Breaking down the linguistic processing hierarchy with meditation: a ERP pilot study
Mitchell de Roo
Supervisor: dr. Ruben Laukonen
Assessor: dr. Marte Otten
Contents
Abstract...4
Covid-19 Project Restrictions...5
Introduction...6
The predictive framework...6
The free energy principle...7
Meditation...8
The relationship between meditation and the predictive framework...8
The goal of this study...10
Methods...12
Participants...12
Stimuli...12
Experimental Procedure...13
EEG acquisition...14
Pre-processing...14
ERP & statistical analysis...15
Results...16 Discussion...21
The N400...22
The N1-P2...23
P300...24
The LPC...24
The auditory processing hierarchy...25
Limitation of this study...25
Future direction/ Points of improvement...26
Conclusion...27
References...28
Abstract
Meditation is growing in popularity in western society along with a rise of empirical evidence indicating its long-lasting changes on brain activity. We will frame meditation in the light of the predictive processing theory and use this to explain the unique effects observed within two different meditation practices. In order to test to what extent meditation can alter auditory processing we have designed three conditions that reflect linguistic, semantic or emotional processing. In order from low to high they represent the auditory processing hierarchy. In this report we will discuss the pilot data, preceding the main experiment, assessing the event-related potential (ERP) data between each condition independent of a meditation condition. In line with our hypothesis we observed a significant effect between emotional stimuli reflecting the N400 component. Furthermore, we report emotional words to elicit a large positive potential (LPC), and pure tones (linguistic condition) to evoke an N1-P2 and P300. The temporal order of appearance per condition confirms their distinct place in the auditory processing hierarchy. At last we will discuss some of the limitations of the current data, points of improvement and future directions.
Covid-19 Project Restrictions
Before diving into this report, I want to discuss its structure and the circumstance
surrounding the internship. Due the covid-19 restrictions we were limited from being able to do any further EEG testing, while in the middle of collecting the pilot data. We were able to collect pilot data of five participants that was limited to stimulus presentation and does not include meditation
conditions. Furthermore, the pilot data is of low power making statistical analyses problematic. Despite this, we have decided to continue this internship and analyse the data as to finalize the methodological pipelines and do exploratory analysis. I have decided to keep the introduction in the framework of the main experiment to provide a broader theoretical context. The methodology and results will however be limited to the analysed pilot data. In the discussion I will discuss the pilot data, but also relate the data to the context of the main experiment and discuss points of
Introduction
As modern society changes at an accelerating rate our brain tries to adjust to this new and complex lifestyle. Although we observe large degrees of plasticity in the brain the extent of this adaptability remains unclear (May & Gaser, 2006). Meditation, an age-old Buddhist practice that shows an exponential growth in popularity in western society over the last 20 years, may reveal new insights on the malleability of the human brain (Van Dam et al. 2018). Its long history has caused a wide diversity of different practices, philosophies, and terminologies. While the different practices cannot be captured in a single overarching explanation their common ground is the focus on
attentional and emotional control (Teper, Segal & inzlicht, 2013; Lutz et al., 2015). Extensive practice towards this state could lead to the ability to voluntary control perception and attention. With
meditation growing in popularity and its supposed potential benefits and effects on human cognition it is therefore important to construct an empirical and comprehensive understanding of meditation and its mechanisms (Van Dam et al. 2018).
The predictive framework
First, we will explore the predictive processing theory, a framework we will later use to explain how neural processes are affected by meditation. The theory offers a unifying framework for cognition, action, perception and attention by understanding the neural processes of the brain in terms of hierarchical probabilistic generative models aiming to reduce prediction error (Clark, 2013; Friston,
2010). The theory states there is a multilevel bidirectional cascade of information processing in which the higher-order levels of the brain try to predict the incoming sensory information of lower order levels with the help of probabilistic generative models. This continuous top-down sensory prediction is opposed by bottom-up sensory processes which determines the mismatch between the prediction and the input – known as the prediction error. In turn, the prediction error is propagated through the system to the higher-order layers allowing for adaptation of their generative model in an attempt to reduce future prediction errors. The models higher up in the hierarchy will encode for more conceptual and abstract concepts of the external world, while lower levels of the hierarchy will be more concrete and unimodal (Friston & Stephan, 2007).
The free energy principle
Karl Friston (2010) elaborates on the predictive processing framework in the free energy principle. The free energy principle is an information theory that states that biological systems that are in equilibrium with their environment aim to minimize their internal probability of surprise, known as entropy. By reducing its internal free energy, the system can more accurately model the causal structures of the external world (Friston & Stephan, 2007). The free energy principle elaborates on the predictive framework by assuming that the bottom-up propagated mismatch is estimated based upon weighting the precision of the sensory signals and the models (or prior beliefs). Thus, the accuracy of the sensory data is situationally dependent on the weighting of top-down and bottom-up information which can be adjusted using attention. For example, the system can choose to assign higher precision to the generative models, rather than the prediction error, to rely more on prior beliefs than sensory input. Besides accounting for precision within the hierarchy, the accuracy of the models can be adjusted via perceptual inference by predicting sensory stimuli based upon its prior beliefs. The free energy principle elaborates on this further by proposing that the agent can also actively search for evidence that confirms its prediction. Furthermore, via active engagement with the environment, in terms of motoric action, the system can confirm its sensory prediction, this is known as active
inference (Aggelopoulos, 2015). For example, touching an empty window to confirm that the glass is indeed missing.
In summary, the free energy principle proposes the brain to continuously update and control its internal models with respect to the weighting of stimulus input and prior beliefs in order to reduce its internal entropy via the help of perceptual and active inference. Long-term meditation could adjust the weighting and precisions of the models and voluntarily alter the agents internal state.
