Elsevier

NeuroImage

Volume 152, 15 May 2017, Pages 647-657
NeuroImage

Oscillatory EEG dynamics underlying automatic chunking during sentence processing

https://doi.org/10.1016/j.neuroimage.2017.03.018Get rights and content

Highlights

  • Sentences—even parts of sentences—are easier to remember than random word sequences.

  • The memory benefit is grounded in highly automatized language comprehension mechanisms.

  • These comprehension mechanism generate meaningful memory chunks irrespective of task demands.

  • Automatic linguistic chunking is reflected in delta oscillation amplitudes Delta activity originates from temporal brain regions involved in syntactic & semantic processing.

Abstract

Sentences are easier to remember than random word sequences, likely because linguistic regularities facilitate chunking of words into meaningful groups. The present electroencephalography study investigated the neural oscillations modulated by this so-called sentence superiority effect during the encoding and maintenance of sentence fragments versus word lists. We hypothesized a chunking-related modulation of neural processing during the encoding and retention of sentences (i.e., sentence fragments) as compared to word lists. Time–frequency analysis revealed a two-fold oscillatory pattern for the memorization of sentences: Sentence encoding was accompanied by higher delta amplitude (4 Hz), originating both from regions processing syntax as well as semantics (bilateral superior/middle temporal regions and fusiform gyrus). Subsequent sentence retention was reflected in decreased theta (6 Hz) and beta/gamma (27–32 Hz) amplitude instead. Notably, whether participants simply read or properly memorized the sentences did not impact chunking-related activity during encoding. Therefore, we argue that the sentence superiority effect is grounded in highly automatized language processing mechanisms, which generate meaningful memory chunks irrespective of task demands.

Introduction

In language comprehension, we depend on the fast and efficient retention of large amounts of incoming information in working memory. For example, to keep track of a constantly evolving conversation, we are able to temporally store recent conversation contents. In general, sentences are recalled very fast and accurately, whereas random word sequences load more heavily on working memory (i.e., the “sentence superiority effect”; cf. Baddeley et al., 2009; Bonhage et al., 2014; Brener, 1940; Jefferies et al., 2004). One possible explanation for the sentence superiority effect is that human memory benefits from linguistic regularities conveyed by grammar, because in general, working memory benefits from the presence of structure amongst to-be-stored items (Bor et al., 2001, Bor and Owen, 2007, Mathy and Feldman, 2012). For example, participants recall regular number sequences better than irregular number sequences (e.g., 1 3 5 7 8 6 4 2 versus 1 8 5 6 2 7 4 3). Thus, structure and relatedness amongst items likely facilitates working memory by enabling the generation of information chunks from single elements (Cowan, 2010, Gobet et al., 2001, Mathy and Feldman, 2012, Miller, 1956, Miller, 1994). Chunking has been shown to significantly expand memory capacity, especially if individual items can be linked on a conceptual level (Gilbert et al., 2014, Gilchrist, 2015).

In language comprehension, chunking might be achieved by utilizing the linguistic regularities set by syntactic rules, which indicate how single words are bound into phrases or sentences (Baddeley et al., 2009, Jefferies et al., 2004). In addition to syntactic rules, semantic relatedness allows us to combine single words into conceptual chunks and derive a sentence-level meaning (Lombardi and Potter, 1992, Potter and Lombardi, 1998, Potter et al., 1990). In line with this notion of semantic chunking, we reported stronger activation of semantic network areas during the encoding of sentences as compared to random word lists (such as middle temporal gyrus, inferior frontal gyrus, inferior parietal lobe, dorso-medial prefrontal cortex; cf. Bonhage et al., 2014). Moreover, during subsequent maintenance, we found evidence that grammatical regularities decrease neural maintenance demands: a substantial activation decrease in a fronto-parietal working memory network (e.g., inferior frontal gyrus, inferior parietal sulcus) for the retention of sentences as compared to random word lists was observed.

However, two important questions remained unanswered: First, is this proposed linguistic chunking an automatized language processing mechanism, or is it rather a specific linguistic working memory mode that is set into action only in the context of a working memory task? In other words: Is chunking-related activity only evident when participants are asked to remember sentences, or does it surface whenever participants read sentences, even without a memory task?

And second, what exactly are the neural mechanisms supporting the observed effects across memory encoding and retention? To answer these questions, in the present study we explored the neurophysiological dynamics associated with sentence encoding and retention using electroencephalography (EEG).

