Comprehending language in the brain

Get Complete Project Material File(s) Now! »

Comprehending language in the brain

We start this thesis by discussing neurocognitive approaches that have been used in studying the neural basis of language. We talk briefly and generally about these approaches but put more emphasis on techniques that are directly related to our study (i.e., electroencephalography – EEG and magnetoencephalography – MEG), to familiarize the readers with them and with the terms that are related to previous studies of subject-verb agreement in the following chapter.

Neurocognitive techniques used to track language processing

Endeavors to understand how our brain processes language started in the nineteenth century. Initially, studies related to the language function in the brain were carried out in patients who had suffered brain damage or neurological disorders such as strokes or aphasia. Brain analysis was thus performed post-mortem. For instance, Paul Broca (1861) discovered the Broca area by studying patients who had suffered aphasia (i.e., a disability that cause difficulty in language comprehension). Studying damaged brains has shown which areas of the brain are involved during language processing. These studies have helped in understanding the impact of brain lesions on language. In other words, this method is fine to isolate brain areas related to a certain function. However, its major drawback is that after suffering brain lesions, patients may present a different cerebral reorganization from one patient to another. Furthermore, this technique does not provide any information about the online processing of language. For instance, as for clinical studies in patients with aphasia, the results could not explain how complex grammatical structures, such as embedded sentences, are processed on-line (Caramazza et al., 1978; Caramazza & Zurif, 1976; Friederici, 1982). Additionally, the results seemed to depend on the population (Sherman & Schweickert, 1989), since the type of aphasia may differ from one patient to another. For example, Sherman & Schweikert (1989) found that their aphasic participants performed well above chance in interpreting passive sentences, while other studies (Caplan & Futter, 1986; Schwartz et al., 1980) suggested that the accuracy of aphasic participants on a similar task was due to chance. Moreover, aphasic participants in Sherman & Schweikert (1989) were still able to perform syntactic processing despite their syntactic deficit. Nowadays, owing to the development of new technologies, we can compensate these limitations thanks to non-invasive neuroimaging techniques that help in collecting data from healthy participants.
These non-invasive neuroimaging techniques measure brain activity while participants perform certain tasks. Online processing can therefore be investigated. These techniques may be divided into two groups: those that focus on neural activities and have high temporal resolution, and those that focus on the structural system of the brain and have high spatial resolution. The former measure brain activity directly, such as electroencephalography (EEG) and magnetoencephalography (MEG). The latter measure it indirectly, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET). Each of these techniques has advantages and disadvantages related to spatial and temporal resolution (Bunge & Kahn, 2009) associated with the measurement approach (see Figure 1. for illustration). As already mentioned, techniques that measure brain activity directly have high temporal resolution but poor spatial resolution, and vice versa for indirect measurement techniques. Each technique is discussed in more detail in the following sub-section: it starts with techniques that have high spatial resolution (fMRI and PET) and then switches to techniques that have high temporal resolution (EEG and MEG). We then describe the results obtained from these techniques, such as brain areas and event-related potential components that are related to language processing.

Functional magnetic resonance imaging

Functional MRI (fMRI) is an advanced technique that makes it possible to record brain activity and map it to high spatial and temporal resolution (Kim & Bandettini, 2010). It can be used to take brain images in high resolution while investigating the brain without dissecting it, as it measures the blood oxygen level-dependent (BOLD) signal (Bandettini, 2020; Heim & Specht, 2019). fMRI measures the metabolic brain response that comes from changes in the hemodynamic response through the BOLD signal: the oxygen is transmitted to neurons by hemoglobin because it has magnetic properties. The BOLD signal consists of hydrogen atoms that are present in water molecules in the brain. These hydrogen atoms then take up the energy provided by the magnetic field that comes from hemoglobin: this magnetic field can either be diamagnetic when the hemoglobin is oxygenated or paramagnetic when the hemoglobin is deoxygenated. This oxygenating process causes differences in the magnetic properties that affect the discharge of hydrogen atoms, it discharges energy at the same radio frequency until they are in a state of balance. The total sum of the discharged radiofrequency energy is then calculated by the MRI scanner. The energy calculated will decay over time due to several factors, one of them being homogeneities in the magnetic field. The latter also affects image intensity. Moreover, the BOLD signal not only depends on blood flow that has radiofrequency excitation but also on the inflow of fresh blood, which has not undergone excitation. Since the metabolic signal depends on the oxygen consumed in related brain areas, the temporal resolution is slower in comparison with EEG and MEG. Normally, the peak of the BOLD signal is recorded around five to six seconds after the stimulus input. Therefore, this technique is good for localization studies but might be not a good choice if one wants to study a process that concerns precise moments in time.