Meditation
Understanding and operationalizing meditation in an empirical context is complex. As meditation covers a long history, the present-day term ‘meditation’ can be related to a broad range of practices, philosophies and terminology. While many of these different branches focus on self-regulating techniques to increase attentional and emotional control generalizing meditation under such a description is an oversimplification (Wenk-Sormaz, 2005; Lutz et al., 2008). The different branches of meditation have been reported to differ in cognitive processes, attentional scope, and conflict monitoring (Lippelt et al., 2014). With such a variety of meditation practice characteristics it is important to find and classify similarities into distinct meaningful categories. Lutz et al. (2008)
proposes a two-way divide of common practices observed within meditation: focussed attention (FA)
and open monitoring (OM). FA aims to sustain selective attention on a chosen object or event while
disengaging from internal and external distractors. OM describes a form of meditation on which one
aims for an open and flexible awareness while aiming for a non-reactive monitoring towards any
internal or external stimuli. It should be noted that over the course of reaching an expert level of
meditation, techniques from both sides are often practiced, usually beginning with FA to reach a
stable present moment awareness, and then OM which is considered a more advanced practice. The
practices can be understood in the context of a continuum. Where FA requires the practitioner to focus
on an object in the present moment independent of distractors, OM demands an even deeper level of
The relationship between meditation and the predictive framework
We can understand meditation as a range of techniques that aim for a voluntarily modulation of attentional and emotional reactivity (Lutz et al., 2008), however, the extent of this regulation and its accompanied phenomenological changes in an empirical context is yet unclear. Predictive processing in conjunction with the free energy principle propose a theoretical paradigm that offers a framework in which the action and effects of meditation can be understood (Pagnoni, 2019).
But how can we understand the different types of meditation practices in light of the free-energy principle? As discussed earlier, the higher-order level generative models send predictions of lower-level input downwards and the lower levels respond with the prediction error. In turn, the models adapt based upon the mismatch. However, the precision weighting towards sensory input or the prior knowledge inherent in the models is dependent on attentional control. Some of these models show rigid mechanisms to account for these prediction errors (Yon et al., 2019). Meditation could voluntarily alter this attentional control adjusting the balance of precision weighting to lean more towards sensory predictions and decrease the stubbornness of the models higher in the processing hierarchy (Pagnoni, 2019).
Previous literature has reported differential processing in expert meditators when using an oddball paradigm (Biedermann et al., 2016; Srinivasan & Baijal, 2007). In these studies meditators and non-meditators are presented a series of similar tones with a low chance for a deviant tone to appear. Such a paradigm is known to evoke the mismatch negativity (MMN) ERP component. The MMN is thought to reflect the process of updating the perceptual representation of mismatched auditory stimuli (Garrido et al., 2009; Winkler, 2007). In other words, the MMN could be a marker for the prediction error that lower order levels propagate upwards. Interestingly, expert meditators show an increased MMN when compared to non-meditators (Biedermann et al., 2016; Srinivasan & Baijal, 2007). These results could indicate that long-term meditations cause an increased accuracy towards the representation of information offered by the lower-level models.
Alternatively, we see that weighting of the priors decreases (Kirk & Montagu, 2015). In a condition paradigm expert meditators and a control group were conditioned with a yellow light cue
followed by a reward. In consecutive trials catch events were introduced with a delayed juice delivery. These catch events cause a prediction error in both the unexpected non-delivery of juice (negative prediction error) and the unexpected time of the delayed delivery (positive prediction error). Here, the expert meditators showed a decreased BOLD response for both the positive and negative prediction error when compared to the control group. This decreased brain activity represents a decreased influence of the prior information achieved via conditioning.
These results show altered perceptual information processing in long-term meditators. However, as earlier discussed meditation differ in their practices. FA aims to attend to an object while disengaging from distractors, and therefore, requires reducing the influence of higher order levels by upweighting the precision of lower order sensory input of the object in focus. Going further on the continuum reaching a level of OM a non-reactive global attentional mind state is required. Through the lens of the free energy principle this requires an equally distributed precision of lower-order sensory input while reducing the precision of the higher-order models. A recent study by Fucci et al. (2018) investigated the difference between FA and OM using an oddball paradigm. Here they used the MMN and the late frontal negativity as markers for predictive processing and attentional monitoring of the sensory environment, respectively. They found an increased MMN in the FA condition compared to the OM. While the OM condition showed an increased late frontal negativity compared to the FA condition. Thus, the continuum observed in meditation practices could therefore be understood as reducing the habit of higher-order hierarchical processing, first by increasing the precision of sensory input (FA) followed by increasing the precision-weighting of the whole field of experience (OM) (Dahl et al., 2015).
The goal of this study
The goal of this study is to research the different effects of both FA and OM practices on auditory attentional processing. This will be done by measuring the electroencephalography (EEG) response to different auditory conditions of novice and expert meditators practicing both FA and OM. To test the extent of meditation on auditory processing we will compare different condition, each
representing a different stage of auditory processing. The stimuli will be compared such that emotional (negative vs neutral words), semantic (negative and neutral words vs pseudowords) and linguistic (words vs pure tones) representation dynamics of linguistic processing can be analysed. At the most extreme level of differentiation we will compare words to pure tones. The difference between pure tones and words in the ERP is already observed as early as 50 ms (Okita et al., 1983; Näätänen & Picton, 1987). This early temporal ERP component represent the lowest level of the hierarchy where auditory information is differentiated on having linguistic properties or not. When we proceed to a later time point in the ERP, we will observe different effect based on the stimulus semantical properties. For example, at this level in the hierarchy the processing semantical
unpredictability elicits a N400 (Ding et al., 2016). Thus, if long term meditators can break down the incoming semantical information, we would observe a diminished difference between the ERP’s of meaningful words and pseudowords. At an even higher layer of the hierarchy is the processing of emotional content. The effects of emotional properties are observed in the late ERP components such as the late positive component (Kotz & Paulmann, 2011). This temporal order in the ERP’s reflects the hierarchical processing of auditory information. Hearing auditory information can be broken down in the temporal order by which the brain encodes this information, first linguistic properties followed by its semantic content and finally it will register the emotional load. Thus, to test how deep long-term meditation affects the auditory processing hierarchy we have decided to compare across three different conditions corresponding with the temporal order of auditory processing as seen in the ERP’s.