While prior EEG studies revealed neural oscillatory correlates of working memory encoding and retention, the neural oscillations underlying the sentence superiority effect across memory encoding and retention are unclear. Outside of sentence comprehension, encoding has been related to increased theta and beta frequency power (Fell et al., 2011, Klimesch et al., 2001, Klimesch et al., 1997, Raghavachari et al., 2006, Sederberg et al., 2007, Sederberg et al., 2003, White et al., 2013). Theta activity is a plausible candidate also from the perspective of language processing, as reading a sentence has been shown to gradually increase theta power (Bastiaansen et al., 2009, Bastiaansen et al., 2002a, Bastiaansen et al., 2002b). Oscillatory activity in the beta band (13–30 Hz) has also been reported to be sensitive to language manipulations – specifically, in their comprehensive review, Weiss and Mueller link beta activity to a range of processes relevant to language processing, such as working memory and binding (Weiss and Mueller, 2012). Moreover, working memory literature suggests that desynchronization in the beta band aids memory encoding (Hanslmayr et al., 2016, Hanslmayr et al., 2012). In addition to theta and beta band, Ding et al. (2015) as well as Meyer et al. (2016) recently observed that oscillatory activity in the delta band (1–4 Hz) relates to the building up of phrases and sentences. Because phrases and sentences are meaningful multi-word units that are linked through syntactic regularities, delta and theta oscillatory activity may directly reflect chunking processes.

During working memory retention of unstructured items on the other hand, increasing memory demands were observed to influence EEG oscillations in across frequency bands, specifically theta– (digits: Jensen and Tesche, 2002; consonants: Onton et al., 2005; Scheeringa et al., 2009), alpha– (consonants: Jensen et al., 2002; Payne and Kounios, 2009), and gamma-band responses (consonants: Howard et al., 2003). It is possible that these theta-, alpha-, gamma effects decrease if to-be-retained words can be chunked into meaningful units in grammatical word sequences instead of being stored word-by-word in ungrammatical word sequences.

In sum, the aim of the present study is two-fold: On the one hand, we aim to examine the oscillatory dynamics underlying the sentence superiority effect (which should be replicated at the behavioral level); on the other hand, we aim to uncover whether or not linguistic chunking is specific to the context of a working memory task, or whether the memory system instead employs an automatic binding process inherent to language comprehension. To solve this question, in a 2×2 factorial design the present study contrasts processing of sentence fragments and random word lists, both during mere reading and during encoding for a downstream recall task.

With respect to encoding, we had different hypotheses with respect to task-dependent and task-independent processes. On the one hand, linguistic chunking based on syntactic regularities and semantic relatedness could be an automatic process; in this case, we expected delta-band oscillatory responses to increase for sentence fragments as compared to random word lists, irrespective of whether the encoded material was to be recalled later. On the other hand, from what we know of oscillatory dynamics during successful memory formation, task-specific increases in theta and beta power could be expected for the encoding of sentence fragments as compared to random word lists. During memory retention—that is, when the encoded material has to be maintained for later recall—we expected higher theta-, and gamma-, but lower alpha- and beta amplitude compared to when participants were given time to pause without retention demands. Importantly, because sentence structure is thought to reduce working memory load, amplitude changes across these frequency bands might be less pronounced during the retention of sentence fragments as compared to the retention of random word lists.

Section snippets

Participants

Twenty right-handed (Oldfield, 1971; adapted German version) native Germans (no professional musicians) participated after giving written informed consent in accordance with the Declaration of Helsinki. One subject was excluded from analysis due to insufficient EEG data quality: Even after automatic artifact rejection (which already removed >30% of the trials) multiple frontal electrodes displayed highly noisy signal throughout the measurement. The remaining 19 participants (9 female; mean age,

Behavioral results

Accuracy (cf. Fig. 2A) was well above chance in all conditions (Z >=3.80, p<0.001). Participants' accuracy was higher for sentence fragments (SF) as compared to word lists (WL; F(1,18)=56.07, p<0.001, part. η2=.76), and for the lexical decision task as compared to the working memory task (F(1,18)=36.36, p<0.001, part. η2=.67). Additionally, STRUCTURE interacted with TASK (F(1,18)=43.93, p<0.001, part. η2=.71; cf. Fig. 2A). Specifically, during the working memory task sentence structure had an

Discussion

Successful conversation requires efficient memory storage of conversation contents. The present study investigated the neural oscillatory correlates of the sentence superiority effect, that is, the enhanced memory performance for recalling words that form a (partial) sentence. Here, on the one hand, we strived to identify the oscillatory dynamics underlying the sentence superiority effect (which should be replicated on the behavioral level). On the other hand, we aimed to figure out whether or

Conclusion

Sentence structure unburdens memory (i.e., the sentence superiority effect) – an effect largely neglected in language comprehension research. This is the first study exploring (a) how the oscillatory dynamics unfold over encoding and retention of sentence fragments versus word lists in order to realize linguistic chunking and (b) whether this linguistic chunking is specific to working memory task or rather relies on basic language comprehension mechanisms and thus takes place irrespective of

Conflict of interest

The authors declare no competing financial interests.

Acknowledgments

The authors are highly grateful to Galina Surova for the excellent data acquisition and to Nelson Trujillo-Barreto for his support regarding EEG data analysis.

References (74)

  • T. Gruber et al.

    Brain electrical tomography (BET) analysis of induced gamma band responses during a simple object recognition task

    NeuroImage

    (2006)
  • S. Hanslmayr et al.

    Oscillations and episodic memory: addressing the synchronization/desynchronization conundrum

    Trends Neurosci.

