AN ELECTROPHYSIOLOGICAL EXAMINATION OF THE EFFECTS OF
SCHIZOTYPAL TRAITS AND CANNABIS USAGE ON NEURAL MARKERS OF
PREDICTIVE PROCESSING
by
Matthew Sargent
BHSc (Hons), University of Northern British Columbia, 2019

THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE
IN
PSYCHOLOGY

UNIVERISTY OF NORTHERN BRITISH COLUMBIA
July 2023
© Matthew Sargent, 2023

ii
Abstract
The brain’s ability to make predictions has long been of interest to cognitive
neuroscientists, who have used techniques including electroencephalography (EEG) to study this
phenomenon. A component of neurophysiological activity, termed the Semantic Predictive
Potential (SPP), may be an indicator of people’s ability to predict the nature of upcoming
semantic content. Interestingly, a body of theoretical and experimental work suggests that people
with schizophrenia and people who regularly use cannabis have difficulty making predictions
about information in their environments. We explored how schizophrenia-like traits and cannabis
use affected the SPP. Participants read sentences which differed in the predictability of their
endings while brain activity was monitored using EEG. Participants then completed
questionnaires assessing their levels of schizophrenia-like traits and cannabis use. We replicate
previous findings suggesting that the SPP is sensitive to semantic predictability, and show that
schizophrenia-like traits and cannabis use interact with semantic predictability to influence the
SPP.

iii
TABLE OF CONTENTS
Abstract ........................................................................................................................................... ii
Table of Contents ........................................................................................................................... iii
List of Tables ...................................................................................................................................v
List of Figures ................................................................................................................................ vi
Acknowledgements ...................................................................................................................... viii
Introduction ......................................................................................................................................1
Chapter One .....................................................................................................................................1
1.1 Schizophrenia .........................................................................................................................1
1.2 The Neuroscience of Schizophrenia .......................................................................................3
1.3 Schizotypy ..............................................................................................................................4
Chapter Two.....................................................................................................................................5
2.1 Cannabis .................................................................................................................................5
2.2 Cannabis and Schizophrenia ..................................................................................................8
Chapter Three...................................................................................................................................9
3.1 Memory ..................................................................................................................................9
3.2 Schizophrenia and Memory..................................................................................................10
3.3 Cannabis and Memory .........................................................................................................11
3.4 Spreading Activation ............................................................................................................12
3.5 Spreading Activation in Thought-Disordered Schizophrenia ..............................................14
3.6 Spreading Activation and Cannabis .....................................................................................15
Chapter Four ..................................................................................................................................16
4.1 Top-Down Inference and Predictive Processing ..................................................................16
4.2 Schizophrenia, Cannabis, and Predictive Processing ...........................................................20
Chapter Five ...................................................................................................................................27
5.1 EEG and Semantic Prediction ..............................................................................................27
5.2 The N400 and the Prediction vs Integration Debate ............................................................28
5.3 Schizophrenia and Semantic Prediction ...............................................................................33
5.4 N400 in Schizophrenia .........................................................................................................35
5.5 Cannabis and Semantic Prediction .......................................................................................37
Chapter Six.....................................................................................................................................39
6.1 Semantic Predictive Potential...............................................................................................39

iv
Chapter Seven ................................................................................................................................42
7.1 Present Study and Experimental Hypotheses .......................................................................43
7.2 Participants ...........................................................................................................................45
7.3 Materials ...............................................................................................................................45
7.4 Task Sentences .....................................................................................................................50
7.5 Design ...................................................................................................................................57
7.6 Sentence-Reading Task Procedure .......................................................................................60
7.7 Experimental Procedure .......................................................................................................61
7.8 EEG Recording and Preprocessing ......................................................................................62
7.9 Analysis Strategy ..................................................................................................................63
Chapter Eight .................................................................................................................................67
8.1 Memory Recognition Task Results ......................................................................................67
8.2 Semantic Predictive Potential Waveforms and Scalp Maps ................................................68
8.3 Semantic Predictive Potential Analysis................................................................................74
8.4 N400 Waveforms and Scalp Maps .......................................................................................85
8.5 N400 Analysis ......................................................................................................................90
Chapter Nine ..................................................................................................................................95
9.1 General Discussion ...............................................................................................................95
9.2 Replications of Previous Findings......................................................................................102
9.3 Limitations .........................................................................................................................103
Conclusion ...................................................................................................................................105
References ....................................................................................................................................107
Appendix A: Study Questionnaires .............................................................................................127
Appendix B: Tables of Means and Post-Hoc Comparisons.........................................................137

v
List of Tables
Table 1. Means and Standard Deviations of Potential Confounding Variables of Target Words. 55
Table 2. P-values From Independent Samples t-Tests Demonstrating Control for Potential
Confounding Variables of Target Words .......................................................................................57
Table 3. Examples of Different Types of Sentences Used in the Experiment. ..............................58
Table 4. Correlation Matrix Showing r Values Between Higher-Order Factors of the SPQ-BRU
and Cannabis Frequency (Taken From the DFAQ-CU) in Our Sample. .......................................68
Table 5. Results From LME Analysis on SPP Amplitude 200-0 ms Preceding the Presentation of
Target Words. ................................................................................................................................75
Table 6. Results From LME Analysis on N400 Amplitude 300-500 ms Following the
Presentation of Target Words. .......................................................................................................91
Table 7. Mean SPP Amplitudes for Each Combination of Fixed Effects 200-0 ms Prior to the
Presentation of Final Words in the Sentence-Reading Task ........................................................138
Table 8. Comparisons of SPP Amplitudes for High Constraint, Low Constraint, and NonSemantic Sentences Across Levels of High and Low Disorganized Thought and Cannabis Use
......................................................................................................................................................140
Table 9. Mean N400 Amplitudes for Each Combination of Fixed Effects 300-500 ms After the
Presentation of Final Words in the Sentence-Reading Task ........................................................141
Table 10. Post-Hoc Comparisons For Congruency x Constraint Interaction of N400 Amplitudes
at Electrode Pz. ............................................................................................................................141
Table 11. Post-Hoc Comparisons For Congruency x Cannabis Frequency Interaction of N400
Amplitudes at Electrode Pz. .........................................................................................................142

vi
List of Figures
Figure 1. Depiction of the Spreading Activation Model. Adapted from Collins & Loftus (1975)
........................................................................................................................................................13
Figure 2. Ambiguous Rat-Man Figure. Adapted from Bugelski & Alampay (1961) ....................17
Figure 3. Images Showing the Progression of the Hollow-Mask Illusion Using a Rotating Hollow
Mask. ..............................................................................................................................................21
Figure 4. Depiction of Differences in the N400 Potential in Response to Congruent and
Incongruent Target Words. Reprinted with permission from León-Cabrera et al. (2019) ............29
Figure 5. Depiction of Differences in the Semantic Predictive Potential Across High Constraint,
Low Constraint, and Non-Semantic Sentences in a Sentence-Reading Task. Reprinted with
permission from León-Cabrera et al. (2019)..................................................................................40
Figure 6. Schematic of an Individual Trial From the Sentence-Reading Task. Adapted From
León-Cabrera et al. (2019) .............................................................................................................59
Figure 7. Topographic Plot Showing Electrode Positions Used in the Experiment. .....................60
Figure 8. Histogram of Frequency of Cannabis Use......................................................................65
Figure 9. Histogram of Frequency of Disorganized Thought. .......................................................66
Figure 10. Semantic Predictive Potential Waveforms Observed Over Electrode F3. ...................70
Figure 11. Epoch Visualizing the Semantic Predictive Potential Across all Electrodes 1100 ms
Preceding the Presentation of the Final Words in Sentences of High, Low, and Non-Semantic
Constraint. ......................................................................................................................................71
Figure 12. Scalp Maps Showing SPP Scalp Topographies............................................................73
Figure 13. Visualization of the Interaction Between Constraint and Disorganized Thought on
SPP Amplitude. ..............................................................................................................................77
Figure 14. Visualization of the Interaction between Constraint, Cannabis Frequency, and
Disorganized Thought on SPP Amplitude .....................................................................................79
Figure 15. Visualization of the Interaction Between Hemisphere and Disorganized Thought on
SPP Amplitude. ..............................................................................................................................81
Figure 16. Visualization of the Interaction Between Hemisphere, Cannabis Frequency, and
Disorganized Thought on SPP Amplitude. ....................................................................................83
Figure 17. N400 Waveforms Observed Over Electrode Pz. ..........................................................86
Figure 18. Epoch Visualizing the N400 Potential Across all Electrodes. .....................................87

vii
Figure 19. Scalp Maps Showing N400 Scalp Topographies. ........................................................89
Figure 20. Visualization of Constraint x Congruency Interaction for N400 Amplitudes Over
Electrode Pz. ..................................................................................................................................92
Figure 21. Visualization of Congruency x Cannabis Frequency Interaction for N400 Amplitudes
Over Electrode Pz. .........................................................................................................................94

viii
Acknowledgements
Science is a collaborative effort, and this thesis would not be possible without the
contributions from my valued supervisory committee, colleagues, and family. Thus, I want to
acknowledge the people who have been most instrumental in my academic journey.
First, I want to extend my sincerest thanks to my supervisor Dr. Heath Matheson. His
expertise and support in multiple domains proved to be invaluable for my study. More
importantly, I want to thank Heath for his continued demonstration of what it means to be an
effective scientist, teacher, and critical thinker. The example Heath sets is not lost on me, and I
am fortunate to have been able to work with him on this project.
Second, I wish to thank Dr. Paul Siakaluk, Dr. Annie Duchesne, and Dr. Andrea
Gingerich for their support and feedback throughout my study. Their wide breadth of experience
gave me an important range of perspectives to consider when planning my methodology and
analysis. I also want to extend a special thanks to Annie for providing the Department of
Psychology with new EEG caps, which ensured I could collect the best possible data for my
project.
Third, I want to thank my close friend and colleague Susie MacRae. Her advice,
generosity, and encouragement have been highly motivating for me throughout my degree. With
a small MSc Psychology cohort at UNBC, I am especially appreciative of the stimulating
discussions and collaborations I have been able to participate in with Susie.
Finally, I wish to thank my parents for their continued support throughout my academic
journey. My Mother’s insistence on my academic excellence from a young age has certainly
contributed to my ability to complete this thesis. I am eternally grateful for my parents’ concern
for my well-being and interest in my academic pursuits.

1
An Electrophysiological Examination of the Effects of Schizotypal Traits and Cannabis
Usage on Neural Markers of Predictive Processing
In this thesis, I investigated the ways in which schizotypal traits and cannabis usage
influence neural markers of semantic prediction. I begin by providing background information
on schizophrenia and cannabis use, and discuss some ways in which schizophrenia and cannabis
use may affect the brain in similar ways. I then present the predictive processing model of the
brain, and discuss a body of neurophysiological evidence which suggests that people with
schizophrenia may abnormally predict the nature of stimuli in their environments. Following
this, I introduce recent neurophysiological research in predictive processing which describes a
novel component of brain activity thought to be related to people’s ability to predict semantic
content.
In my study, participants completed a task adapted from recent literature, which allowed
me to investigate neurophysiological components of predictive processing. I collected
information on participant’s levels of schizophrenia-like thought and cannabis usage, and
explored the ways these factors influenced the neural correlates of predictive processing. I show
that schizophrenia and cannabis use interact in complex ways to influence brain activity related
to predictive processing.
Chapter One
1.1 Schizophrenia
Schizophrenia is a multifaceted mental health disorder which affects thoughts,
motivation, affect, and cognitive abilities (Patel et al., 2014). The disorder can be debilitating for
patients, and difficult for their families, due to the characteristic symptoms experienced by
people with schizophrenia. Many symptoms of schizophrenia are classically categorized as either