EEG will be analysed using event-related potentials (ERP). We expect the differences between the conditions to be characterized via dissimilarities observable in various ERP components. For the emotional comparison we expect that the negative-laden emotional stimuli elicit a larger LPC than neutral words (Kotz & Paulmann, 2011; Rostami et al., 2016). When comparing on a semantic level we hypothesize that the semantic content of the pseudowords will cause an N400 effect compared to the other word conditions (Leinonen et al., 2009; Ding et al., 2016). Finally, at the highest level of contrast, and lowest level of the hierarchy, we expect the pure tones to differentiate
from the words during early processing such that the N1 and P2 components will be be larger in the pure tone conditionthan the word conditions (Okita et al., 1983; Näätänen & Picton, 1987).
Methods
Participants
Five Dutch speaking participants took part in the pilot study. Participants reported normal bilateral hearing and no personal or family history of epilepsy or neurological conditions. Participants were recruited via VU lab and rewarded with course credits or were part of the research group. These participants were not selected based on meditation expertise, neither did we control for this. The experiment is approved by the research ethics committee of the Vrije Universiteit Medical Centre.
Stimuli
Stimuli consisted of the following four categories; neutral words (kiosk, aspect, sleutel), negative words (kanker, misbruik, kotsen), pseudowords (arpen, elijt, polosk), and pure tones each containing 15 unique Dutch (pseudo)words or tones. Pseudowords were created by mixing the syllables of neutral words. All words and pseudowords were synthesized into speech using Natural Reader at the speed setting -3 (https://www.naturalreaders.com/online/). Pure tones were generated using Ableton live 10 Operator (https://www.ableton.com). Pure tones were created using sine waves and randomized in frequency and intensity. Editing of the pure tones to adjust length was done using audacity (https://www.audacity.com).
Data on word valence, arousal, power, age of acquisition, frequency (log10), frequency (per million), length (number of letters), N% and noun or verb usage was acquired via a study from Moors et al. (2013). Negative and neutral words were matched on all dimensions except valance, arousal & power. Negative words were intentionally chosen to have a significantly higher arousal, valence and power to represent negative emotional load. Phonological neighbourhood (PTAN) and frequency (PTAF) was tested using Clearpond and matched to negative and neutral words (Marian et al., 2012). All words and pseudowords were analysed on neighbourhood density and bigram frequency using
Wordgen (Duyck, 2004). Words and pseudowords were matched on bigram frequency but not neighbourhood density.
We analysed the pitch, intensity (dB) and length (seconds) of each word condition and the pure tones using Praat (Boersma et al., 2020). Words across all conditions were matched on pitch, intensity, and length. Statistical data per condition and testing results between conditions are shown in Table 2 and 3 in the Appendix.
After pilot testing, we observed limitations in our stimuli production and presentation. At first, we discovered that OpenSesame lags in its auditory stimuli presentation (Bridges et al., 2020). This causes a delay between logging the auditory stimulus presentation and the moment actual auditory stimulus presentation, to compensate for this lag we added a general 40 ms delay to all stimuli before pre-processing. Secondly, all word conditions were designed without removing the silence at the start of each word. This resulted in a 0 to 235 ms variance of silence before audibility of the stimuli. We have had two separate observers calculate the lag of each word, compared both observations and averaged over both observations to acquire the lag per word (large differences in lag were assessed and discussed until a consensus was reached). The final averaged lag unique for each word was also added before pre-processing.
Experimental Procedure
Each participant was informed of the general overview of the experiment and asked to reduce movements to a minimum and passively listen to the auditory stream presented while keeping their eyes closed. We did not ask the participants to meditate. After addressing remaining questions all participants were given and signed a consent form.
Stimuli were presented in a closed dimly lit soundproof room via binaural speakers set on a comfortable intensity level. No practice trials were included. Stimuli order was randomized using OpenSesame (Mathôt et al., 2012). The representation order of all conditions was randomized such that no category or word was repeated. Every stimulus, per condition, was included a total of six
times. Within each category we also randomized the word order. Interstimulus interval was randomized between 1500 and 1700 ms.
EEG acquisition
EEG data was recorded at 512 Hz via a 64 -A/AgCl electrodes according to the extended 10/20 system using a BioSemi ActiveTwo system. Additionally, electrodes were recorded from both earlobes and used as a reference signal. Impedance level was not measured since high impedance levels do not affect statistical power in cool and dry environments (Kappenman & Luck, 2010). Vertical and horizontal eye movement was measured using the electrodes positioned 2 cm below or above the right eye or 1 cm lateral from both eyes, respectively.
Pre-processing
Acquired data was analysed and pre-processed using Python (Python Software Foundation, https://www.python.org/) using custom written scripts (https://github.com/dvanmoorselaar/DvM). and the MNE toolbox (https://www.mne-tools.com). Before epoch selection data was high pass filtered above 0.1 Hz, using a ‘firwin’ filter (Gramfort et al., 2013). Epochs for each trial were selected 200 ms prior to stimulus onset and up to 2000 ms after stimulus onset. An extra 500 ms on both sides of the epoch was included to control for filter artefacts. Automatic-artefact rejection procedure was adapted from Fieldtrip toolbox (Oostenveld et al., 2011). The data was first bandpass filtered from 110 to 140 Hz, afterwards we applied Hilbert transformation and box smoothing. Artefacts detection was set on a 0.05 ms duration and a minimum of 1 artefact per epoch. Automatic-trial rejection procedure was done by transforming the data and calculating the within-subject cut off z-variance. Based upon this individual Z-score variance bad epochs were marked. Additionally, we carried out independent component analysis (ICA) using the Picard method from the MNE toolbox.
