trends in Cognitive Sciences
,
12
,
474
480
.
Haase
,
A.
(
1990
).
Snapshot FLASH MRI—Applications to T1, T2, and chemical-shift imaging.
Magnetic Resonance in Medicine
,
13
,
77
89
.
Hagoort
,
P.
,
Baggio
,
G.
, &
Willems
,
R. M.
(
2009
).
Semantic unification.
In M. Gazzaniga (Ed.)
,
The cognitive neurosciences.
Cambridge, MA
:
MIT Press
.
Hanley
,
J. R.
, &
Thomas
,
A.
(
1984
).
Maintenance rehearsal and the articulatory loop.
British Journal of Psychology
,
75
,
521
527
.
Henke
,
K.
,
Weber
,
B.
,
Kneifel
,
S.
,
Wieser
,
H. G.
, &
Buck
,
A.
(
1999
).
Human hippocampus associates information in memory.
Proceedings of the National Academy of Sciences, U.S.A.
,
96
,
5884
5889
.
Humphries
,
C.
,
Binder
,
J. R.
,
Medler
,
D. A.
, &
Liebenthal
,
E.
(
2006
).
Syntactic and semantic modulation of neural activity during auditory sentence comprehension.
Journal of Cognitive Neuroscience
,
18
,
665
679
.
Jefferies
,
E.
,
Lambon Ralph
,
M. A.
, &
Baddeley
,
A. D.
(
2004
).
Automatic and controlled processing in sentence recall: The role of long-term and working memory.
Journal of Memory and Language
,
51
,
623
643
.
Kaan
,
E.
, &
Swaab
,
T. Y.
(
2002
).
The brain circuitry of syntactic comprehension.
Trends in Cognitive Sciences
,
6
,
350
356
.
Kubler
,
A.
,
Dixon
,
V.
, &
Garavan
,
H.
(
2006
).
Automaticity and reestablishment of executive control—An fMRI study.
Journal of Cognitive Neuroscience
,
18
,
1331
1342
.
Kumaran
,
D.
, &
Maguire
,
E. A.
(
2006
).
An unexpected sequence of events: Mismatch detection in the human hippocampus.
PLoS Biology
,
4
,
2372
2382
.
MacKay
,
D. G.
,
Stewart
,
R.
, &
Burke
,
D. M.
(
1998
).
HM revisited: Relations between language comprehension, memory, and the hippocampal system.
Journal of Cognitive Neuroscience
,
10
,
377
394
.
Majerus
,
S.
,
Poncelet
,
M.
,
Van der Linden
,
M.
,
Albouy
,
G.
,
Salmon
,
E.
,
Sterpenich
,
V.
,
et al
(
2006
).
The left intraparietal sulcus and verbal short-term memory: Focus of attention or serial order?
Neuroimage
,
32
,
880
891
.
Makuuchi
,
M.
,
Bahlmann
,
J.
,
Anwander
,
A.
, &
Friederici
,
A. D.
(
2009
).
Segregating the core computational faculty of human language from working memory.
Proceedings of the National Academy of Sciences, U.S.A.
,
106
,
8362
8367
.
Marshuetz
,
C.
,
Smith
,
E. E.
,
Jonides
,
J.
,
DeGutis
,
J.
, &
Chenevert
,
T. L.
(
2000
).
Order information in working memory: fMRI evidence for parietal and prefrontal mechanisms.
Journal of Cognitive Neuroscience
,
12
,
130
144
.
McDermott
,
K. B.
,
Petersen
,
S. E.
,
Watson
,
J. M.
, &
Ojemann
,
J. G.
(
2003
).
A procedure for identifying regions preferentially activated by attention to semantic and phonological relations using functional magnetic resonance imaging.
Neuropsychologia
,
41
,
293
303
.
Meyer
,
P.
,
Mecklinger
,
A.
,
Grunwald
,
T.
,
Fell
,
J.
,
Elger
,
C. E.
, &
Friederici
,
A. D.
(
2005
).
Language processing within the human medial temporal lobe.
Hippocampus
,
15
,
451
459
.
Miller
,
G. A.
, &
Selfridge
,
J. A.
(
1950
).
Verbal context and the recall of meaningful material.
The American Journal of Psychology
,
63
,
176
185
.
Moses
,
S. N.
, &
Ryan
,
J. D.
(
2006
).
A comparison and evaluation of the predictions of relational and conjunctive accounts of hippocampal function.
Hippocampus
,
16
,
43
65
.
Mugler
,
J. P.
, &
Brookeman
,
J. R.
(
1990
).
3-Dimensional magnetization-prepared rapid gradient-echo imaging (3dmp-Rage).
Magnetic Resonance in Medicine
,
15
,
152
157
.
Murray
,
D. J.
,
Rowan
,
A. J.
, &
Smith
,
K. H.
(
1988
).
The effect of articulatory suppression on short-term recognition.
Canadian Journal of Psychology—Revue Canadienne De Psychologie
,
42
,
424
436
.
Oldfield
,
R. C.
(
1971
).
The assessment and analysis of handedness: The Edinburgh Inventory.
Neuropsychologia
,
9
,
97
113
.
Pallier
,
C.
,
Devauchelle
,
A. D.
, &
Dehaene
,
S.
(
2011
).
Cortical representation of the constituent structure of sentences.
Proceedings of the National Academy of Sciences, U.S.A.
,
108
,
2522
2527
.
Paulesu
,
E.
,
Frith
,
C. D.
, &
Frackowiak
,
R. S. J.
(
1993
).
The neural correlates of the verbal component of working memory.
Nature
,
362
,
342
345
.