2
positive or negative. Positive symptoms in schizophrenia are commonly conceptualized as
symptoms which are “added to” a person’s subjective experience, including hallucinations,
delusions, and disorganized thinking (Kay et al., 1989). Hallucinations refer to false perceptions,
while delusions refer to false thoughts or beliefs (Fletcher & Frith, 2009). Negative symptoms of
schizophrenia, conversely, refer to symptoms which are “subtracted from” a person’s subjective
experience, and may include diminished affect and motivation as well as deficits in cognitive
function. Schizophrenia is also associated with premature mortality; it is estimated that people
with schizophrenia die 25 years prematurely compared to the general population, often due to
comorbidities including heart disease (Kilbourne et al., 2009).
It is estimated that the prevalence of schizophrenia in the Canadian population is ~1%
(Canadian Mental Health Association (CMHA), 2014). The British Columbia Schizophrenia
Society estimates that the prevalence in British Columbia is consistent with the 1% estimate
provided by CMHA (British Columbia Schizophrenia Society, 2022). In Northern British
Columbia, the Prince George Branch of the British Columbia Schizophrenia Society serves the
community and surrounding regions by providing support and resources to people with
schizophrenia.
The etiology of schizophrenia suggests a strong role of genetic heritability in
predisposition to the disorder (Patel et al., 2014). Indeed, in second-degree relatives of someone
with schizophrenia, prevalence rises to 4-6%, while in identical twins, prevalence rises to 46%
(McDonald et al., 2003). Environmental and social factors, including social isolation and
childhood trauma, also appear to have a role in the onset of schizophrenia (Patel et al., 2014). It
has been further suggested that exposure to psychoactive drugs, including cannabis, may
exacerbate rates of schizophrenia in at-risk populations (Patel et al., 2020). Symptoms of

3
schizophrenia tend to become present in males when they reach their early 20s, and in females in
their late 20s or early 30s, although the prevalence of the disorder appears to be relatively equal
in males and females (Patel et al., 2014).
People with schizophrenia experience a wide range of cognitive deficits (Bowie &
Harvey, 2006). A meta-analysis found that the intelligence (IQ) of people with schizophrenia
was approximately half a standard deviation lower than the IQ of people in the general
population (d = -.54) (Woodberry et al., 2008). People with schizophrenia also display
impairments in executive function, short- and long-term memory, language functioning, and
attention (Fioravanti et al., 2012). In summary, schizophrenia is associated with canonical
positive and negative symptoms, increased morbidity, and widespread deficits to cognitive
function.
Studying the brain dynamics and anatomy of people with schizophrenia may allow
researchers to better understand the atypical neural patterns of people with schizophrenia.
Neuroscientists have made important discoveries about how brain structure and dynamics vary
with schizophrenia, although more work remains. Understanding how the brain operates
differently in people with schizophrenia compared to neurotypical people is important for
guiding clinical research on schizophrenia and informing public practice.
1.2 The Neuroscience of Schizophrenia
The brains of people with schizophrenia often show widespread structural deficits in gray
and white matter, in addition to enlarged ventricles (Birur et al., 2017). Further, functional
connectivity analyses, which are used in determining which parts of the brain are communicating
and coupling during cognitive performance, have revealed important alterations in brain
dynamics in schizophrenia (Rogers et al., 2007). In particular, a meta-analysis of resting-state

4
functional connectivity analyses, which involve assessing functional connectivity while
participants are not engaged in cognitively demanding tasks, have revealed disrupted
connectivity patterns associated with schizophrenia. These include decreased within-network
connectivity in the default mode network (involved in self-projection and episodic memory),
somatomotor network (involved in sensorimotor function), core network (involved in task
control), auditory network (involved in auditory system) and self-referential network (involved
in self-referential processing) (Li et al., 2019; Liao et al., 2010). Importantly, within-network
hyperconnectivity was not found in any observed networks in schizophrenia patients compared
to controls, suggesting that within-network hypoconnectivity may be related to the cognitive
deficits experienced by people with schizophrenia (Li et al., 2019). Indeed, diminished withinnetwork resting-state functional connectivity in parts of the default mode network has been
observed in patients who have undergone a first episode of psychosis but are without a
psychiatric disorder diagnosis and have experienced the presence of positive and negative
symptoms for less than one year (Alonso-Solís et al., 2013). This suggests that functional
connectivity abnormalities underlying schizophrenia may manifest in earlier stages of the
disorder. In sum, people with schizophrenia experience a wide range of cognitive deficits, in
addition to the positive and negative symptoms characteristic of the disorder, and show brain
abnormalities in anatomy and functional connectivity compared to healthy controls.
1.3 Schizotypy
Although only a subset of the population is diagnosed with schizophrenia, there have
been efforts to quantify the extent to which neurotypical people without a schizophrenia
diagnosis experience schizophrenia-like symptoms (Cohen et al., 2010; Davidson et al., 2016;
Raine, 1991). Schizotypy has been described as a ‘personality organization’ which can manifest

5
in select phenotypic characteristics of schizophrenia, or cause people to have a predisposition to
schizophrenia (Kotov et al., 2018; Meehl, 1990; Meehl, 1962). Attempts have been made to
quantify the extent to which healthy people display schizotypal traits, including constricted
affect, odd speech, suspiciousness, and unusual perceptions (Cohen et al., 2010; Raine, 1991).
The Schizotypal Personality Questionnaire (SPQ) and Schizotypal Personality Questionnaire
Brief Revised (SPQ-BR) have been used to measure the extent to which people display
schizotypal personality traits (Cohen et al., 2010; Raine, 1991). Importantly, it has been
demonstrated that meaningful between-subject variation exists on the SPQ-BR, both in
populations of people with schizophrenia and in healthy populations. Davidson et al. (2016)
surveyed a large population of healthy undergraduates (N = 2552) and found that schizotypal
traits were reported somewhat commonly among the sample. For example, 25.1% of the
respondents endorsed the statement “I am an odd, unusual person”, while 17.7% endorsed “Are
your thoughts sometimes so strong that you can almost hear them?”. The SPQ has been shown to
have high construct validity, and has been used to measure schizotypy in many studies (the
original paper has over 2000 citations) (Raine, 1991). It has been found that the SPQ-BR, which
is a shorter revised version of the SPQ, takes less time to administer and results in less
incomplete responses than the SPQ (Cohen et al., 2010). In sum, schizotypy appears to be a trait
which varies among healthy people, and can be quantified using measures such as the SPQ and
SPQ-BR.
Chapter Two
2.1 Cannabis
Since recreational cannabis use was legalized in Canada in 2018, there has been a surge
in cannabis consumption among Canadians. In Canada, the percentage of individuals who

6
reported using cannabis at least once in the last three months increased by almost 2% from 2018
(pre legalization) to 2019 (post legalization) (Statistics Canada, 2020). Young Canadians are the
most prevalent users of cannabis, with 33.3% of 18–24-year-olds reporting having used cannabis
at least once in the last three months in 2019, compared to 16.8% of all Canadians aged 15 and
older. In addition, legal cannabis retailers in British Columbia generated almost seven times as
much revenue in June 2020 than in June 2019, further suggesting that cannabis consumption is
on the rise (Statistics Canada, 2021). Cannabis use in Canada is increasing, and it is crucial that
we understand how the substance affects every aspect of our cognition and health.
Cannabis is an interesting topic of research for cognitive psychologists and
neuroscientists. Cannabis contains psychoactive ingredients, meaning it has the potential to cause
changes in thinking, emotion, perception, cognition, and behavior; areas of interest to cognitive
psychologists (Atakan, 2012; Beirness & Porath-Waller, 2015). Further, cannabis is known to
cause changes to brain structure and activity, which makes it a compelling topic of study for
cognitive neuroscientists (Beirness & Porath-Waller, 2015; Skosnik et al., 2012). Subjective
effects of cannabis can include relaxation, improved mood, increased appetite, heightened
appreciation of art, and increased talkativeness, as well as delusions and feelings of paranoia,
guilt, anxiety, and depression (Iversen, 2019; Zeiger et al., 2010).
Although cannabis is made up of over 400 compounds, it is thought that Δ-9Tetrahydrocannabinol, or Δ-9-THC, is the main psychoactive ingredient (Beirness & PorathWaller, 2015). Δ-9-THC is believed to be responsible for many of the subjective and
physiological effects of the drug, including the “high” experienced during acute cannabis
intoxication (Atakan, 2012; Beirness & Porath-Waller, 2015). Acute cannabis intoxication refers
to the period shortly after ingesting cannabis, in which the subjective effects are being

7
experienced. Δ-9-THC acts on cannabinoid receptors which are found in the brain as part of the
brain’s endocannabinoid system; in particular, the CB1 cannabinoid receptor is a primary
binding site for Δ-9-THC molecules. CB1 receptors are expressed in multiple areas of the brain,
including the frontal cortex, basal ganglia, hippocampus, and cerebellum (Boggs et al., 2016).
It has been suggested that acute cannabis intoxication may inhibit motor coordination,
inhibit memory encoding, cause impairments in executive functioning, and alter critical thinking
(Crean et al., 2011; Doss et al., 2018; Ramaekers et al., 2006). Additionally, research has studied
the effects of acute cannabis intoxication on attention, and found that Δ-9-THC may actually
improve performance on sustained attention tasks as well as divided attention tasks (Haney et al.,
1999; Hart et al., 2001). Lane et al. (2005) found that acute Δ-9-THC intoxication caused risky
decision making. It has also been suggested that cannabis may impair creativity; specifically, it
has been found that acute Δ-9-THC intoxication hinders performance on divergent thinking
tasks, rendering participants less able to devise novel uses for objects (Kowal et al., 2015).
It is important to note that there are similarities and differences between long-term
cannabis use and acute cannabis intoxication with respect to their effects on cognition. Regular
cannabis use can cause residual effects on cognition even after the subjective effects of the drug
have subsided (Beirness & Porath-Waller, 2015). It has been suggested that regular cannabis use
may have implications for learning, memory, cognitive flexibility, and other executive functions
(Beirness & Porath-Waller, 2015). Notably, memory and executive functioning are also impaired
during acute cannabis intoxication, as mentioned previously, suggesting a degree of similarity
between acute cannabis intoxication and chronic cannabis use. Some authors (Crean et al., 2011;
Beirness & Porath-Waller, 2015) have argued that many of the impairments to executive
functions caused from cannabis use may be attenuated after approximately one month of

8
abstinence from cannabis. Crean et al. (2011) elaborate by suggesting that some cognitive
deficits, in domains such as working memory and attention, may be negligible following
sustained cannabis abstinence, but deficits to verbal fluency and decision making may persist
even after long periods of abstinence from the substance.
2.2 Cannabis and Schizophrenia
Scientists have long been interested in the association between schizophrenia and
cannabis use (Pearson & Berry, 2019). Both chronic schizophrenia and cannabis intoxication
have been associated with paranoia, delusional thinking, auditory and visual hallucinations, and
reduced motivation (Fletcher & Frith, 2009; Lawn et al., 2016; Raine, 1991; Wainberg et al.,
2021; Zeiger et al., 2010). Further, people with schizophrenia and people under the influence of
cannabis exhibit similar performance on some cognitive tasks, such as reduced susceptibility to
the binocular depth inversion illusion (an illusion in which inverted percepts often appear
normal) and differential spreading activation in lexical decision tasks (tasks canonically used to
investigate priming effects), when compared to healthy controls (Koethe et al., 2006; Morgan et
al., 2010; Moritz et al., 2001).
It has been proposed that cannabis may have a causal role in the development of
schizophrenia. A recent systematic review concluded that cannabis increases the likelihood of
developing schizophrenia in at-risk populations, and may exacerbate symptoms of schizophrenia
in people who already have the disorder (Patel et al., 2020). Further, a longitudinal study found
that daily cannabis users have a rate of psychotic symptoms which is 1.6-1.8 times higher than
non-users (Fergusson et al., 2005). Even in non-clinical populations, it has been found that
schizotypal traits increase during acute cannabis intoxication, and studies have additionally

9
reported that regular cannabis users display higher levels of schizotypal traits, even when not
under the influence of the substance (Dumas et al., 2002; Morgan et al., 2010).
Importantly, populations with schizophrenia have high rates of cannabis use (Patel et al.,
2020). Patel et al. (2020) suggest that a schizophrenia diagnosis may predict the use of cannabis,
rather than the converse. People with schizophrenia may use cannabis to manage the symptoms
of schizophrenia, and may benefit particularly from using cannabis with high cannabidiol (CBD)
content (Patel et al., 2020). CBD, in contrast to Δ-9-THC, generally does not cause acute
psychoactive effects, and may even partially attenuate the subjective effects of Δ-9-THC (Patel
et al., 2020). Patel et al. (2020) discuss how CBD shows promising results for treatment-resistant
schizophrenia, and note that studies have found improved cognitive performance in people with
schizophrenia who use cannabis compared to people with schizophrenia who have not used
cannabis (Hanna et al., 2016). Overall, the direction of causality in the relationship between
cannabis use and schizophrenia is not clear, and compelling arguments have been made for
causality in both directions of the relationship.
Chapter Three
3.1 Memory
Many cognitive psychologists believe there are different subtypes of memory, which may
rely on different neural systems for activation (although some authors argue that there is
considerable overlap between these systems; see Addis (2018) for a relevant discussion). A
distinction in the literature is often made between semantic memory and episodic memory; while
semantic memory is generally construed as being concerned with understanding of objects,
places, and people (e.g., I know that the Eiffel Tower is in Paris), episodic memory is focused on
events and personal experiences (e.g., I can remember the time I went to see the Eiffel Tower)