ERP & statistical analysis
ERP and statistical analysis were carried out using python 3 in Jupyter notebook
peristimulus period. The epochs were arranged per condition and averaged over condition and participant. Since the ERP components relevant to our research are not equally distributed over the scalp we have included several electrode selections in our analysis. We have included the following electrode groups: frontal (Fp1, AF7, AF3, F1, F3, F5, F7, Fpz, Fp2, AF8, AF4, AFz, Fz, F2, F4, F6 & F8), central (F1, Fz, F2, FC1, FCz, FC2, C1, Cz & C2), centroparietal (CP5, CP3, CP1, P1, P3, P5, P7, P9, Pz, CPz, CP6, CP4, CP2, P2, P4, P6, P8, P10, C1, Cz & C2) and peak electrodes (C4, CP6 & AF7). Peak electrodes were determined using pairwise comparison across all conditions to calculate peak amplitude.
ERP-data was tested using non-parametric permutation cluster using an alpha of 0.05. We have chosen permutation testing since it allows us to test significance at each time-point with few assumptions of the data, while also automatically controlling for family wise error rate (Maris & Oostenveld, 2007). Data was analysed pairwise between each condition per each group of electrodes. Number of permutations was set to 4000.
Results
Figure 1.
ERP plot of frontal electrodes
Shown is the ERP plot for all four condition grand averaged over particpants and trials for the frontal electrode (Fp1, AF7, AF3, F1, F3, F5, F7, Fpz, Fp2, AF8, AF4, AFz, Fz, F2, F4, F6 & F8).
Figure 1 shows the grand average ERP for all conditions in the frontal electrodes. We observed a negative deflection of the pure tones around 100 ms followed by a positive deflection at the 200 ms timepoint. This is in line with our hypothesis that in the linguistic condition the pure tones will deviate already early on. Later in time around 600 ms we observe a greater negative deflection of the
pseudowords compared to the other conditions. This could indicate a late N400. During this same time point we also observe a greater positive deflection of the negative words, as in line with our expected LPC in ethe emotional condition. The patterns observed in the data are in line with our hypothesis, however, after permutation testing none of these patterns turned out to be significant. This suggest that we cannot state a significant difference in the current data at any time point between the conditions in the frontal electrodes.
Figure 2.
ERP plot of central electrodes
Shown is the ERP plot for all 4 condition grand averaged over particpants and trials for the central electrode (F1, Fz, F2, FC1, FCz, FC2, C1, Cz & C2).
Visual inspection of the central electrodes ERP shows a similar early negative peak followed by a positive deflection of the pure tones as in the frontal electrode. Furthermore, around the 300 ms timepoint we observe a second positive peak. This peak was not expected in our hypothesis and could reflect the P300. For the semantic condition we expected the pseudowords to elicit a N400. Within the central electrode we indeed observe a negative deflection of the pseudowords starting around 500 ms. This suggest a pattern that is in line with our hypothesis however, at a later point in time. At last, we expected a larger late positive deflection in the negative words. However, the central electrodes do not show this larger positivity. Within the central electrode permutation testing did not report any
significant results. This suggest that only the sematic and linguistic condition in the central electrode show a pattern in line with our hypothesis, however, we cannot state a significant difference.
Figure 3.
ERP plot of centro-parietal electrodes
Shown is the ERP plot for all 4 condition grand averaged over particpants and trials for the centro-parietal electrode (CP5, CP3, CP1, P1, P3, P5, P7, P9, Pz, CPz, CP6, CP4, CP2, P2, P4, P6, P8, P10, C1, Cz & C2).
When visually inspecting the centroparietal ERP, the first thing that stands out is a generally flattened pattern for all conditions. Alhough we observe diminished positive and negative peaks, the patterns visible in the frontal and central groups electrodes are nonetheless recognizable. The pure tones show the N1-P2 pattern as hypothesized, while also revealing a P300-like pattern. The pseudowords do show the largest negativity from 400 to 600 ms, however, this difference is somewhat negligible. The negative words show a rather more distinguishable positive deflection between the 600 ms and 1000 ms time mark, as in line with our hypothesis for the emotional condition. Again, we did not observe any significant after permutation testing. This suggests no significant difference between conditions in the centroparietal group of electrodes. Although the patterns are less apparent, they are still in line
Figure 4.
ERP plot of peak electrodes
Shown is the ERP plot for all 4 condition grand averaged over particpants and trials for the peak electrode (C4, CP6 & AF7). Significant difference is visualized with a purple line on the x-axis (p- 0.02 between 619 to 775 ms).
Electrodes showing the largest amplitude difference between each condition were identified as peak electrodes. For the pure tone condition, we observe the same N1-P2 pattern as observed in all other electrode groups. Here, we also observe the positive peak at the 300 ms mark, suggesting a P300 component. The semantic condition also seems in line with our hypothesis, where the pseudowords deflect downwards after the 400 ms timepoint. And at last we expected the negative words to show a larger late positivity compared to the neutral words. This is confirmed in the peak electrodes where the negative words are the most positive condition after the 600 ms timepoint. Permutation testing in the peak electrode did report a significant effect between the neutral and pseudo condition. This is effect is observed between 619 to 775 ms, p = 0.02. This suggest a significant larger negative peak of the pseudowords compared to the neutral words between these timepoints. This is somewhat later than
the expected N400 but in line with our expectation. For the other condition we did not observe a significant difference. Thus, we can only state that the patterns are in line with our hypothesis, but we cannot report other significant differences.
Discussion
In this paper, we have proposed the predictive processing theory as a framework to explain the adjusted auditory perception observed in expert meditators. Previous literature has observed long-term meditation to affect brain activity during auditory processing as measured with EEG and fMRI (Biedermann et al., 2016; Fucci et al. 2018). However, it is unclear to what extent and depth
meditation and the different techniques alter auditory processing. To test this, we have designed 3 conditions representing a different level of the hierarchy as reflected by their temporal appearance in the ERP. In order from low to high: linguistic, semantic & emotion. The similarity observed in expert meditators within the ERP observed per condition in turn indicates the depth to which meditation can break down the auditory processing hierarchy. As observed in previous literature, expert meditators demonstrate an increased MMN in oddball paradigms, reflecting an increased prediction error
(Biedermann et al., 2016; Garrido et al., 2009; Srinivasan & Baijal, 2007). Whilst the prior knowledge as seen in a conditioning paradigm is diminished (Kirk & Montagu, 2015). In general, we expect meditation to diminish high level predictions while simultaneously increase low-level predictions.