Penhune
,
V. B.
,
Zatorre
,
R. J.
, &
Evans
,
A. C.
(
1998
).
Cerebellar contributions to motor timing: A PET study of auditory and visual rhythm reproduction.
Journal of Cognitive Neuroscience
,
10
,
752
765
.
Perham
,
N.
,
Marsh
,
J. E.
, &
Jones
,
D. M.
(
2009
).
Syntax and serial recall: How language supports short-term memory for order.
Quarterly Journal of Experimental Psychology
,
62
,
1285
1293
.
Potter
,
M. C.
, &
Lombardi
,
L.
(
1990
).
Regeneration in the short-term recall of sentences.
Journal of Memory and Language
,
29
,
633
654
.
Potter
,
M. C.
, &
Lombardi
,
L.
(
1998
).
Syntactic priming in immediate recall of sentences.
Journal of Memory and Language
,
38
,
265
282
.
Pylkkanen
,
L.
(
2008
).
Mismatching meanings in brain and behavior.
Language and Linguistics Compass
,
2
,
712
738
.
Rogalsky
,
C.
, &
Hickok
,
G.
(
2011
).
The role of Broca's area in sentence comprehension.
Journal of Cognitive Neuroscience,
,
23
,
1664
1680
.
Rottschy
,
C.
,
Langner
,
R.
,
Dogan
,
I.
,
Reetz
,
K.
,
Laird
,
A. R.
,
Schulz
,
J. B.
,
et al
(
2012
).
Modelling neural correlates of working memory: A coordinate-based meta-analysis.
Neuroimage
,
60
,
830
846
.
Salmon
,
E.
,
Van der Linden
,
M.
,
Collette
,
F.
,
Delfiore
,
G.
,
Maquet
,
P.
,
Degueldre
,
C.
,
et al
(
1996
).
Regional brain activity during working memory tasks.
Brain
,
119
,
1617
1625
.
Santi
,
A.
, &
Grodzinsky
,
Y.
(
2010
).
fMRI adaptation dissociates syntactic complexity dimensions.
Neuroimage
,
51
,
1285
1293
.
Schiffer
,
A. M.
,
Ahlheim
,
C.
,
Wurm
,
M. F.
, &
Schubotz
,
R. I.
(
2012
).
Surprised at all the entropy: Hippocampal, caudate and midbrain contributions to learning from prediction errors.
PLoS One
,
7,
,
e36445
.
Schweppe
,
J.
, &
Rummer
,
R.
(
2007
).
Shared representations in language processing and verbal short-term memory: The case of grammatical gender.
Journal of Memory and Language
,
56
,
336
356
.
Schweppe
,
J.
,
Rummer
,
R.
,
Bormann
,
T.
, &
Martin
,
R. C.
(
2011
).
Semantic and phonological information in sentence recall: Converging psycholinguistic and neuropsychological evidence.
Cognitive Neuropsychology
,
28
,
521
545
.
Sederberg
,
P. B.
,
Schulze-Bonhage
,
A.
,
Madsen
,
J. R.
,
Bromfield
,
E. B.
,
McCarthy
,
D. C.
,
Brandt
,
A.
,
et al
(
2007
).
Hippocampal and neocortical gamma oscillations predict memory formation in humans.
Cerebral Cortex
,
17
,
1190
1196
.
Squire
,
L. R.
(
1992
).
Memory and the hippocampus—A synthesis from findings with rats, monkeys, and humans.
Psychological Review
,
99
,
195
231
.
Stowe
,
L. A.
,
Broere
,
C. A. J.
,
Paans
,
A. M. J.
,
Wijers
,
A. A.
,
Mulder
,
G.
,
Vaalburg
,
W.
,
et al
(
1998
).
Localizing components of a complex task: Sentence processing and working memory.
NeuroReport
,
9
,
2995
2999
.
Vandenberghe
,
R.
,
Nobre
,
A. C.
, &
Price
,
C. J.
(
2002
).
The response of left temporal cortex to sentences.
Journal of Cognitive Neuroscience
,
14
,
550
560
.
Ward
,
D.
(
2000
).
AlphaSim: Estimate statistical significance via Monte Carlo simulation.
.
Wildgruber
,
D.
,
Ackermann
,
H.
,
Klose
,
U.
,
Kardatzki
,
B.
, &
Grodd
,
W.
(
1996
).
Functional lateralization of speech production at primary motor cortex: A fMRI study.
NeuroReport
,
7
,
2791
2795
.
Yarkoni
,
T.
,
Speer
,
N. K.
, &
Zacks
,
J. M.
(
2008
).
Neural substrates of narrative comprehension and memory.
Neuroimage
,
41
,
1408
1425
.
Yetkin
,
F. Z.
,
Hammeke
,
T. A.
,
Swanson
,
S. J.
,
Morris
,
G. L.
,
Mueller
,
W. M.
,
Mcauliffe
,
T. L.
,
et al
(
1995
).
A comparison of functional MR activation patterns during silent and audible language tasks.
American Journal of Neuroradiology
,
16
,
1087
1092
.
Zarahn
,
E.
,
Rakitin
,
B.
,
Abela
,
D.
,
Flynn
,
J.
, &
Stern
,
Y.
(
2005
).
Positive evidence against human hippocampal involvement in working memory maintenance of familiar stimuli.
Cerebral Cortex
,
15
,
303
316
.
Zhou
,
Y. D.
,
Ardestani
,
A.
, &
Fuster
,
J. M.
(
2007
).
Distributed and associative working memory.
Cerebral Cortex
,
17(Suppl 1)
,
i77
i87
.