10
(Nastase & Haxby, 2016). A body of literature suggests that semantic memory can be accessed
in different ways depending on context, and that semantic information which is relevant is
reactivated differently depending on current goals, recent and long-term experience, and the
passage of time during which semantic memories are reconstructed (Yee & Thompson-Schill,
2016).
3.2 Schizophrenia and Memory
Memory impairments have been observed in schizophrenia; in particular, deficits to
episodic memory are well documented (Guo et al., 2019). Functionally, it has been suggested
that the dorsolateral prefrontal cortex (DLPFC), which is thought to be involved in complex
cognitive operations, is important for episodic memory (Ragland et al., 2015). Guo et al. (2019)
suggest that hypoactivity in the DLPFC may be a potential mechanism by which episodic
memory is impaired in schizophrenia. Indeed, a meta-analysis found hypoactivity in the DLPFC
among people with schizophrenia during episodic memory encoding and retrieval tasks (Ragland
et al., 2009).
Interestingly, as Guo et al. (2019) note, findings of dysfunctional activity in the medial
temporal lobe (MTL) during episodic memory encoding and retrieval are inconsistent across
studies, with some studies reporting hypoactivity or hyperactivity in different regions of the
MTL during episodic memory tasks (Jessen et al., 2003; Ragland et al., 2004). The MTL has an
important role in memory; in particular, it has been proposed as a binding site where perceptual
details from other parts of the neocortex are bound to create coherent long-term memories
(Squire & Zola-Morgan, 1991). It is curious that widespread hypoactivity in the MTL has not
been consistently observed in people with schizophrenia, and suggests that the mechanism by

11
which people with schizophrenia experience memory deficits is more nuanced than widespread
hypoactivity.
Semantic memory is also compromised in schizophrenia (Mckay et al., 1996; Tan et al.,
2020). Mckay et al. (1996) found that people with schizophrenia performed worse than controls
on a battery of semantic memory tasks, including naming pictures, defining features of stimuli,
and category fluency. Category fluency tasks are often used as an index of semantic memory
capability, and involve participants naming as many exemplars as possible for a given category
in a given span of time, often one minute. Tan et al. (2020) found in a recent meta-analysis that
performance on category fluency tasks was consistently impaired in people with schizophrenia.
Interestingly, Mckay et al. (1996) found that people with schizophrenia performed comparably to
controls in dichotomizing stimuli as living vs. humanmade, but performed worse when
categorizing items at lower conceptual levels (i.e. categorizing birds vs. sea creatures) or
attributional levels (i.e. categorizing fierce vs nonfierce animals). This suggests that semantic
memory in schizophrenia may be intact at higher conceptual levels, but deficits may become
apparent when more concrete and detailed memory representations are probed.
3.3 Cannabis and Memory
The link between cannabis use and its potential effects on memory is not fully
understood. Some studies have reported that regular cannabis use predicts poor verbal recall on
memory tasks (Battisti et al., 2010). Morrison et al. (2009) found that intravenous Δ-9-THC
hindered performance on the digit span task, a task designed to study working memory capacity.
McKetin et al. (2016) found that regular cannabis users showed impaired immediate recall on
verbal tests, but no significant difference in delayed recall ability compared to controls. Further,

12
McKetin et al. (2016) found no evidence that cannabis use was associated with greater decline in
recall ability over time after analyzing the results of 4- and 8-year follow-ups.
Importantly, the idea that cannabis use has pernicious effects on memory is more readily
accepted in the context of adolescent populations. A study found that adolescents who regularly
used cannabis performed more poorly on verbal recall and recognition tasks (Solowij et al.,
2011). Further, the age of onset of cannabis use, duration, quantity, and frequency of cannabis
use were correlated with the extent to which performance on recall and recognition tasks was
impaired. In sum, although the effects of cannabis on memory have been studied, little is known
about the effects on semantic memory associated with cannabis use, and the neural correlates
which may underlie semantic memory alterations associated with cannabis use have not been
fully explored.
3.4 Spreading Activation
It has been proposed that semantic memory can be represented with network models
(Collins & Loftus, 1975; Zemla & Austerweil, 2018). Such models postulate that concepts in
memory are represented by nodes in a network, and these nodes are connected with edges.
Concepts which are closely semantically related (i.e. “doctor” and “nurse”) will have a shorter
semantic distance between them than concepts which are unrelated (i.e. “doctor” and “orange”),
such that the nodes of closely related concepts will be connected by relatively short edges (or
with fewer edges and nodes between them) compared to the nodes of unrelated concepts. Figure
1 shows an adapted version of the spreading activation model (Collins & Loftus, 1975).
Figure 1

13
Depiction of the Spreading Activation Model. Adapted From Collins & Loftus (1975).

Note. Concepts in the model, represented by nodes, are linked to other concepts via edges. For
example, the concept Green is linked to the concepts Apples, Pears, Leaves, Trees, and Plants.
Spreading activation refers to the idea that neural activation spreads throughout semantic
networks, from node-to-node, starting from the initial node at which a semantic search begins
(Collins & Loftus, 1975; Quillian, 1969); hence, activation ‘spreads’ from node to nearby nodes.
Lexical decision tasks have been used since the early 1970s to assess spreading activation
through semantic priming paradigms (Meyer & Schvaneveldt, 1971). Typically, in lexical
decision tasks, a pair of words is presented quickly in succession, and people must make a
judgement about whether the second (target) word is a real word or a non-word. If the target
word is semantically related to the first (prime) word (i.e. prime: doctor, target: nurse), people
are faster to make judgements about whether the target word is a real word (e.g. Neely, 1977).
Non-words often consist of a string of letters which is phonetically plausible, i.e. able to be

14
pronounced, but contain no meaning. If a prime word is able to facilitate a response to the target
word, it is taken to suggest that a spread of activation has occurred throughout nearby nodes,
which caused the target word to become activated in some capacity in response to the prime,
allowing for facilitated processing (e.g. Morgan et al., 2010).
Another feature of semantic priming tasks which may be manipulated to investigate
semantic networks is stimulus onset asynchrony (SOA). SOA refers to the length of time after
the prime word before the target word is presented. For example, Morgan et al. (2010) used a
short SOA of 250 ms and a long SOA of 750 ms to tease apart how semantic priming may differ
as a function of SOA. It has been suggested that short SOA priming tasks (< 500 ms SOA) are a
better metric of measuring automatic spreading activation, while long SOA tasks (> 500 ms
SOA) may be used to examine more conscious and deliberate processes (Moritz et al., 2001).
3.5 Spreading Activation in Thought-Disordered Schizophrenia
In people with schizophrenia, thought disorder refers to the incoherent or disorganized
speech observed in some people with the condition (Pomarol-Clotet et al., 2008). Authors have
proposed different metrics for identifying the presence of thought disorder in schizophrenia.
Moritz et al. (2001) quantified thought disorder using the Positive and Negative Disorganized
Symptoms Scale, which consisted of a semi-structured interview in which speech abnormalities
were assessed. Other authors have measured thought disorder by quantifying the extent to which
bizarre answers are given in response to prompts (Nordgaard et al., 2021). Interestingly,
increased semantic priming has been found in people with thought-disordered schizophrenia
using short SOA semantic priming tasks (Moritz et al., 2001). Moritz et al. (2001) suggest that
disinhibited semantic networks are responsible for increased semantic priming in thoughtdisordered schizophrenia. Excessive spreading activation among nodes in a semantic network

15
may lead to concepts which would normally be distally related being activated in concert with
one another, causing the speech of someone with thought disorder be composed of more distally
related concepts.
It is important to acknowledge the presence of thought disorder in the finding of
increased semantic priming in schizophrenia (Moritz et al., 2001). When the group with
schizophrenia was split between those with and without thought disorder, the enhanced semantic
priming effect remained significant only for those with thought disorder. Indeed, a meta-analysis
of semantic priming in schizophrenia suggested that schizophrenia itself is not associated with
increases in semantic priming, but rather people with both schizophrenia and thought disorder
may display increased semantic priming compared to healthy controls (Pomarol-Clotet et al.,
2008). It may be that disorganized thought associated with schizophrenia, rather than chronic
schizophrenia per se, is associated with disinhibited semantic networks and automatic semantic
priming.
Importantly, it has been found that people with thought-disordered schizophrenia may
exhibit reduced semantic priming during tasks with longer SOAs of 500ms or more (e.g. Besche
et al., 1997). This contrasts with the increased semantic priming observed in people with
thought-disordered schizophrenia during shorter SOA tasks. In summary, spreading activation
appears to be influenced by thought-disordered schizophrenia, although whether the semantic
networks of people with thought-disordered schizophrenia exhibit increased or decreased
priming compared to neurotypical individuals appears to depend on the SOA.
3.6 Spreading Activation and Cannabis
A study investigating semantic processing in cannabis users also found that acute
cannabis intoxication leads to increased semantic priming in short SOA conditions during lexical

16
decision tasks (Morgan et al., 2010). The authors interpret this finding to suggest that acute
cannabis intoxication leads to increased automatic spreading activation among nodes in semantic
networks. Cannabis users also displayed increased semantic priming in long SOA conditions
compared to controls, even when abstinent from the drug for a period of 3-5 days. The finding
that cannabis intoxication is associated with increases in short SOA semantic priming while
regular cannabis use is associated with increases in long SOA semantic priming suggests that
cannabis intoxication may affect automatic spreading activation processes, while regular
cannabis use may result in alterations to more deliberate semantic retrieval processes (Morgan et
al., 2010). It is possible that the disinhibited semantic networks account of spreading activation
proposed by Moritz et al. (2001) may explain increased spreading activation both in
schizophrenia and cannabis intoxication.
Chapter Four
4.1 Top-Down Inference and Predictive Processing
Traditional models of perceptual processing propose that perception is a process guided
primarily by our interactions with the environment, causing perceptual information to impinge
on sensory receptors (e.g. Sinnett et al., 2016). This information is passed through the cortex to
cells which respond to increasingly complex stimuli, which the brain interprets and uses to form
a coherent representation of the world. For example, if someone is out on a hike and sees a deer
in the woods, a proponent of traditional models of perceptual processing might suggest that light
is reflected off the deer and reaches the photoreceptor cells of the eye. This information is
ultimately transmitted through the visual pathway in the brain, where cells first process the lowlevel visual features contained in the visual field (i.e., lines, edges), followed by interpreting
more complex shapes, and eventually recognizing the information which impinged on the

17
photocells as a deer. Thus, this “bottom-up” perceptual information is a primary driver for
perception. However, it is proposed that “top-down” processing also exerts influences on
perception. Top-down processing refers to the brain’s biasing of perception based on beliefs and
expectations, typically accumulated through experience with the world. A classic example of
top-down processing is the rat-man illusion (Figure 2; see Bugelski & Alampay, 1961). In the
illusion, participants are presented with a line drawing which can be interpreted as either a
curled-up rat or the face of a man. Bugelski & Alampay (1961) found that when participants
were shown pictures of humans before being shown the rat-man figure, they were more likely to
perceive the drawing as a man, whereas if they were shown pictures of animals, they were more
likely to perceive the drawing as a rat. This suggests that the brain is shaping perception based on
prior expectations; in this case, after seeing a series of drawings of animals, it is likely that
another drawing of an animal would be shown. The ambiguous rat-man figure being perceived
one of two different ways suggests that perception can be guided by top-down processes.
Figure 2
Ambiguous Rat-Man Figure. Adapted from Bugelski & Alampay (1961).