The current study is in its pilot phase.We have tested all conditions on five participants independent of any meditation condition. We report a significant larger negativity of the pseudowords when compared to the neutral words 200 ms later than the expected N400 timepoint. We did not observe any other significant effects. However, visible inspection of the ERP data shows a unique pattern per word condition. Pure tones seem to deviate already from all other conditions quite early on and display distinctive components,whilet the neutral and negative words show similar patterns only differentiating in their late ERP. Especially in the peak electrodes we observe the different conditions to branch out with respect to their place in the auditory processing hierarchy, pure tones (low in the
hierarchy) differentiate early on whilst both neutral and negative words (high in the hierarchy) differentiate later in time.
The N400
The N400 is an ERP component that reaches a negative peak around 400 ms and has been related to semantic deviance (Ding et al., 2016). An auditory N400 is generally observed the strongest in the parietal and central electrodes (Federmeier et al., 2007). We expected a difference in the N400 between the words and the pseudowords condition as a marker of semantical processing. Our permutation test did observe a significant larger negative effect of the pseudo compared to the negative condition within the peak electrodes from 619 to 775 ms. The peak electrodes observing this effect are located central (C4), centroparietal (CP6) and anterior frontal (AF7), and thus generally fit the expected localization of N400. Therefore, the significant negative peak could be related to the N400 even though its timing is to be questioned.
Since the pseudowords were created by swapping syllables of existing words, the auditory processing of such pseudowords do not reveal their pseudo-semantic nature until at least the second syllable is perceived. Therefore, its uniqueness point, the point where the unique lexical nature of the word is realized, appears later in time. Compared to written text, auditory information takes a longer time to process due to auditory information being smeared over time, whereas written text can be observed and processed at once. From previous literature it has been observed that an auditory N400 can be delayed in time (Praamstra & Stegeman, 1993). It is therefore likely that our observed effect is a delayed N400 as a result of its auditory nature and its low-level properties, syllables, being
unaltered.
Interestingly, the anterior frontal electrode is also included in the significant N400 effect. Furthermore, visual inspection of the centroparietal electrode ERP shows a small N400 like trend. The component is more visible in central electrodes and extends to some frontal electrodes.
How can we explain this aberrant localization of the N400 component observed in the pseudowords condition? Generally, auditory N400 seem to show different localization compared to visual evoked N400, were they tend to be more frontally distributed and last longer (Kutas &
Federmeier, 2011). Another possible explanation is the FN400 – family of the N400. The FN400 reflects a more frontal distributed ERP component related to semantic processing and recognition (Bridger et al., 2012). Although it is normally observed within recognition tests and exclusively related to semantic processes an alternative explanation has been brought forward associating the FN400 to be sensitive to differences of the stimuli’s semantic features in memory– as seen in pseudowords (Voss & Federmeier, 2011; Bermúdez-Margaretto et al., 2015). We must consider that this hypothesis has only been tested with visual stimuli and might not generalize to auditory stimuli. Nonetheless, we can hypothesize that the more frontal distributed N400 component could be related, possibly partly, to the pseudowords different semantic features.
The N1-P2
The pure tones seem to deviate already quite early on compared to the other conditions. Showing a larger negative peak in the first 100 ms, followed by a bigger positive deflection. This pattern is in line with N100 and P200 component, otherwise known as the N1-P2. The N100 is an ERP component that has been associated with sound detection of attention catching properties (Joos et al., 2014). The P200 component is a known follow-up, representing later sensory processing (Lijffijt et al., 2009). The N1-P2 is known to be elicited by pure tones (Okita et al., 1983; Näätänen & Picton, 1987).
Furthermore, the N1 is known to be evoked by abrupt changes of auditory properties such as intensity, pitch or quality (Hyde, 1997). As the pure tones are discriminated from all other conditions due to its lack of semantic or linguistic information, being only a pure frequency wave, they are the deviant condition. Furthermore, the N1 has been known to be affected by the steepness of the sound’s stimulus onset during the first milliseconds (Biermann & Heil, 2000). The pure tones differ in their auditory properties causing a steeper attack curve of the sound – a quicker growth in loudness during the first few milliseconds of the sound. Thus, the N1 observed in our data can also be explained by the vast difference in auditory properties of the pure tones compared to the other conditions.
P300
At last, the pure tones are the only condition to show a third positive deflection around 300 ms. This is especially clearly visible in the peak electrodes plots where we observe two positive slopes embedding a small dip (see figure 4). This trend seems in line with the P300 component found in the ERP literature (Polich, 2007). The P300 component is normally observed in oddball-paradigms to which the odd stimuli elicit a positive deflection around 300 ms. (Katayama & Polich, 1998). As earlier discussed, the pure tones are the deviant condition and could be perceived by participants as the odd stimuli in our presentation set-up. There is a 25% chance to be exposed to pure tones compared to a 75% chance of words stimuli regardless of their semantic or emotional properties. Thus, the positive component visible around 300 ms could be related to the unequal distribution of stimulus property presentation. Corroborating research by Sarang & Telles (2005) did observe an increased peak amplitude of the P300 in an oddball paradigm after meditation. However, they only compared activity before and after meditation.
The observation of the pure tones producing an oddball paradigm is interesting for our research question. Previous literature has observed an increased MMN during meditation in an oddball paradigm which is explained by an increased accuracy to sensory information during
meditation (Biedermann et al., 2016; Srinivasan & Baijal, 2007). Fucci et al. (2018) expanded on this reporting an MMN increase only in the FA condition. If the P300 can be fully explained by being the oddball we can hypothesize to observe an increased P300 only in the FA condition.