PET scan

Unlike fMRI, PET uses radioactive isotopes (e.g., carbon, nitrogen and oxygen molecules) to measure the changes in brain blood flow. It measures brain function through regional cerebral blood flow (rCBF), metabolism, neurotransmitters and radiolabeled drugs (Berger, 2003). This method uses radioactivity by detecting the isotope that is injected into the vein. The isotope enters the brain after around 30 seconds and in the following 30 seconds, the radiation reaches a peak that can be detected by a PET scanner. For experimental purposes, the measurements can be done up to ten times or more, with ten to fifteen minutes’ intervals. The strongest activation is normally found in an area that is constantly activated during an experiment. In terms of resolution, PET has a lower spatial and temporal resolution than fMRI, so the latter may also be performed for localization purposes. The temporal resolution is also very poor since there is a delay of several seconds. The acquired data is therefore an indirect measure of blood flow and, akin to fMRI, it does not measure neural activity. Nevertheless, PET also has advantages, as it is not as sensitive to movement as fMRI thus PET data can therefore not easily be affected by a movement artifact. Moreover, PET provides information about the neuronal metabolism and the neurotransmission

Electroencephalography

Electroencephalography (EEG) is the oldest non-invasive technique that measures brain activity. It was first developed by Hans Berger (Gloor, 1969) who was a physiologist. In the beginning, he worked with patients who had skull defects from the war. He then discovered that brain activity could also be recorded from a normal scalp. Importantly, at that time, he only used two electrodes to record brain activity over the frontal and occipital sites. Rather than locating the neuronal sources, he was more focused on the notion of integrated activity of the whole brain. He observed its repetitive electrical activity and its particular rhythmicity, and labeled them as alpha and beta activities. The frequency of alpha activities ranges between 8 and 12 Hz, and it is related to processes that require attention. The frequency of beta activities ranges between 15 and 30 Hz, and it is related to working memory and motor processes.
The waves identified from recorded electrical currents came from many synchronous pyramidal neurons. The current itself is produced by neurons that communicate with other neurons. When the information is received by the other neurons, it results in a postsynaptic potential (PSP). In PSP, there is a temporary change in the electric polarization of the membrane of neurons because excitatory or inhibitory neurotransmitters are released. These voltage differences generate radial electric polarization (i.e., electrical field that contains positive and negative current, see Figure 2a) that reaches the scalp surface and is then picked up by the EEG electrodes. To record this electrical activity, ground and reference electrodes are also needed (Luck, 2014). The ground electrode functions as a virtual ground that eliminates the surplus of static electricity in participants. The ground electrode is used only for the aforementioned purpose, so while it does not record any signal, it is required to allow the recording of other electrodes and reference channels. The reference electrode is required for noise reduction purposes. It is normally placed on the mastoid, i.e. the backbone of the ear. Exogenous electrodes are also used to detect artifacts that arise from biological signals, such as eye blinks and eye movements, which are usually unavoidable during experimental trials.
Although EEG can measure the electric activity to the millisecond, it lacks spatial resolution. This is due to the fact that EEG is more sensitive to the radial dipole, which is close to the scalp surface. The signal also degraded when it passes through the skull. It is therefore more difficult to localize the source, which is also known as the inverse problem. Moreover, the number of electrodes that cover the scalp area in EEG was limited. This is also considered as a disadvantage for source localization because with a small number of electrodes, it is hard to pinpoint the source of activity. However, EEG caps with up to 260 electrodes are now available. The more electrodes are used in the experiments, the better the possibility to achieve source localization, although 64-128 electrodes are adequate for this purpose. Several mathematical methods have also been developed to enhance the accuracy of source localization (e.g., LORETA and sLORETA).

Table of contents :