18
Note. The figure can be interpreted either as the face of a man with glasses on, or a mouse with
its tail curled.
However, some theorists have proposed a radical overhaul of the way cognitive
psychologists think about perceptual processing. Theories of predictive coding, or predictive
processing, propose that the brain is not primarily guided by incoming bottom-up sensory
information, but is instead a “predictive machine”, aiming to constantly predict what is
happening in the environment, and affirming or adjusting its predictions based on actions
occurring within the environment (e.g. Friston, 2010; Clark, 2013). In other words, the brain
generates probabilistic models of ‘what is out there’, and tests them against its self-generated
sensory input (Friston, 2010). A critical pillar of this theory is that the probabilistic models are
hierarchically structured; models which are more spatially and temporally precise occur towards
the bottom of the hierarchy; conversely, towards the top of the hierarchy, predictions and
generative models become increasingly spatially and temporally abstract. Radical predictive
processing models of perception suggest that most sensory information is not passed upwards
through the hierarchy towards higher-order association cortices of the brain, because the brain’s
probabilistic model is able to “explain away” the expected sensory information. Instead, when
the brain’s prediction is not congruent with incoming sensory information, a “prediction error”,
or residual, is signaled, and it is this prediction error (deviation from the expected model) which
is passed upwards through the neural hierarchy towards higher-level association cortices (Clark,
2013). Such errors help update future predictions about incoming information from the
environment.
A proponent of predictive processing theories might propose that someone walking in the
forest who sees a deer has generated a prediction of a deer, based on previous experiences

19
walking in the woods, and matches this prediction of the deer against the incoming perceptual
input. Because the visual properties (i.e., lines/edges/shapes of which the deer is composed)
generated by the incoming perceptual information are consistent with the predictions made of a
deer, no radical adjustments need to be made to the model, and the perceptual information can be
“explained away” at low levels of the neural hierarchy.
The precision, (or “gain”, which is the inverse of variance) of prediction errors is another
key concept within the predictive framework (Mathys et al., 2011). When sensory input is
processed, prediction errors are propagated upwards through the cascade of predictions in the
event of a mismatch between the brain’s probabilistic model and the sensory information
obtained in the world (e.g. Clark, 2013). When this happens, theorists have argued that the brain
places differing degrees of “confidence” in the error signals (e.g. Mathys et al., 2011; Clark,
2013). Errors which are “noisy” or unreliable are assigned a lower weighting than prediction
errors which are processed with a higher degree of certainty. For example, if someone is out on a
hike during a bright clear day and hears a rustling in the forest, the brain might form a model
with the expectation that a deer is most likely responsible for the rustling, based on previous
experiences walking in the woods. When the person sees a bear step out of the woods instead, a
prediction error with high precision is generated, and attention is drawn to this prediction error –
this is most definitely not the deer the brain was predicting, and the brain’s model of the world
needs to be quickly updated.
Now imagine that the same person is walking through the woods at night, and again hears
a rustling which their brain expects to most likely be a deer. This time, the brain sees a shadow
against the trees, but it is dark out, and it is not apparent as to what caused the shadow. The
context (dark, low visibility) establishes that the reliability of the bottom-up sensory input is

20
impaired. In this case, a prediction error is generated with a lower precision, as the bottom-up
perceptual input is noisy. This noisy prediction error has less influence on the brain’s model, and
is not able to make changes to the probabilistic model as readily as a more precise prediction
error. Here, it may take longer for the person to realize the rustling is not coming from a deer
after all.
4.2 Schizophrenia, Cannabis, and Predictive Processing
The hollow mask illusion, depicted visually in Figure 3, has long been used to study the
effects of top-down influences on perception. Typically, when people view a mask shaped like a
human face from the inside (i.e., looking at the reverse, concave part of the mask), they often
perceive it as being convex (e.g. Corlett et al., 2009). In other words, their perception is similar
to how they might perceive the outside of the mask, or a regular human face. This provides an
example of the brain using prior beliefs and expectations to estimate the cause of the perceptual
input; in the real world, we encounter convex faces much more frequently than concave faces.
When we encounter a hollow mask, the brain is making its best estimate of the veridical nature
of the inside of the mask, which is that the inside of the mask is convex (Corlett et al., 2009).
Figure 3
Images Showing the Progression of the Hollow-Mask Illusion Using a Rotating Hollow Mask.

21

22

Note. Still images taken from a GIF from https://imgur.com/gallery/jQNUqIl. These images
depict a hollow mask, which typically appears convex during its rotation to neurotypical
observers. However, people with schizophrenia and people under the influence of cannabis tend
to perceive the mask as concave during parts of its rotation, which some theorists suggest
indicates a reduced influence of top-down processes.
Intriguingly, some behavioral evidence suggests that both schizophrenia and cannabis
usage may affect the way in which predictive processing occurs. Further, it may be that the
predictive processing process is affected in similar ways by schizophrenia and cannabis usage.
Research has found that the binocular depth inversion illusion (BDII), which is conceptually
related to the hollow mask illusion, has less of an effect on people with schizophrenia or healthy
people under the influence of Δ-9-THC (Emrich et al., 1991; Koethe et al., 2006). In the BDII, a
three-dimensional object is pseudoscopically presented to participants, meaning that visual
information which would normally be presented to the right eye is instead presented to the left,
and vice versa (Koethe et al., 2006). Theoretically, presentation of visual information in this way
should lead to inverted depth perception of objects (i.e., objects which are typically convex

23
should appear concave, and vice versa), much in the same manner as the hollow mask illusion.
However, people with schizophrenia and people under the influence of cannabis tend to report
that objects, as well as faces, presented in this manner retain the same percepts as when the
objects are presented normally (i.e., people with schizophrenia and people under the influence of
Δ-9-THC tend to perceive faces which are presented pseudoscopically as convex, when
theoretically the percept should be inverted and the face should appear concave) (Emrich et al.,
1991; Koethe et al., 2006). From the perspective of predictive processing, one possible
explanation for these effects is that the precision of prediction errors may be improperly
integrated if people have schizophrenia or are under the influence of Δ-9-THC. Perhaps the
precision of prediction errors, in the case of BDII and hollow mask illusion, is weighted more
highly in schizophrenia and cannabis intoxication than under neurotypical conditions. This could
lead to an increased influence of bottom-up perceptual input, which would allow probabilistic
models to be more readily influenced by perceptual input. Indeed, some authors have argued that
improperly balanced regulation between prior models and sensory input are responsible for
hallucinations (Stephan et al., 2009). However, it is important to note that others have argued
that an increased confidence in prior models is actually responsible for hallucinations (Collerton
et al., 2005). From this perspective, it may be that overconfidence in an aberrant model causes an
individual to perceive the world through the framework of this rigid model, without
incorporating the necessary prediction errors which would be signaling that the brain’s model is
incomplete or insufficient. Other accounts have suggested that both weak and strong prior beliefs
may be present in different perceptual modalities and interact in complex ways to give rise to the
phenotypic characteristics of psychosis (Sterzer et al., 2018). While predictive processing

24
accounts of schizophrenia provide compelling explanations for the symptoms of schizophrenia,
the literature remains inconclusive on important theoretical considerations.
Fletcher & Frith (2009) suggest that both delusions and hallucinations experienced by
people with schizophrenia can be elegantly characterized through a predictive processing
framework. They propose that sensory evidence encountered in the environment is improperly
integrated with prior models. This results in false prediction errors being propagated up the
neural hierarchy to higher levels of abstraction (Fletcher & Frith, 2009). They propose that the
confidence of prediction errors is incorrectly regulated in people with schizophrenia, such that
prediction error signals which ought to have a high estimate of confidence may instead be
propagated with a low estimate of confidence and vice-versa. In neurotypical individuals, a
“noisy” prediction error may not be accepted with a great degree of confidence. However, in
people with schizophrenia, noisy prediction errors may be given high confidence, while
prediction errors which should represent a clear signal are given less confidence, causing the
brain to allow prediction errors to change its models in aberrant ways. In schizophrenia, this may
manifest as stimuli which would generally be considered uninteresting or unimportant seeming
intriguing or highly relevant, while stimuli which are veridically important are ignored (Fletcher
& Frith, 2009). It may be that, when prediction errors which would normally be considered noisy
are maintained and given a high degree of confidence, the brains of people with schizophrenia
attempt to explain this unaccounted uncertainty by generating delusional beliefs. In a world
where incidental stimuli are deemed meaningful, models of the world must change to incorporate
the meaningfulness in ways which result in aberrant and unusual beliefs and thoughts
(delusions). Indeed, people with schizophrenia commonly report that they feel everyone is
plotting against them, which may be a consequence of having many unimportant stimuli seem

25
important in some way and having to generate a belief system which accounts for these
“coincidences” (Fletcher & Frith, 2009).
Friston et al. (2016) further suggest that a failure to properly regulate the precision of
prediction errors is responsible for pathophysiological characteristics of schizophrenia including
delusions and hallucinations. They propose that pyramidal cells, which are the most abundant
type of neuron in the cerebrum, and are believed to be involved in complex cognitive functions,
have an integral role in predictive processing and in schizophrenia (Benavides-Piccione et al.,
2021; Elston, 2003; Friston et al., 2016). Importantly, it has been suggested that variation in
pyramidal cell characteristics may be responsible for their roles in many different cognitive
functions (Elston, 2003). Friston et al. (2016) suggest that generally, prediction errors are
propagated from superficial (closer to the scalp) to deep (deeper in the cortex) pyramidal cells.
Further, they suggest that superficial pyramidal cells are involved in updating expectations at
lower levels of a predictive processing hierarchy, while deep pyramidal cells are hypothesized to
encode high-level expectations and beliefs. In schizophrenia, these prediction errors are
propagated with unduly high precision, which effectively results in attention being directed
towards prediction errors which should be of little interest. Indeed, the idea that attention can be
directed by varying the precision of prediction errors has been widely discussed (e.g. den Ouden
et al., 2012). Friston et al. (2016) propose that prefrontal deficiencies in regulating the precision
of prediction errors may underlie hallucinations and delusions.
Cannabis is also thought to affect predictive processes. Disruptions to high-level
probabilistic models may be responsible for some of the subjective effects of cannabis (Corlett et
al., 2009). The idea that top-down influences on perception (probabilistic models) are weakened
under the influence of psychoactive substances is supported by recent theoretical and

26
experimental advancements in the study of psychedelic drugs. Importantly, cannabis is not
typically classified as a psychedelic drug, so comparisons between Δ-9-THC and the so-called
“classic psychedelics” should be made with some reservation. There are, however, some
similarities between the mental states invoked by cannabis and those invoked by psychedelics, so
a brief discussion will be included (Nichols, 2016).
As previously discussed, it has been suggested that pyramidal cells are integral to
predictive processing (Friston et al., 2016; Carhart-Harris & Friston, 2019). Classic psychedelics
exert their effects largely through acting on serotonin 5-HT2A receptors, found abundantly on
deep-layer pyramidal cells in the cortex (particularly layer V) (Carhart-Harris & Friston, 2019).
Importantly, CB1 receptors, a primary binding site for cannabinoids including Δ-9-THC, are also
found in pyramidal cells (Hill et al., 2007). This suggests a potential mechanism by which
cannabinoids like Δ-9-THC act to influence the specification of predictions, which may
ultimately disrupt normal top-down processing.
Empirical evidence is emerging which suggests a role of psychedelics in increasing
bottom-up informational transmission, at least for the visual modality (Alamia et al., 2020).
Alamia et al. (2020) discuss previous research from their lab on travelling alpha waves, which
are rhythmic oscillations in electrical brain activity spreading across cortical regions. “Forward”
travelling alpha waves spread from occipital to frontal regions of the brain during visual input.
Conversely, “backward” travelling waves tend to predominate in the absence of perceptual input,
spreading from frontal to occipital regions, and may reflect the transmission of top-down prior
beliefs (Alamia & VanRullen, 2019). Alamia et al. (2020) found that injection with N,N
Dimethyltryptamine (DMT), a powerful psychedelic, increased forward travelling cortical waves
and decreased backward travelling waves during eyes-closed rest. This may reflect DMT altering

27
predictive processing during eyes-closed rest, by reconfiguring the brain state to give more
weight to bottom-up sensory input and thus reduce the influence of top-down beliefs. In
summary, evidence has suggested that schizophrenia and drugs including cannabis and classic
psychedelics may affect the brain’s predictive processing capabilities. Future empirical work
grounded in neurophysiology, motivated by a predictive processing framework, may allow
researchers to better elucidate the mechanisms by which perceptual processing is altered through
schizophrenia and cannabis usage.
In the following section, I review event-related potential research to motivate the idea
that disruptions to predictive processing caused by schizotypy and cannabis use will have
observable and quantifiable neurophysiological manifestations.
Chapter Five
5.1 EEG and Semantic Prediction
Researchers have used neuroimaging techniques, including electroencephalography
(EEG), to study predictive processing. EEG is a non-invasive form of brain imaging which
measures voltage changes across the scalp, and is particularly sensitive to the dipoles produced
by cortical pyramidal cells (Kirschstein & Köhling, 2009). Using EEG, researchers can
investigate activity of the brain with high temporal resolution. Specifically, research involving
event-related potentials (ERPs) is used to test hypotheses about cognitive processing. ERPs elicit
characteristic signals in EEG data which enable researchers to observe changes in brain activity
with temporal precision on the order of milliseconds. ERPs are often sensitive to the expectancy
of a particular type of stimulus, or violations thereof. For example, the P300 component is an
ERP component, canonically appearing around 300ms following the onset of a particular
stimulus, which is thought to be related to perceptual processing, and varies in amplitude if a
stimulus is presented which is incongruent with a set of previously presented stimuli, commonly