The LPC
We hypothesized that comparing the neutral words and negative words conditions would elicit a higher LPC that reflects emotional stimuli processing (Kotz & Paulmann, 2011; Rostami et al., 2016). The LPC has been found to be widely distributed across the scalp (Kanske & Kotz, 2007). Visual inspection of the ERP plots does reveal trends of a higher positive late component of the negative condition, especially in the frontal electrodes. This is in line with the literature that observes negative words to show a more frontal positivity and parieto-occipital negativity (Grass et al., 2016).
The auditory processing hierarchy
The stimuli have been designed to observe the effect of meditation on different levels of the auditory processing hierarchy. Here, low levels of the hierarchy represent stimuli that can be differentiated already early in time and high levels late in time. A visible inspection the ERP results shows the stimuli to differentiate at different time points. This temporal branching, as best observed in the peak electrodes, corresponds with this hierarchy. The pure tones already deviate in their pattern very early on representing the lowest level of the auditory processing hierarchy as reflected by the N1-P2 component. Around the 400 ms mark, and significant at the 600 ms mark, we observe the pseudowords to deflect from a pattern that before this time point was similar to the neutral and negative words. This is in line with our hypothesis suggesting semantic deviance to be reflected by the N400. Furthermore, it suggests that stimuli that differ in their semantical content, but both contain linguistic content showing differential processing later point than pure tones. Lastly, we observe the neutral and negative words show a similar pattern in that they only diverge later in time. This is also in line with our hypothesis that emotional content compared to semantic and linguistic content is processed last, and thus corresponds with the highest level of the hierarchy. The current ERP data shows different stages in which the different conditions are processed and validate our stimuli to be useful for testing how meditation alters auditory processing at different levels in the hierarchy.
Limitation of this study
The current data presents limitations in several dimensions. First and foremost is the low power of the pilot study as we were only able to collect data of five participants before we were restrained to collect more data due to the covid-19 restriction. A low power is a huge bottleneck for statistical analysis and limits the possibility of observing significant effects. Many small effects can be missed due to their small effect size. Furthermore, it also increases the chance of false positives (Ioannidis, 2005). Second, one of our participants included in the pilot data is actively involved in the study, and therefore has an a-priori knowledge of the goal and components (e.g. stimuli) of the study.
Third, we have tried to compensate for the OpenSesame lag but it its actual delays remains arbitrary (with a variance of 8.4 ms) (Bridges et al., 2020). Fourth, the lag to account for the silence at the start of the words is still based on subjective interpretation of two observers and therefore likely inacurate. At last, we were not able to include any meditation condition. Long-term meditation leads to long-term trait effects that adjust brain activity even when one is not meditating (Cahn & Polich, 2006). And on the other side, novice meditators are known to require more attentional processes and show different brain activity while meditating (Lutz, 2008). Therefore, the current observed ERP
components and effects could be different in the main experiment.
Future direction & Points of improvement
Although the current data has methodological and statistical limitations it has proved valuable for insight on our hypothesis and a better construction of the main experiment.
We hypothesize long-term meditation to affect auditory processing as reflected by reducing the influence of priors or increasing the accuracy of sensory information. To research the extent of this modulation in the hierarchy we have designed three conditions that can be compared at a
linguistic, semantic or emotional level. In this order we expect the different conditions to represent the auditory processing hierarchy. Based on the results of this pilot data we do observe the conditions to show this temporal order of processing. The chosen conditions therefore are suitable for researching the effect of meditation at these different levels of the hierarchy.
Furthermore, we have hypothesized to observe an N1-P2 effect in pure tones, N400 in pseudowords and an LPC in the negative words. Despite having only observed a significant difference in the semantic condition representing the N400, we do observe an N1-P2 and LPC in the right conditions. This could indicate the possibility of confirming our hypothesis when not limited by the current power.
Besides the expected components we also observe a P300 component in the pure tones. Its appearance could be explained by the emergence of an oddball paradigm by cause of the pure tones deviant properties. This is interesting to include as a hypothesis for the main experiment. As previous
literature in meditation using an oddball paradigm has focussed on the MMN, it is interesting to see if similar results are observed in the P300 component.
The current pilot study has exposed two limitations in our stimulus design. We have discovered that the program used for auditory presentation is subjected to lag (Bridges et al., 2020) and some of our stimuli included a silence before the start of the word. We have tried to compensate for these constraints before pre-processing. It is recommended for the main experiment to remove these limitations before testing. To account for the silence at the start of each word we recommend Praat to analyse the intensity of each stimuli separately and remove the silence/quiet part at the starting of a word given a certain threshold. As for the OpenSesame lag this requires a somewhat more innovative solution. Two options are available here. At first, the stimulus data can be exported on an external hard drive before presentation and thereby removing the OpenSesame lag that is observed while logging the data. Another option would be to move to a more suitable auditory presentation program such as Matlab (htwww.mathworks.com/products/matlab).
Conclusion
To conclude, in our pilot data we have observed a single significant effect between the pseudo and negative words that seem to reflect an auditory N400 component. Furthermore, we observe components in our ERP data that could indicate different early processing of the pure tones, N1-P2, P300, and late processing of the emotional stimuli, the LPC. From this data we conclude that the temporal order in which the condition seem to differentiate are suitable to reflect different levels in the auditory processing hierarchy and can be used for the main experiment.
References
Aggelopoulos, N. C. (2015). Perceptual inference. Neuroscience & Biobehavioral Reviews, 55,
375-392.
Bermúdez-Margaretto, B., Beltrán, D., Domínguez, A., & Cuetos, F. (2015). Repeated exposure to
“meaningless” pseudowords modulates LPC, but not N (FN) 400. Brain topography, 28(6), 838-851.
Biedermann, B., De Lissa, P., Mahajan, Y., Polito, V., Badcock, N., Connors, M. H., ... & McArthur,
G. (2016). Meditation and auditory attention: An ERP study of meditators and
non-meditators. International Journal of Psychophysiology, 109, 63-70.