Preface
Chapter 1 Comprehending language in the brain
1.1 Neurocognitive techniques used to track language processing
1.1.1 Functional magnetic resonance imaging
1.1.2 PET scan
1.1.3 Electroencephalography
1.1.4 Magnetoencephalography
1.2 Results from brain studies
1.2.1 Brain regions related to language processing
1.2.2 Event-related potential components
Chapter 2 Grammatical agreement
2.1 Forming verb agreement through morphosyntax
2.1.1 Morphology
2.1.1 Syntax
2.2 Subject-verb agreement
2.2.1 Agreement features
2.2.1.1 Person
2.2.1.2 Number
2.2.1.3 Gender
2.2.2 Agreement mechanism: a linguistic perspective
2.2.2.1 French verb agreement
2.2.3 Agreement mechanism: a neurocognitive perspective
2.2.3.1 Left anterior negativity
2.2.3.2 P600
2.2.3.3 Biphasic LAN-P600
Chapter 3 Mental representations
3.1 What are mental representations?
3.2 Language representations
3.2.1 Abstract representations
3.2.1.1 The organization of abstract representations
3.2.2 Associative representations
Chapter 4 Language and cognitive flexibility
4.1 What is cognitive flexibility?
4.2 Cognitive flexibility in language processing
4.3 Flexibility vs. automaticity in grammatical agreement: evidence from electrophysiological studies
Chapter 5 Prediction
5.1 From perception to prediction
5.2 Prediction in language
5.2.1 Phonological prediction
5.2.2 Word prediction in spoken language comprehension
5.3 Prediction vs. integration
5.4 Studying prediction in agreement processing
Chapter 6 Aims and hypotheses
Chapter 7 Study 1: EEG experiment to explore the nature of representations
7.1 Predictions
7.2 Methods
7.2.1 Participants
7.2.2 Materials
7.2.2.1 Stimuli recording
7.2.3 Experimental procedure
7.2.4 EEG data acquisition
7.2.5 EEG data pre-processing
7.3 ERP analysis
7.4 Results
7.4.1 Behavioral results
7.4.2 ERP results
7.4.2.1 Time window between 100 and 160 ms
7.4.2.2 Time window between 300 and 600 ms
7.4.2.3 Time window between 650 and 850 ms
7.4.2.4 Time window between 920 and 1120 ms
7.5 Discussion
7.5.1 N100
7.5.2 Anterior and fronto-anterior negativity
7.5.3 Late P600
7.5.4 The role of representations in subject-verb agreement
7.6 Conclusions
Chapter 8 The second study: EEG study with two experiments to explore flexibility in accessing the representations
8.1 Predictions
8.2 Methods (Experiment 2)
8.2.1 Participants
8.2.2 Materials
8.2.3 Experimental procedure
8.2.4 EEG data acquisition
8.2.5 EEG data pre-processing
8.3 ERP analysis
8.4 Results
8.4.1 Experiment 2
8.4.1.1 Behavioral results
8.4.1.2 ERP results
8.4.1.2.1 Time window between 100 and 160 ms
8.4.1.2.2 Time window between 300 and 600 ms
8.4.1.2.3 Time window between 650 and 850 ms
8.4.1.2.4 Time window between 920 and 1120 ms
8.4.1.3 Discussion
8.4.2 Comparsion of Experiment 1 and Experiment 2
8.4.2.1 Behavioral results
8.4.2.2 ERP results
8.4.2.2.1 Time window between 100 and 160 ms
8.4.2.2.2 Time window between 300 and 600 ms
8.4.2.2.3 Time window between 650 and 850 ms
8.4.2.2.4 Time window between 920 and 1120 ms
8.4.2.3 Discussion
8.4.2.3.1 N100
8.4.2.3.2 N400 and late anterior negativity
8.4.2.3.3 Late P600
8.4.2.2.4 Flexibility in agreement processing
8.5 Conclusions
Chapter 9 Study 3: MEG study to investigate prediction in agreement processing
9.1 Predictions
9.2 Method
9.2.1 Participants
9.2.2 Materials
9.2.3 Procedure
9.2.4 MEG data acquisition
9.2.5 MEG data pre-processing
9.3 Statistical analyses
9.3.1 Whole-brain analysis
9.3.2 fROI analysis
9.4 Results
9.4.1 Behavioral results
9.4.2 MEG results
9.4.2.1 Results from the whole-brain analysis
9.4.2.1.1 Left hemisphere
9.4.2.1.2 Right hemisphere
9.4.2.2 Results from the fROIs analysis
9.5 Discussion
9.5.1 Abstract representations in temporal lobe
9.5.2 Associative representations and prediction
9.5.3 IFG and syntactic processing
9.5.4 Motor area and anterior cingulate cortex
9.6 Conclusions
Chapter 10 General discussion
10.1 Summary of results
10.2 Representations in agreement processing
10.2.1 Abstract representations
10.2.2 Associative representations and top-down processing
10.2.2.1 Motor area and prediction
10.3 Flexibility in accessing the representations
10.4 Future directions
10.5 Conclusions
References
Appendix A List of the stimuli
Appendix B Statistical table
Appendix C Submitted paper

GET THE COMPLETE PROJECT