28
studied in “oddball” paradigms (Luck, 2014). In summary, ERP research has allowed researchers
to make marked discoveries about the brain’s perceptual processing abilities, particularly with
respect to temporal dynamics.
5.2 The N400 and the Integration vs Prediction Debate
The N400 is an ERP component which researchers have studied to understand semantic
memory. Although there is variation in the paradigms researchers use to study the N400, a
common task involves participants reading sentences which either have endings which are
semantically congruent (i.e., “while I was visiting my hometown I had lunch with several old
friends”) or endings which are semantically incongruent (i.e., “while I was visiting my
hometown I had lunch with several old shirts”). Immediately following the presentation of an
incongruent final target word in a sentence (i.e., a word which violates the preceding semantic
context of the sentence), an N400 with a large amplitude can be observed, compared to a smaller
N400 amplitude after reading sentences with congruent endings. Figure 4 shows N400
waveforms in response to congruent and incongruent sentence endings (León-Cabrera et al.,
2019). The difference between the amplitudes of the N400 potential in response to congruent
compared to incongruent endings refers to the strength of the N400 effect. Further, the strength
of the N400 effect is related to the level of semantic constraint of the sentence (Grisoni et al.,
2021; Kutas & Hillyard, 1984). Semantic constraint refers to the likelihood that a sentence will
have an ending which is highly predictable (i.e., “The goalkeeper managed to catch the ball” is
highly constrained while “As a present she gave her son a ball” is low in semantic constraint)
(León-Cabrera et al., 2019). Thus, the strength of the N400 effect is greater when sentences are
high compared to low in semantic constraint. These findings indicate that neural processing is

29
sensitive to both the semantic constraint of sentences and the congruency of the final words of
sentences within the preceding semantic context.
Figure 4
Depiction of Differences in the N400 Potential in Response to Congruent and Incongruent
Target Words (León-Cabrera et al., 2019). Copyright 2019 by Patricia León-Cabrera, Amanda
Flores, Antoni Rodríguez-Fornells, and Joaquín Morís. Reprinted with permission.

Note. Negative is plotted upward.
DeLong et al. (2005) attempted to elucidate whether the brain aims to predict the
upcoming semantic information of sentences, or whether it merely integrates new information
with the previous content of the sentence, where the difficulty of integration varies with the
extent to which neural resources are needed to integrate new content into a sentence
representation. They presented sentences one word at a time, where the final word in the
sentence (a noun) was preceded by either the article “a” or “an”, depending on the appropriate
article for the final noun. The authors were able to observe an N400 effect (difference in N400
amplitude between conditions) when comparing ERPs of sentence fragments which ended in
different articles. DeLong et al. (2005) give the example of the sentence “The day was breezy so

30
the boy went outside to fly a kite”. An unexpected, although plausible, ending to the same
sentence could be “The day was breezy so the boy went outside to fly an airplane”. When
comparing ERP amplitudes in response to sentence fragments ending in the more likely expected
article (“The day was breezy so the boy went outside to fly a”) to sentence fragments ending in
the more unexpected article (“The day was breezy so the boy went outside to fly an”), the
authors noticed an N400 effect. These amplitudes were inversely related to the Cloze probability
of the sentence. Note that the Cloze probability of a sentence indicates how likely it is that the
sentence will be completed with a particular word; for example, the sentence “Bob proposed and
gave her a diamond ring” has a high Cloze probability, as ring is very likely to be the word that
fits logically as the final item when reading the preceding sentence (Block & Baldwin, 2010).
Conversely, the sentence “Sandra enjoyed going for walks right before dinner” has a low Cloze
probability, as there are many words which could fit logically after reading “Sandra enjoyed
going for walks right before…” DeLong et al. (2005) argued that since “a” and “an” serve
exactly the same purpose in the English language, there should be no difference in the difficulty
with which either article may be integrated into the representation of a sentence (unless “an” is
more difficult to integrate due to being longer, in which case a larger amplitude may be expected
for “an” in each instance, but this was not the case). However, because differences in the N400
effect were found in response to articles which were related to the Cloze probability of the
sentence, the authors argue that the brain is indeed attempting to predict upcoming words when
reading sentences. Notably, however, this finding has been criticized due to failed replication
attempts (Pulvermüller & Grisoni, 2020).
Importantly, there are multiple layers at which written language can be interpreted and
predicted, which convey different pieces of information. For example, semantics generally refers

31
to the meaning of words or sentences. Conversely, phonemes refer to the sounds made by letters
or particular combinations of letters in words, while syntax refers to the organization of words,
with the goal of creating coherent messages. All of these factors possess information which may
be meaningful or relevant to the brain when people are reading. Kuperberg & Jaeger (2016)
discuss language comprehension through a predictive processing framework. They argue that the
available evidence suggests that the brain is indeed predicting the upcoming content of language
at multiple layers of representation. For the brain to do this, they suggest, it must aim to predict
so-called “message-level representations”. When interpreting language, the brain does not aim to
specifically predict the semantic categories, or phonological content, or syntactic organization of
the following words; rather the goal is to understand the full meaning or message of the input. To
do so, the brain must iteratively update its beliefs regarding which is the most probable
“solution” of the input it is encountering. Kuperberg & Jaeger (2016) suggest the brain employs
a form of predictive processing like that proposed by Clark (2013) and Friston (2010) when
comprehending language. This idea allows for speculation that there may be message-level
representations in the brain of events, with high-level representations of objects (which would be
semantic in nature), as well as lower-level representations of syntax and phonemes in sentences.
Specifically, Kuperberg & Jaeger (2016) elaborate on the distinction between whether the
brain is actually pre-activating components of a representation during language processing, or
whether the information gleaned by processing incoming language merely serves to facilitate the
processing of the following informational content. For example, if one were to read the sentence
fragment “The day was breezy so the boy went outside to fly a…”, enhanced processing of the
word “kite” can be explained under at least two conflicting hypotheses. Kuperberg & Jaeger
(2016) suggest that the hypothesis <boy flies kite> is likely to be stored by the brain, with a high

32
degree of confidence, prior to encountering the word “kite”. Under the facilitation hypothesis, it
is possible that the brain waits for the bottom-up input (encountering the word “kite”), which
activates the word “kite” and its phonological neighbors in memory. Then, the prior hypothesis
of <boy flies kite> is used to facilitate the selection of the word “kite” from among the
phonological neighbors. This facilitation hypothesis does not commit to the idea that a
representation of “kite” is pre-activated before encountering the word itself. Conversely, under
the pre-activation hypothesis, the brain would pre-activate the phonological representation of
“kite” before actually encountering the word.
Ultimately, Kuperberg & Jaeger (2016) argue that the brain is indeed pre-activating
components of a representation prior to actually encountering bottom-up input, consistent with
the pre-activation hypothesis. ERP research, such as that previously discussed by DeLong et al.
(2005), suggests that the brain is generating predictions during the reading of sentences, and
constructing representations of the preceding content at different levels prior to encountering the
relevant bottom-up input. Predictive processing theorists might propose that the brain is
constantly updating and affirming or adjusting its predictions; when reading language, the brain
may have high-level predictions about the semantics of the text, while it has low-level
predictions about the geometry of the lines and edges involved in constructing the shapes of
different letters.
As Kuperberg & Jaeger (2016) note, additional support for the pre-activation hypothesis
comes from studies using eye-tracking. Altmann & Kamide (1999) conducted a study where
participants were asked to judge whether an action was plausible, given the objects in a scene
which were presented on a computer monitor (e.g., whether “the person will light the fire” was a
plausible sentence depended on whether or not a fireplace was present on screen). Participants

33
were not instructed to make speedy judgements. Intriguingly, (Altmann & Kamide, 1999) found
that the onset with which people made saccadic eye-movements towards an object was faster
when only one object in a scene was a possible referent following the verb of a sentence
compared to when multiple objects could be referred to following the verb. For example,
participants saccades towards a picture of a cake placed amidst (non-edible) distractor objects
after hearing the sentence “the boy will eat the cake” occurred earlier during sentence reading
than when they heard the sentence “the boy will move the cake”. Importantly, the probability
with which participants fixated on the cake was significantly higher when hearing “the boy will
eat the cake” than “the boy will move the cake” even before the presentation of the final noun
(cake). This further suggests that the brain is predicting possible outcomes of sentences during
language comprehension, and is using contextual information and prior beliefs to guide behavior
even before the onset of a noun which qualifies the preceding information. In summary, a body
of behavioral and electrophysiological evidence suggests that the brain attempts to predict the
semantic content of language prior to encountering bottom-up perceptual input, and this is
consistent with predictive processing theories of perceptual processing.
5.3 Schizophrenia and Semantic Prediction
Like the N400, the amplitude of the mismatch negativity (MMN) ERP component is
consistently impaired in people with schizophrenia (Erickson et al., 2016). The MMN is a
component typically evoked in auditory oddball paradigms, in which ‘deviant’ stimuli (which
occur in ~20% of trials) are presented among ‘standard’ stimuli (which occur in ~80% of trials)
(Luck, 2014). The deviant stimulus elicits a negative voltage change compared to the standard
stimulus between 100-200 ms following the onset of the deviant stimulus (Luck, 2014).
Importantly, in people with schizophrenia, the difference in MMN between standard and deviant

34
stimuli is less than in healthy controls (Erickson et al., 2016). This may suggest that people with
schizophrenia are less sensitive to violations of expectancy than healthy controls.
Repetition of stimuli can also reduce the MMN. Previously, it has been proposed that
reduced MMN to standard stimuli compared to deviant stimuli in healthy participants may be
due to adaptation to the standard stimulus, which ultimately results in a reduced neural response
upon encountering the standard stimulus (Shin et al., 2012). This is consistent with ideas in
predictive processing – if a stimulus is repeatedly presented, the brain will be less surprised by
its presentation, and be able to “explain away” the presence of the stimulus, resulting in reduced
neural firing. A hypothesis which stems from the finding of reduced MMN response in
schizophrenia is that people with schizophrenia have impairments in adapting to repeated stimuli
and updating models of the world (see also Shin et al., 2012).
As previously discussed, hallucinations in schizophrenia may arise from an improper
regulation of the precision of prediction errors. When the brain produces speech, it must produce
a copy of the signal which “tells itself” that the speech it is hearing is self-produced, so it does
not treat self-produced speech as external speech (e.g. Ford et al., 2007). Suppression of the
amplitude of the N1 component in response to producing vs listening to speech has been used as
an indicator of how the brain is able to account for its own speech production; in neurotypical
people, the N1 is suppressed when producing compared to listening to speech (Ford et al., 2007).
Ford et al. (2007) found that the N1 was less suppressed in people with schizophrenia when
talking compared to listening to speech, which the authors suggest may reflect a failure in selfmonitoring. Fletcher & Frith (2009) note that self-generated actions should be predictable and
thus unsurprising, while the lack of N1 suppression in people with schizophrenia suggests the
possibility that self-generated speech may be, at some level, more surprising and salient in

35
schizophrenia than in neurotypical people. By extension, self-generated thoughts in people with
schizophrenia may be less predictable, and it is possible this may also be due to aberrant
prediction error signaling. The brain may be improperly able to attribute its own thoughts to
itself due to prediction error mismatches in high level association cortices, leading the brain to
believe something surprising or unusual is happening, and misattributing thoughts or “inner
voices” to other sources. Indeed, many auditory hallucinations reported in schizophrenia are of
voices commenting on one’s actions (Fletcher & Frith, 2009). In conclusion, ERP research
suggests that people with schizophrenia may experience impairments in adapting to repeated or
self-generated stimuli. It may be that people with schizophrenia have an aberrant ability to
predict upcoming stimuli due to improper regulation of the precision of prediction errors with
probabilistic models.
5.4 N400 in Schizophrenia
Schizophrenia’s effects on the N400 potential have been thoroughly investigated,
although findings are complex (Mohammad & de Lisi, 2013). In a meta-analysis, Wang et al.
(2011) report a reduced N400 effect in patients with schizophrenia in short SOA paradigms
(operationally, the N400 amplitude difference in response to congruent vs incongruent words in
patients with schizophrenia is reduced compared to healthy controls). Two opposing theoretical
views may account for this phenomenon (Mohammad & de Lisi, 2013; Wang et al., 2011). The
first viewpoint states that people with schizophrenia may experience an impairment in activating
related semantic connections, such that little facilitation occurs for semantically related words
due to an inefficient or impaired activation of the related nodes in the semantic network.
Alternatively, the second viewpoint suggests that people with schizophrenia may experience
semantic network hyper-priming, such that even unrelated concepts are facilitated due to