    (2016)
  • U. Hassler et al.

    Induced gamma band responses in human EEG after the control of miniature saccadic artifacts

    NeuroImage

    (2011)
  • E. Jefferies et al.

    Automatic and controlled processing in sentence recall: the role of long-term and working memory

    J. Mem. Lang.

    (2004)
  • E. Kaan et al.

    The brain circuitry of syntactic comprehension

    Trends Cogn. Sci.

    (2002)
  • W. Klimesch

    EEG alpha and theta oscillations reflect cognitive and memory performance: a review and analysis

    Brain Res. Rev.

    (1999)
  • W. Klimesch et al.

    Brain oscillations and human memory: EEG correlates in the upper alpha and theta band

    Neurosci. Lett.

    (1997)
  • W. Klimesch et al.

    Theta synchronization during episodic retrieval: neural correlates of conscious awareness

    Cogn. Brain Res.

    (2001)
  • M. Köster et al.

    Theta-gamma coupling during episodic retrieval in the human EEG

    Brain Res.

    (2014)
  • A.G. Lewis et al.

    Fast oscillatory dynamics during language comprehension: unification versus maintenance and prediction?

    Brain Lang.

    (2015)
  • L. Lombardi et al.

    The regeneration of syntax in short term memory

    J. Mem. Lang.

    (1992)
  • F. Mathy et al.

    What's magic about magic numbers? Chunking and data compression in short-term memory

    Cognition

    (2012)
  • L. Meyer et al.

    Left parietal alpha enhancement during working memory-intensive sentence processing

    Cortex

    (2013)
  • L. Meyer et al.

    Frontal–posterior theta oscillations reflect memory retrieval during sentence comprehension

    Cortex

    (2015)
  • H. Nolan et al.

    FASTER: fully automated statistical thresholding for EEG artifact rejection

    J. Neurosci. Methods

    (2010)
  • R.C. Oldfield

    The assessment and analysis of handedness: the Edinburgh inventory

    Neuropsychologia

    (1971)
  • J. Onton et al.

    Frontal midline EEG dynamics during working memory

    NeuroImage

    (2005)
  • S. Palva et al.

    New vistas for alpha-frequency band oscillations

    Trends Neurosci.

    (2007)
  • L. Payne et al.

    Coherent oscillatory networks supporting short-term memory retention

    Brain Res.

    (2009)
  • M.C. Potter et al.

    Syntactic priming in immediate recall of sentences

    J. Mem. Lang.

    (1998)
  • M.C. Potter et al.

    Regeneration in the short-term recall of sentences

    J. Mem. Lang.

    (1990)
  • R. Scheeringa et al.

    Trial-by-trial coupling between EEG and BOLD identifies networks related to alpha and theta EEG power increases during working memory maintenance

    NeuroImage

    (2009)
  • N.J. Trujillo-Barreto et al.

    Bayesian model averaging in EEG/MEG imaging

    NeuroImage

    (2004)
  • A. Von Stein et al.

    Different frequencies for different scales of cortical integration: from local gamma to long range alpha/theta synchronization

    Int. J. Psychophysiol.

    (2000)
  • E. Wascher et al.

    Frontal theta activity reflects distinct aspects of mental fatigue

    Biol. Psychol.

    (2014)
  • C.L. Wiggs et al.

    Properties and mechanics of perceptual priming

    Curr. Opin. Neurobiol. Opin. Neurobiol.

    (1998)
  • S. Yuval-Greenberg et al.

    Transient induced gamma-band response in EEG as a manifestation of miniature saccades

    Neuron

    (2008)
  • Cited by (43)

    • Oscillatory delta and theta frequencies differentially support multiple items encoding to optimize memory performance during the digit span task

      2022, NeuroImage
      Citation Excerpt :

      In EEG literature, several studies studied “chunking" in the scope of language processing (Bonhage et al., 2017; Gilbert et al., 2014, 2015). According to these studies, delta-theta oscillatory responses (Bonhage et al., 2017), as well as N400 and P300 components (Gilbert et al., 2014), may reflect chunking-related encoding during language processing. Additionally, Nogueira et al (2015) showed increased late positive slow waves during the encoding of the words when the items were chunked (Nogueira et al., 2015) compared to a control condition.

    • Event-related responses reflect chunk boundaries in natural speech

      2022, NeuroImage
      Citation Excerpt :

      Speech unfolds as a continuous flow of acoustic information where extraction of meaning has been proposed in theoretical models to be achieved by chunking speech into smaller units (Christiansen and Chater, 2016; Sinclair and Mauranen, 2006). Chunking language is understood to be an automatized, integrated, and multi-level process (e.g., Bonhage et al., 2017). A considerable body of research involving short, controlled stimuli suggests that segmentation of oral speech is associated with neural correlates at multiple timescales simultaneously (Doelling et al., 2014; Ghitza and Greenberg, 2009; Giraud and Poeppel, 2012; Gross et al., 2013; Henke and Meyer, 2021; Peelle et al., 2013).

    View all citing articles on Scopus
    View full text