Biermann, S., & Heil, P. (2000). Parallels between timing of onset responses of single neurons in cat
and of evoked magnetic fields in human auditory cortex. Journal of neurophysiology, 84(5),
2426-2439.
Boersma, Paul & Weenink, David (2020). Praat: doing phonetics by computer [Computer program]. Version 6.1.09, retrieved 26 January 2020 from http://www.praat.org/
Bridger, E. K., Bader, R., Kriukova, O., Unger, K., & Mecklinger, A. (2012). The FN400 is
functionally distinct from the N400. Neuroimage, 63(3), 1334-1342.
Bridges, D., Pitiot, A., MacAskill, M. R., & Peirce, J. (2020). The timing mega-study: comparing a
range of experiment generators, both lab-based and online.
Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive
science. Behavioral and brain sciences, 36(3), 181-204.
Cahn, B. R., & Polich, J. (2006). Meditation states and traits: EEG, ERP, and neuroimaging
studies. Psychological bulletin, 132(2), 180.
Curran, T., Tucker, D. M., Kutas, M., & Posner, M. I. (1993). Topography of the N400: Brain
electrical activity reflecting semantic expectancy. Electroencephalography and Clinical
Neurophysiology/Evoked Potentials Section, 88(3), 188-209.
Dahl, C. J., Lutz, A., & Davidson, R. J. (2015). Reconstructing and deconstructing the self: cognitive
Ding, N., Melloni, L., Zhang, H., Tian, X., & Poeppel, D. (2016). Cortical tracking of hierarchical
linguistic structures in connected speech. Nature neuroscience, 19(1), 158.
Duyck, W., Desmet, T., Verbeke, L., & Brysbaert, M. (2004). WordGen: A Tool for Word Selection and Non-Word Generation in Dutch, German, English, and French. Behavior Research Methods, Instruments & Computers, 36(3), 488-499.
Federmeier, K. D., Wlotko, E. W., De Ochoa-Dewald, E., & Kutas, M. (2007). Multiple effects of
sentential constraint on word processing. Brain research, 1146, 75-84.Friston, K. J., &
Stephan, K. E. (2007). Free-energy and the brain. Synthese, 159(3), 417-458.
Friston, K. (2010). The free-energy principle: a unified brain theory? Nature reviews
neuroscience, 11(2), 127-138.
Frishkoff, G. A., Tucker, D. M., Davey, C., & Scherg, M. (2004). Frontal and posterior sources of
event-related potentials in semantic comprehension. Cognitive Brain Research, 20(3),
329-354.
Fucci, E., Abdoun, O., Caclin, A., Francis, A., Dunne, J. D., Ricard, M., ... & Lutz, A. (2018).
Differential effects of non-dual and focused attention meditations on the formation of
automatic perceptual habits in expert practitioners. Neuropsychologia, 119, 92-100.
Garrido, M. I., Kilner, J. M., Stephan, K. E., & Friston, K. J. (2009). The mismatch negativity: a
review of underlying mechanisms. Clinical neurophysiology, 120(3), 453-463.
Gramfort, A., Luessi, M., Larson, E., Engemann, D. A., Strohmeier, D., Brodbeck, C., ... &
Hämäläinen, M. (2013). MEG and EEG data analysis with MNE-Python. Frontiers in
neuroscience, 7, 267.
Grass, A., Bayer, M., & Schacht, A. (2016). Electrophysiological correlates of emotional content and
volume level in spoken word processing. Frontiers in human neuroscience, 10, 326.
Hyde, M. (1997). The N1 response and its applications. Audiology and Neurotology, 2(5), 281-307.
Ioannidis, J. P. (2005). Why most published research findings are false. PLos med, 2(8), e124. Johnson, B. W., & Hamm, J. P. (2000). High-density mapping in an N400 paradigm: evidence for
Joos, K., Gilles, A., Van de Heyning, P., De Ridder, D., & Vanneste, S. (2014). From sensation to
percept: the neural signature of auditory event-related potentials. Neuroscience &
Biobehavioral Reviews, 42, 148-156.
Kanske, P., & Kotz, S. A. (2007). Concreteness in emotional words: ERP evidence from a hemifield
study. Brain research, 1148, 138-148.
Kappenman, E. S., & Luck, S. J. (2010). The effects of electrode impedance on data quality and
statistical significance in ERP recordings. Psychophysiology, 47(5), 888-904.
Katayama, J. I., & Polich, J. (1998). Stimulus context determines P3a and P3b. Psychophysiology,
35(1), 23-33.
Kiehl, K. A., Hare, R. D., McDonald, J. J., & Brink, J. (1999). Semantic and affective processing in
psychopaths: An event-related potential (ERP) study. Psychophysiology, 36(6), 765-774.
Kirk, U., & Montague, P. R. (2015). Mindfulness meditation modulates reward prediction errors in a
passive conditioning task. Frontiers in Psychology, 6, 90.
Kotz, S. A., & Paulmann, S. (2011). Emotion, language, and the brain. Language and Linguistics
Compass, 5(3), 108-125.
Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: finding meaning in the N400
component of the event-related brain potential (ERP). Annual review of psychology, 62,
621-647.
Leinonen, A., Grönholm-Nyman, P., Järvenpää, M., Söderholm, C., Lappi, O., Laine, M., & Krause,
C. M. (2009). Neurocognitive processing of auditorily and visually presented inflected words
and pseudowords: evidence from a morphologically rich language. Brain research, 1275,
54-66.
Lijffijt, M., Lane, S. D., Meier, S. L., Boutros, N. N., Burroughs, S., Steinberg, J. L., ... & Swann, A.
C. (2009). P50, N100, and P200 sensory gating: relationships with behavioral inhibition,
attention, and working memory. Psychophysiology, 46(5), 1059-1068.
Lippelt, D. P., Hommel, B., & Colzato, L. S. (2014). Focused attention, open monitoring and loving
Lutz, A., Slagter, H. A., Dunne, J. D., & Davidson, R. J. (2008). Attention regulation and monitoring
in meditation. Trends in cognitive sciences, 12(4), 163-169.