36
unusually strong connections between what would otherwise be relatively distal concepts
(consistent with semantic priming research described earlier). This account is consistent with the
suggestion that disinhibited semantic networks underlie hyper-priming in schizophrenia (Moritz
et al. 2001).
In long SOA tasks (>500ms SOA), people with schizophrenia often exhibit a reduced
N400 effect compared with healthy controls, which may reflect deficits in integrating new
information within the preceding semantic context (Wang et al., 2011). The idea that people with
schizophrenia experience deficits in using context to guide semantic interpretation has been
supported by empirical evidence from other studies. Salisbury et al. (2002) found that people
with schizophrenia show greater N400 amplitudes to subordinate homographs compared to
controls. For example, when presented with the sentence “The toast is sincere”, people with
schizophrenia displayed N400 potentials in response to the word “sincere” which were
comparable to those displayed when they read sentences they judged as nonsensical. In this
sentence, the word “toast” refers to the act of drinking to one’s health, rather than its more
common semantic function as a piece of warm browned bread. Control subjects displayed
reduced N400 effects in response to sentences containing these subordinate homographs
compared to sentences they judged as nonsensical. This finding further suggests difficulties in
using semantic context to guide comprehension in schizophrenia (Wang et al. (2011).
Variation in the magnitude of the N400 effect has also been observed in healthy people
from general populations who differ on trait schizotypy. Kiang & Kutas (2005) presented
participants with category descriptions (i.e. ‘a type of fruit’) followed by either a high-typicality
exemplar (i.e. ‘apple’), a low-typicality exemplar (i.e. ‘cherry’), or a non-exemplar (i.e. ‘clamp’).
They found that the size of the N400 effect generated when comparing non-exemplars to both

37
types of category exemplars was reduced in people high in schizotypy. Kiang & Kutas (2005)
propose that high schizotypy leads to hypoactivation of the semantic network for related targets
and hyperactivation of the semantic network for unrelated targets.
5.5 Cannabis and Semantic Prediction
Few studies have investigated the relationship between cannabis use and neural markers
of semantic prediction. One study aimed to explore differences in the N400 potential in regular
cannabis users during a typical N400 semantic priming task (Kiang et al., 2013). The authors
hypothesized that cannabis would influence semantic priming, resulting in a reduced N400 effect
for cannabis users compared to controls, due to previous literature finding reduced semantic
priming effects in schizophrenia at short SOAs. In the study, participants were presented with
pairs of words which were either semantically related (metal-steel), unrelated (donkey-purse), or
a word followed by a non-word (dress-zores) with a SOA of 750 ms. Interestingly, and contrary
to the author’s hypotheses, cannabis users exhibited reduced N400 potentials to both related and
unrelated target words. The amplitude of the N400 potential decreased globally irrespective of
whether a related or an unrelated word was presented. Consequently, the discrepancy between
contextually related and unrelated target words (i.e., the strength of the N400 effect) was similar
between cannabis users and controls. However, the authors did find a correlation between levels
of delusion-like ideation and the extent to which N400 amplitudes were attenuated, both to
related and unrelated target words, with higher levels of delusion-like ideation being associated
with greater decreases in N400 potentials.
Ahmed (2019) examined the N400 in a typical semantic priming paradigm in cannabis
users who were at a clinical high risk for psychosis (i.e., individuals presenting with symptoms
of psychosis, who were deemed at risk for developing psychosis but without having yet

38
experienced a full psychotic episode). Their study included four groups; a healthy control group,
a group of healthy regular cannabis users, a group of people at clinical high-risk for psychosis
who did not use cannabis, and a group of regular cannabis users at a clinical high-risk for
psychosis. Ahmed (2019) found no significant difference in the magnitude of the N400 effect
across the four groups. However, when comparing groups at clinical high-risk for psychosis
(including both those who used cannabis and those who did not) and healthy controls (including
both those who used cannabis and those who did not), they observed a trend effect in which the
magnitude of the N400 effect decreased during long SOA trials (p = .07) in people at clinical
high-risk for psychosis. The authors interpret this finding to suggest that clinical high-risk for
psychosis may inhibit semantic priming at long SOAs, but cannabis use does not modulate this
phenomenon.
Another study investigated the effects of acute cannabis intoxication on the N400
potential by presenting a list of familiar and unfamiliar words to participants (Ilan et al., 2005).
Familiar words had been previously presented in another part of the experiment, while unfamiliar
words were shown to participants for the first time. Acute cannabis intoxication reduced the
amplitude of the N400 potential in a dose-dependent fashion, with higher doses of Δ-9-THC
resulting in larger decreases of the N400 potential across all words. Further, there was a main
effect of word type, with the N400 response being larger for unfamiliar than familiar words.
However, there was no interaction between word type and drug condition on the amplitude of the
N400, suggesting that the N400 effect was not different between cannabis and control
conditions. In conclusion, although research on cannabis and the N400 is limited, the literature
reviewed thus far suggests that cannabis does not modulate the extent to which people

39
experience semantic priming in N400 paradigms. However, no studies on cannabis so far have
employed N400 paradigms involving reading full sentences.
Chapter Six
6.1 Semantic Predictive Potential
Recent work in predictive neural processing has found that semantic constraint
modulates a specific neural component which can be measured through EEG, and is observable
prior to the presentation of a final target word while people read sentences (León-Cabrera et al.,
2017, 2019; Pulvermüller & Grisoni, 2020). Importantly, as previously discussed, the extent to
which a sentence has a predictable ending can be quantified using Cloze probability values.
León-Cabrera et al. (2019) had participants read sentences which differed in their level of
semantic constraint (“The goalkeeper managed to catch the ball” vs “As a present she gave her
son a ball”). The first sentence is highly semantically constrained (and has a high Cloze
probability); there are not a lot of other words we would normally expect to hear after the phrase
“The goalkeeper managed to catch the…”. The second is lowly semantically constrained (and
has a low Cloze probability); after hearing the phrase “As a present she gave sentence her son
a…” there are many possible concepts we can imagine taking the place of the final word in the
sentence. The authors found a difference in neural activity leading up to the last word in the
sentence (before the presentation of the final word) while participants read the two types of
sentences, shown in Figure 5. Specifically, there was a greater negativity present in the EEG
signal when participants read sentences which were high in constraint compared to sentences low
in constraint. This suggests that this neural activity is modulated by levels of semantic constraint.
Figure 5

40
Depiction of Differences in the Semantic Predictive Potential Across High Constraint, Low
Constraint, and Non-Semantic Sentences in a Sentence-Reading Task (León-Cabrera et al.,
2019). Copyright 2019 by Patricia León-Cabrera, Amanda Flores, Antoni Rodríguez-Fornells,
and Joaquín Morís. Reprinted with permission.

Note. Semantic Predictive Potential waveform over electrode FC3 for high constraint, low
constraint, and non-semantic sentences. Negative is plotted upwards. Notably, the waveform for
high constraint sentences displays greater negativity than the waveform for low constraint
sentences.
Pulvermüller & Grisoni (2020) term this component the Semantic Predictive Potential
(SPP). Intriguingly, more recent work from their lab suggests that the brain regions involved in
knowledge representation of specific concepts may be involved in the generation of the SPP
(Grisoni et al., 2021). Grisoni et al. (2021) had participants read sentences which differed in their

41
level of semantic constraint (measured by Cloze probability) and of which the final word in the
sentence was either a type of animal or a type of tool. Prior to the presentation of the final word,
activity was more prominent in parieto-occipital regions for sentences which were highly
constrained to indicate that the final word was likely to be an animal compared to a tool, while
activity was more prominent in prefrontal and premotor areas when the final word of the
sentence was more likely to be a tool compared to an animal. This provides strong evidence that
the brain is preactivating object-level representations prior to the presentation of perceptual
input.
A recent study by Shao et al. (2021) further explored the neural correlates of semantic
prediction using fMRI. They found that brain regions are differentially recruited to support
comprehension of semantic category-specific information even before words are presented, and
that the brain regions involved may differ as a result of semantic constraint manipulations.
Specifically, they found that activity in the left inferior frontal gyrus was greater when
participants read high-constraint compared to low-constraint sentences. This is consistent with a
body of literature suggesting the inferior frontal gyrus has important roles in language production
and comprehension (e.g. Ishkhanyan et al., 2020). Interestingly, Shao et al. (2021) found greater
activity in the left supramarginal gyrus and parahippocampal place area preceding tool related
and place related words, respectively. It is well established that the parahippocampal place area
is involved in the cognitive processing of buildings and places (Epstein et al., 1999). Perhaps the
recruitment of the parahippocampal place area prior to encountering bottom-up input which
affirms the presence of place-related content can be taken to suggest that probabilistic models are
being formed by domain-specific cortical regions.

42
Grisoni et al. (2021) found that N400 potentials were more negative overall in low
constraint compared to high constraint sentences, replicating previous findings (e.g. Kutas &
Hillyard, 1984). They suggest that while the semantic predictive potential may be a neural
marker of prediction, the N400 may reflect processes related to prediction error computation. If
the brain is able to confidently and correctly predict upcoming perceptual input, prediction errors
should be low, and increase when predictions are made with less confidence. Indeed, Grisoni et
al. (2021) found a negative correlation between the amplitude of the semantic predictive
potential and the amplitude of the N400 at select electrodes.
Interestingly, León-Cabrera et al. (2019) also found that the N400 effect occurred later in
the epoch for low constraint compared to high constraint sentences. They suggest that this may
reflect facilitated processing for high constraint compared to low constraint sentences, or
alternatively, that preactivation may have occurred more readily in high constraint sentences
which allowed the brain to make a judgement more quickly about whether the preactivated
lexical content was comparable with the novel bottom-up perceptual input.
In conclusion, the SPP is a recently discovered neural marker of semantic expectancy,
which allows researchers to study the brain’s predictions of upcoming semantic content.
Although research on the semantic predictive potential is still in its infancy, it is sensitive to
manipulations of Cloze probability and appears to have neural origins which are domainspecific. Related research on the N400 suggests that the semantic predictive potential may be
sensitive to schizotypy and cannabis use.
Chapter Seven
7.1 Present Study and Experimental Hypotheses

43
Before introducing my experiment, I will briefly summarize some of the key arguments
made thus far. Firstly, schizophrenia is a complex multifaceted mental health disorder, which
shares some phenotypical characteristics with cannabis use. In recent years, cognitive
psychologists have been interested in empirically testing predictive processing theories, which
suggest that the brain interprets the world by predicting upcoming perceptual input and affirming
or adjusting its predictions upon encountering perceptual input. Some theoretical work has
suggested that people with schizophrenia may be improperly able to regulate and integrate
prediction errors with priors, which renders these people less effective at making predictions
about the state of their world. Testing the brain’s ability to predict upcoming percepts is possible
through ERP research; indeed, ERP studies have suggested that people with schizophrenia may
have difficulty incorporating contextual details when predicting upcoming content. Recently, the
semantic predictive potential has been discovered, which appears to be a neural marker sensitive
to the predictability of semantic information.
If schizophrenia results in a failure to properly regulate the precision of prediction errors
during perception, the brain may improperly assign confidence to prediction errors during
sentences, which renders its updating of probabilistic models less effective, and ultimately causes
the brain to be less confident in predicting the nature of upcoming semantic content. If high
levels of semantic constraint would normally cause the brain to “hone in” on a particular
prediction which manifests as a slow potential shift (semantic predictive potential) during
semantic processing, then in schizophrenia it is possible that the difference in this semantic
predictive potential when reading sentences of high vs low semantic constraint will be
diminished. Given this, I hypothesized that the semantic predictive potential would be modulated

44
by schizotypy, with greater reductions in the semantic predictive potential between highly
predictive (i.e. high constraint) and less predictive (low constraint) sentences.
Further, I postulated that cannabis use will influence the SPP. Consistent with the idea
that cannabis use may be associated with increased spreading activation, I speculated that regular
cannabis use would result in processing which is less sensitive to high levels of semantic
constraint. One hypothesis that stems from this reasoning is that a brain under the influence of
cannabis is “expecting” a greater range of possibilities from a sentence due to increased
spreading activation, and as a result is less “surprised” when the final word in a sentence is
inconsistent with the original semantic content. Importantly, Morgan et al. (2010) suggested that
acute cannabis use was associated with increased automatic spreading activation due to increased
priming in trials with short SOA, while chronic cannabis use was associated with increased
priming on trials with long SOA. As discussed in section 3.4, it has been argued that longer
SOAs (~750 ms) are associated with attentional and controlled processing rather than automatic
spreading activation (e.g. Morgan et al., 2010). During the sentence reading task used by LeónCabrera et al. (2019), participants are presented words of the sentence 500 ms apart, and there is
a 1000 ms pause before the presentation of the final word. Thus, the sentence reading task I
employed may be sufficient to invoke both automatic and controlled processes.
Further, I was interested in the interaction between cannabis and schizotypy, given the
similarities on some cognitive tasks between people with schizophrenia and people under the
influence of cannabis, and the high prevalence of cannabis use in populations with schizophrenia
(Koethe et al., 2006; Patel et al., 2020). This was largely exploratory; arguments could be made
that cannabis use may exacerbate abnormal prediction error signaling in people high in

45
schizotypy by further intensifying deficits in prediction error signaling. I speculated that
cannabis and schizotypy may interact to influence amplitudes of the SPP and N400.
In this study, I investigated the effects of schizotypal thought and cannabis use on neural
markers of semantic predictive processing. Specifically, I measured the extent to which people
experienced schizotypal thought and their history of cannabis use, and investigated how these
variables influenced the semantic predictive potential and the N400 on a sentence reading task
(León-Cabrera et al., 2019).
7.2 Participants
Participants (n = 18, 11 female) were recruited through UNBC’s SONA system. All
participants were enrolled in at least one eligible psychology course and received course credit
for participation. Participants ranged in age from 18-41 (M = 22.1 years). Additionally,
participants were considered eligible for the study only if English was their first language
learned. One participant’s data was excluded from the analysis due to poor quality. Another
participant’s data was excluded due to them failing to correctly answer questions in the
Schizotypal Personality Questionnaire Brief-Revised Updated (SPQ-BRU) which were designed
to ensure they were paying attention to the questions being asked. Ultimately, the data from 16
participants was used for the analysis.
7.3 Materials
Schizotypy. To measure schizotypal thought, I used the 32-item SPQ-BRU (Cohen et al.,
2010; Davidson et al., 2016). The SPQ-BR was developed to improve on the original 74-item
SPQ by reducing the time it takes for psychologists to administer, and improving completion
rates (Raine, 1991). Although other attempts have been made to shorten the length of SPQ, such
as the 22-item SPQ-Brief, the psychometric validity of the SPQ-Brief has been the subject of

46
much debate, as discussed by Cohen et al. (2010). The SPQ-BRU was designed to further
improve on the SPQ-BR by employing a consistent use of pronouns throughout the
questionnaire. The SPQ-BR has some questionnaire items using first person pronouns “I…” and
some using second-person pronouns “You…”. The SPQ-BRU changed the wording of some
items in the SPQ-BR such that every item is presented using first-person pronouns. Further, the
SPQ-BRU subdivides two of the seven subordinate factors of the SPQ, resulting in a nine-factor
subordinate solution. Davidson et al. (2016) found that the wording change improved factor
loadings of questionnaire items and decreased residual variance. The SPQ-BRU consists of 34
items, which load onto nine subordinate factors and can be represented with either three or four
factor higher-order solutions (Cohen et al., 2010). The four higher-order factors are social
anxiety, interpersonal, disorganized, and cognitive-perceptual. In the three-factor solution, social
anxiety is instead classified as a lower-order factor. The goodness of fit of the three and four
higher-order factor solutions and nine subordinate factors has since been replicated, and has been
shown to effectively fit the SPQ-BR in both Spanish and American populations (Callaway et al.,
2014; Davidson et al., 2016; Fonsesca-Pedrero et al., 2016). Davidson et al. (2016) showed that
the four-factor solution was a better model fit than the three-factor solution for the SPQ-BRU,
although the nine-factor subordinate solution was a better fit than both, and the only model with
good fit. The three and four factor higher-order solutions both had acceptable fit (Davidson et al.,
2016).
Cohen et al. (2010) report at least acceptable internal consistency for each of the seven
subfactors on the SPQ-BR (α values range from 0.70-0.86). Davidson et al. (2016) report ω
reliability values ranging from 0.69-0.88 for each of the nine subfactors of the SPQ-BRU.
Notably, Davidson et al. (2016) suggest that schizotypy varies meaningfully in healthy control

47
populations as measured by the SPQ-BRU. It has also been found that scores on SPQ-BRU
measures correlate with clinical outcomes. Davidson et al. (2016) found that the four higher
order factors of the SPQ-BRU were independently related to personal history of psychiatric
medications, psychiatric diagnoses, family history of psychiatric medication use, and family
history of psychiatric inpatient treatment. When all four higher order factors were inputted in a
logistic regression model, the model explained a significant amount of variance in psychiatric
history for each of the four variables described in the previous sentence (R2 values range from
0.112-0.205, p-values range from <.0005-.003). The variance in psychiatric history explained by
the SPQ-BRU was a motivator for my use of the questionnaire.
Cannabis Use. To measure cannabis consumption, I used the Daily Sessions, Frequency,
Age of Onset, and Quantity of Cannabis Use Inventory (DFAQ-CU) (Cuttler & Spradlin, 2017).
This inventory was designed in response to the previous lack of psychometrically sound
assessments for measuring age of onset of cannabis use, frequency of cannabis use, and quantity
of cannabis use. The questionnaire draws upon inspiration from previous cannabis use
questionnaires to construct a reliable, valid assessment of cannabis use measures. Other cannabis
use measures include the popular Timeline Followback Method (TLFB), which involves
participants selecting days on a calendar and trying to remember whether they used a substance
on that day and estimating how much of the substance they used (Sobell & Sobell, 1992;
Robinson et al., 2014). Cuttler & Spradlin (2017) acknowledge the validity and reliability of the
TLFB, but criticize limitations including recall bias, noting that people who regularly use
substances have an easier time filling in the whole calendar than people who use a substance
more sporadically. Further, the questionnaire, when adopted to measure cannabis use, asks
participants how many joints they smoked on each day. Cuttler & Spradlin (2017) argue that this

48
may lead to inaccurate estimations of the amount of cannabis being consumed, stating that only
10% of their sample reported using joints as their primary method of cannabis consumption.
People who use other methods of cannabis consumption such as bongs, edibles, or oils may have
a difficult time providing an accurate estimate of their cannabis consumption when forced to
estimate with respect to the amount of joints smoked. In the DFAQ-CU, questions are included
which ask people to estimate their amount of cannabis consumption, and specifically probe
participants on their consumption of dried cannabis including joints, consumption of edibles, and
consumption of cannabis concentrates. Additionally, the questionnaire includes a visual
depiction of dried cannabis and joints for participants to reference when estimating the quantity
of cannabis consumption. Cuttler & Spradlin (2017) note a strength of a previous assessment, the
Marijuana Smoking History Questionnaire (MSHQ), in including pictures of joints for
participants to refer to when estimating the amount of cannabis they consume (Bonn-Miller &
Zvolensky, 2009). However, as discussed by Cuttler & Spradlin (2017), the images of joints in
the MSHQ are not real images, but crude pixelated computer-generated images of joints, which
may complicate estimations of cannabis consumption. Cuttler & Spradlin (2017) further argue
that the MSHQ suffers from other drawbacks, including that the Likert scales for some items on
the questionnaire include no midpoint references (i.e., one item asks participants to rate on a 6point Likert scale how many days in their lifetime they have used cannabis, with 0 days at one
end of the scale and 300+ days at the other, but offers no indication as to how many days of
cannabis consumption should correspond to the middle points of the scale).
Another motivation for my use of the DFAQ-CU is to encourage the use of consistent
measurement standards across studies. Many studies develop their own method for measuring
cannabis consumption, which often consist of only a few short questions, limiting

49
generalizability and interpretability, and promoting indifference towards inconsistent
measurement standards. For example, Dumas et al. (2002) measured cannabis use using a short
interview in which they asked participants whether they had ever used cannabis, whether they
currently use cannabis, and if so, whether they use cannabis more than twice per week. Although
the authors found important correlations between schizotypal traits and cannabis use, a more
nuanced understanding of the relationship between cannabis use and schizotypal traits could be
gleaned by using a detailed questionnaire such as the DFAQ-CU. Further, I would argue that
when discussing and collaborating with other cannabis researchers, interpretation may be
facilitated by using consistent measures across studies.
The DFAQ-CU consists of 24 core items which load onto six factors. There are
additional items which may be used depending on responses to other questions; for example, if
participants answer ‘today’ to “When was the last time you used cannabis?”, additional questions
are incorporated to probe participants about their current level of cannabis intoxication. The six
factors are daily sessions, frequency, age of onset, marihuana quantity, concentrate quantity, and
edible quantity. Cuttler & Spradlin (2017) administered the DFAQ-CU to a sample which
consisted of 2,062 cannabis-using undergraduate students after ineligible participants were
excluded from the analysis. The results showed high convergent validity between the ‘frequency’
factor of the DFAQ-CU and the frequency measures used by the TLFB and MHSQ, which
participants also completed. r = .705 and r = .856 for the TLFB and MHSQ respectively, p <
.001 for both correlations. Additionally, the frequency factor showed high predictive validity for
cannabis use disorder symptoms and problems, which were assessed through other measures
typically used for classifying problematic cannabis use, including the Cannabis Abuse Screening
Test and Cannabis Use Disorders Identification Test Revised. Discriminant validity was assessed

50
by examining correlations with scores on the Alcohol Use Disorders Identification Test
(AUDIT). The DFAQ-CU factors frequency, age of onset, and edible quantity showed
significant correlations with the AUDIT; correlations range from r = -.267 for age of onset to r =
.295 for edible quantity, p < .001 for both correlations. Importantly, however, cannabis use has
been associated with an increased likelihood of using other substances, including alcohol (Keith
et al., 2015). Chronbach’s alpha values for internal reliability of the six factors of the
questionnaire range from α = .95 for ‘frequency’ to α = .69 for ‘daily sessions’. Cuttler &
Spradlin (2017) note that only two items load onto the ‘daily sessions’ factor, which may
contribute to its relatively questionable reliability rating.
Additionally, I administered a demographic questionnaire to participants. This
demographic questionnaire has been previously used for research studies in our lab, and assesses
standard demographic information including age, sex, and current educational attainment.
7.4 Task Sentences
Importantly, prior studies using this paradigm (e.g. León-Cabrera et al., 2017, 2019) used
Spanish sentences, so requesting the original sentence lists from the authors was unfeasible for a
study being conducted at a Canadian university. Sentences were either eight or nine words in
length. León-Cabrera et al. (2017, 2019) used sentences of only eight words; however, the
available English databases did not contain an adequate number of eight-word sentences with the
appropriate lexical characteristics to justify using exclusively eight-word sentences. Appropriate
sentences were carefully selected from two databases of sentence completion norms (Block &
Baldwin, 2010; Peelle et al., 2020). The Block & Baldwin (2010) database contained sentence
stems with matching target words and Cloze probability values for each sentence stem-target
word pairing. The Peelle et al. (2020) database contained these parameters, in addition to each

51
target word given by participants for each sentence, and the frequency with which these
responses appear in the dataset.
In my study, each target word (i.e., the final word of the sentence) was selected to appear
both in high and low constraint conditions; this will match the target words across high and low
levels of contextual constraint. For example, the target word “mail” appeared in both the highconstraint sentence “She received the acceptance letter in the mail” (Cloze probability = 0.96)
and the low-constraint sentence “Kevin found his keys hidden underneath the mail” (Cloze
probability = 0.02). The aim of using this method was to effectively control for potential
confounds including length of the target words, number of syllables, concreteness, imageability,
and frequency across the high and low constraint conditions, as the same target words appeared
in both conditions. This technique has been previously employed with successful results (Grisoni
et al., 2021; León-Cabrera et al., 2019). In addition, the target words being selected were all
nouns, and all appeared in both the MRC Psycholinguistic Database and Lancaster Sensorimotor
Norms (Coltheart, 1981; Lynott et al., 2020). The MRC Psycholinguistic Database provides
numerical ratings for words on dimensions including concreteness, imageability, familiarity, and
lexical frequency. I obtained this information on each target word both for reporting purposes as
well as controlling for these variables when I composed a list of non-congruent words (discussed
further below). The Lancaster Sensorimotor Norms provides ratings for a wide range of
sensorimotor dimensions including how much the word is experienced through different
perceptual modalities, which I collected for potential future analyses.
In addition, I controlled for another source of potential variation which does not appear to
be discussed in some of the previous literature. For their studies, León-Cabrera et al. (2017,
2019) used a set of sentences and target words obtained from a previous study performed by one