Lutz, A., Jha, A. P., Dunne, J. D., & Saron, C. D. (2015). Investigating the phenomenological matrix
of mindfulness-related practices from a neurocognitive perspective. American
Psychologist, 70(7), 632.
Marian, V., Bartolotti, J., Chabal, S., & Shook, A. (2012). CLEARPOND: Cross-linguistic
easy-access resource for phonological and orthographic neighborhood densities. PloS one, 7(8).
Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing of EEG-and MEG-data. Journal
of neuroscience methods, 164(1), 177-190.
Mathôt, S., Schreij, D., & Theeuwes, J. (2012). OpenSesame: An open-source, graphical experiment
builder for the social sciences. Behavior research methods, 44(2), 314-324.
May, A., & Gaser, C. (2006). Magnetic resonance-based morphometry: a window into structural
plasticity of the brain. Current opinion in neurology, 19(4), 407-411.
Moors, A., De Houwer, J., Hermans, D., Wanmaker, S., Van Schie, K., Van Harmelen, A. L., ... &
Brysbaert, M. (2013). Norms of valence, arousal, dominance, and age of acquisition for 4,300
Dutch words. Behavior research methods, 45(1), 169-177.
Näätänen, R., & Picton, T. (1987). The N1 wave of the human electric and magnetic response to
sound: a review and an analysis of the component structure. Psychophysiology, 24(4),
375-425.
Okita, T., Konishi, K., & Inamori, R. (1983). Attention-related negative brain potential for speech
words and pure tones. Biological Psychology, 16(1-2), 29-47.
Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J. M. (2011). FieldTrip: open source software for
advanced analysis of MEG, EEG, and invasive electrophysiological data. Computational
intelligence and neuroscience, 2011.
Pagnoni, G. (2019). The contemplative exercise through the lenses of predictive processing: a promising approach. In Progress in brain research (Vol. 244, pp. 299-322). Elsevier.
Praamstra, P., & Stegeman, D. F. (1993). Phonological effects on the auditory N400 event-related
brain potential. Cognitive Brain Research, 1(2), 73-86.
Rostami, H. N., Ouyang, G., Bayer, M., Schacht, A., Zhou, C., & Sommer, W. (2016). Dissociating
the influence of affective word content and cognitive processing demands on the late positive
potential. Brain topography, 29(1), 82-93.
Sarang, S. P., & Telles, S. (2006). Changes in P300 following two yoga-based relaxation
techniques. International Journal of Neuroscience, 116(12), 1419-1430.
Srinivasan, N., & Baijal, S. (2007). Concentrative meditation enhances preattentive processing: a
mismatch negativity study. Neuroreport, 18(16), 1709-1712.
Teper, R., Segal, Z. V., & Inzlicht, M. (2013). Inside the mindful mind: How mindfulness enhances
emotion regulation through improvements in executive control. Current Directions in
Psychological Science, 22(6), 449-454.
Van Dam, N. T., van Vugt, M. K., Vago, D. R., Schmalzl, L., Saron, C. D., Olendzki, A., ... & Fox,
K. C. (2018). Mind the hype: A critical evaluation and prescriptive agenda for research on
mindfulness and meditation. Perspectives on psychological science, 13(1), 36-61.
Voss, J. L., & Federmeier, K. D. (2011). FN400 potentials are functionally identical to N400
potentials and reflect semantic processing during recognition
testing. Psychophysiology, 48(4), 532-546.
Wenk-Sormaz, H. (2005). Meditation can reduce habitual responding. Alternative therapies in health
and medicine, 11(2), 42-59.
Winkler, I. (2007). Interpreting the mismatch negativity. Journal of Psychophysiology, 21(3-4),
147-163.
Yon, D., de Lange, F. P., & Press, C. (2019). The predictive brain as a stubborn scientist. Trends in
Appendix
Table 1.
Statistical data on different properties per word condition
Negative Neutral Combined Pseudo Pure
M SD M SD M SD M SD M SD Pitch 172.95 8.31 176.73 13.66 178.18 13.18 174.31 25.11 Intensity 75.46 1.45 76.30 1.17 76.44 1.28 76.32 1.14 Length (sec) 1.20 0.10 1.15 0.11 1.15 0.12 1.11 0.11 Valence 1.66 0.21 4.12 0.27 Arousal 5.31 0.43 4.08 0.68 Power 4.68 0.74 4.13 0.30 AoA 7.88 1.37 7.03 1.50 Freq log10 1.15 0.55 0.75 0.56 Freq /million 36.77 62.73 13.57 21.45 Length 6.67 1.07 6.20 0.75 N% 0.12 0.20 0.69 1.80 PTAN 5.27 7.78 10.33 5.51 PTAF 10.33 20.93 19.30 27.40 Neighbourhood density 2.90 3.46 1.20 1.72 Bigram frequency 56624.33 26014.25 48130.73 22517.29
Showing mean (M) and standard deviation (SD) of all condition (and words combined) per each controlled word parameter.
Table 2.
Statistical data on t-test of different properties between condition
Negative vs Neutral Words vs pseudo words Words vs Pure Tones
t-value DF p-value t-value DF p-value t-value DF p-value Pitch 0.88 28 0.39 0.81 25 0.42 0.08 17 0.94 Intensity 1.66 28 0.11 1.29 30 0.21 1.08 33 0.29 Length (sec) 1.37 28 0.18 0.66 25 0.51 1.73 29 0.09 Valence 26.77 28 8.58E-22* Arousal 5.71 28 2.02E-06* Power 2.63 28 0.01* AoA 1.56 28 0.13 Freq log10 1.92 28 0.07 Freq /million 1.31 28 0.20 Length 1.33 28 0.19 N% 1.17 28 0.25 PTAN 0.60 28 0.55 PTAF 0.97 28 0.34 Neighborhood density 2.15 43 0.04* Bigram frequency 1.10 32 0.28
Showing t-test values over all compared word conditions with an alpha of 0.05. Asterix shows significant p-values.