52
of the authors (Mestres-Missé et al., 2007). The sentences they use are reported as having
average Cloze probabilities of 0.760 and 0.061 for high and low constraint sentences
respectively. When Mestres-Missé et al. (2007) generated this set of sentences, they developed
three different sentences which matched each target word and differed in their level of contextual
constraint. For example, with the target word “fly”, the high constraint sentence was “The Tsetse
is a special kind of fly”, the medium constraint sentence was “With this insecticide you can’t kill
the fly”, and the low constraint sentence was “The most disgusting insect is the fly” (note that the
actual sentences used in these studies are in Spanish). When testing the Cloze probabilities of
these sentences, the authors had participants (N = 196) read each sentence and answer with the
first word which came to their mind and fit the sentence. In sentences with higher levels of
constraint, more participants were likely to converge on a single answer, resulting in high Cloze
probabilities for these sentences. The authors reported the percentage of people who responded
with the target word as the Cloze probability for that sentence, but did not discuss whether a
word other than the target word appeared more frequently than the target word itself in any
instances. For a hypothetical example, with the sentence stem “The most disgusting insect is
the”, perhaps 20% of participants responded with “fly”, and 25% responded with “cockroach”. It
appears the authors would have coded this sentence as “The most disgusting insect is the fly” (as
“fly” was the target word), with a Cloze probability of 0.20. However, this does not account for
the fact that the highest convergence upon a single word with this sentence stem is actually 0.25,
not 0.20. This seems like an important detail in studies concerned with how predictable words
are in a sentence context. Nonetheless, León-Cabrera et al. (2017, 2019) found that the level of
semantic constraint invoked by the different groups of sentences was sufficient for observing
changes in the semantic predictive potential and N400.

53
I accounted for this by collecting Cloze probabilities of both the sentence completion
with the target word and the sentence completion with the word which is most commonly given
as a final word for that sentence. To do this, I selected target words which appeared in the full
database by Peele et al. (2020), finding suitable high and low constraint sentences which fitted
the target word and recorded the Cloze probabilities for these sentences, then finding the word
which completes the sentence with the highest Cloze probability across the entire database and
recording this Cloze probability. I monitored the highest possible Cloze probability of the low
constraint sentences by ensuring the semantic constraint of sentences in the low constraint
condition was coded as the highest Cloze probability for any word following the given sentence
stem, rather than the Cloze probability value given to a specific stem-word pair. For example,
referencing the sentences in the previous paragraph where “The most disgusting insect is the fly”
has a Cloze probability of 0.20 and “The most disgusting insect is the cockroach” has a Cloze
probability of 0.25, I used the Cloze value of 0.25 even if “fly” is the final word of the sentence.
I maintained a liberal threshold (0.60) for this highest possible Cloze probability, for both
practical and theoretical reasons. Practically, there are a finite number of English sentences
available in the databases I used which met the eligibility criteria outlined earlier, and finding a
sufficient number of suitable sentences remained a challenge. Theoretically, León-Cabrera et al.
(2017, 2019) were able to find differences in the semantic predictive potential using the
sentences obtained by Mestres-Missé et al. (2007), who did not discuss this potential issue.
Additionally, waveform differences in indexes of semantic comprehension appear to be apparent
even when comparing sentences with high and medium levels of constraint. Kutas & Hillyard
(1984) found differences in N400 waveforms when comparing sentences with average Cloze
probabilities of 0.92 and 0.63. If the N400 is indeed related to the semantic predictive potential,

54
as suggested by Grisoni et al. (2021), it may be possible to detect differences in the semantic
predictive potential even when comparing sentences with high and (arguably) medium levels of
semantic constraint.
As previously mentioned, I obtained sentences from the Block & Baldwin (2010) and
Peele et al. (2020) databases, as neither database alone contains a sufficient number of
appropriate sentences for the task. Notably, only high constraint sentences are obtained from the
Block & Baldwin (2010) database, while both high and low constraint sentences are obtained
from the Peele et al. (2020) database. Unfortunately, a potential issue arises when crossreferencing the Block & Baldwin (2010) and Peele et al. (2020) databases. Block & Baldwin
(2010) instructed participants to complete sentences stems with “the single word they thought
would best complete each sentence”, while Peele et al. (2020) instructed participants to complete
sentence stems “by typing in the first word that enters your mind”. This difference in the way
participants completed sentences may have implications as a potential confound if stimuli from
both databases are being used. However, an argument may be made that the first word which
enters one’s mind and the word which best completes the sentence would likely have substantial
overlap, especially if the brain is predicting upcoming lexical content by preactivating a
representation of a specific word. Indeed, there is variation in the instructions researchers give to
participants when asking them to complete sentences in Cloze probability studies, an issue which
has been previously discussed (Staub et al., 2015). Staub et al. (2015) suggest that Cloze
probability is reflective of the relative level of activation of a word, which may be influenced by
the preceding sentence stem in addition to other top-down factors such as the broader lexical
context. They further argue that words which are the best fit for a sentence are not always easily
retrieved due to the low frequency of certain words, giving the example “To determine how fast

55
his engine was revving, the race car driver checked his tachometer”. Many subjects would
recognize tachometer as the best fit for the sentence, but may be unlikely to easily produce the
word during a Cloze probability task due to the low lexical frequency of the word. In my study,
this problem may be ameliorated to an extent by ensuring the average frequency of target words
is relatively high, which may help maximize overlap between ‘best completion’ and ‘first word
which comes to mind’. In a recent study, Van Engen et al. (2020) defined low, mid, and high
frequency target words as falling within Log HAL frequency ranges of 5.1-6.8, 6.9-9.3, and
10.0-11.9, respectively. The target words I selected have average Log HAL frequency values of
10.0, 9.9, and 9.9 for congruent, incongruent, and non-semantic words respectively.
The average cloze probabilities for the sentences in the present study are .89 for high
constraint sentences and .05 for low constraint sentences. In addition, the highest possible
average cloze probability for the low constraint sentences in the present study, as described
above, is .28. This discrepancy between these highest possible cloze probabilities and given
cloze probabilities supports my intuition that the chosen word for any reported stem-target word
pair may not necessarily reflect the word pairing associated with the highest cloze probability for
that sentence stem. León-Cabrera et al. (2017, 2019) used high and low constraint sentences with
average cloze probabilities of .76 and .06, respectively.
Table 1
Means and Standard Deviations of Potential Confounding Variables of Target Words.
Target Word

Variable

M

SD

Congruent

Imageability

557.71

61.84

Familiarity

558.51

43.60

Concreteness

537.73

86.78

56

Incongruent

Non-Semantic

Number of Letters

4.28

.81

Number of Syllables

1.04

.19

Log HAL Frequency

9.95

1.38

Imageability

560.36

66.10

Familiarity

548.79

41.99

Concreteness

556.67

77.28

Number of Letters

4.33

.79

Number of Syllables

1.05

.22

Log HAL Frequency

9.92

1.16

Imageability

550.70

57.59

Familiarity

547.48

39.60

Concreteness

542.60

63.04

Number of Letters

4.38

.59

Number of Syllables

1.05

.22

Log HAL Frequency

9.93

.67

Note. Number of Letters and Syllables, and Imageability, Concreteness, and Familiarity ratings
were obtained from the MRC Psycholinguistic Database (Coltheart, 1981). Log HAL Frequency
ratings were obtained from the English Lexicon Project (Balota et al., 2007).
Table 2
P-values From Independent Samples t-Tests Demonstrating Control for Potential Confounding
Variables of Target Words.

57
Variable

Congruent vs

Congruent vs Non-

Incongruent vs Non-

Incongruent (p)

Semantic (p)

Semantic (p)

Imageability

.79

.55

.43

Familiarity

.15

.18

.87

Concreteness

.15

.75

.32

Number of Letters

.69

.49

.72

Number of Syllables

.70

.75

.99

Log HAL Frequency

.89

.92

.98

Note. Number of Letters and Syllables, and Imageability, Concreteness, and Familiarity ratings
were obtained from the MRC Psycholinguistic Database (Coltheart, 1981). Log HAL Frequency
ratings were obtained from the English Lexicon Project (Balota et al., 2007).
7.5 Design
I created a sentence-reading task which I adapted from León-Cabrera et al. (2019). In a 2
x 2 within-subjects design, participants read sentences which varied in 1) their level of semantic
constraint and 2) the congruency of the final word of the sentence. There were three groups of
sentence stems: one group of sentences with high semantic constraint, one group with low
sematic constraint, and one group consisting of non-semantic sentences (e.g. “The goalkeeper
managed to catch the ball” (high constraint) vs “As a present she gave her son a ball” (low
constraint)). The critical difference between these two groups of sentences is the Cloze
probability, which is an indicator of how likely the sentence is to have a specific word complete
the sentence. High constraint sentences have a high Cloze probability, while low constraint
sentences have a low Cloze probability. The non-semantic sentences were created by scrambling

58
the letters in select sentences in the low-constraint condition, with the goal of creating sentences
which were phonetically plausible but semantically meaningless. For example, “As a present she
gave her son a ball” becomes “Sa a tenserp hes vage reh nos a ball”. This was the same
procedure used by León-Cabrera et al. (2019) for generating non-semantic sentences.
Importantly, the final word in the non-semantic sentences remained semantically meaningful; its
letters were not scrambled. Additionally, for each sentence in the high- and low- constraint
conditions, the final word was either congruent or incongruent with the preceding semantic
information in the sentence (e.g. “I spread butter over my toast” (congruent) vs “I spread butter
over my socks” (incongruent)). To create incongruent endings, I replaced the target word in each
sentence with a different word which makes the sentence meaning implausible. I selected
incongruent words which match the congruent target words on imageability, familiarity,
concreteness, average word length, number of syllables, and lexical frequency (León-Cabrera et
al., 2019).
Table 3
Examples of Different Types of Sentences Used in the Experiment. Adapted From León-Cabrera
et al. (2019)
Condition

Sentence Fragment

Final Word

High Constraint Congruent

Sharing combs with infected people

lice

gave Lorie
High Constraint Incongruent

Sharing combs with infected people

yolk

gave Lorie
Low Constraint Congruent

When buying used clothing always
beware of

lice

59
Low Constraint Incongruent

When buying used clothing always

yolk

beware of
Non-Semantic

Raka vener flet reh osehu thiwout ehr

watch

The experiment consisted of ten blocks, each of which contained twenty trials. Each
sentence belonged to one of five categories: high-constraint congruent, low-constraint congruent,
high-constraint incongruent, low-constraint incongruent, and non-semantic. In total, each
participant viewed 40 sentences from each of the five conditions, for a total of 200 sentences.
After each block, I employed a memory recognition test to ensure participants were attending to
the stimuli being presented, and to measure behavioral differences in semantic prediction (LeónCabrera et al., 2019). In this test, ten words were presented, half of which were target words in
the previous block and half of which were novel words to the participant. Participants were asked
to make binary judgements about whether they saw the word in the previous block by pressing
one of two keys on a keyboard.
Figure 6
Schematic of an Individual Trial From the Sentence-Reading Task. Adapted From León-Cabrera
et al. (2019)

Figure 7
Topographic Plot Showing Electrode Positions Used in the Experiment.

60

Note. Electrodes show numbers used by the Electrical Geodesics Inc. 400 system (e.g. E19) as
well as their referents in the international 10/20 system where appropriate (e.g. Pz).
7.6 Sentence-Reading Task Procedure
Before the start of each trial, a screen with an image of an eye was displayed for 2000 ms,
during which participants were encouraged to blink, in an effort to minimize blinking during the
trial itself. A fixation cross preceded the presentation of each sentence for a time period between
1350 and 1750 ms (León-Cabrera et al., 2019). Sentences were presented sequentially, one word
at a time, in white text on a black background. Each word was presented for 200 ms before

61
disappearing, with a 300 ms inter-word interval in which participants saw only the black screen.
Before the presentation of the final word of the sentence, a 1000 ms interval occurred, in which I
expected to observe the semantic predictive potential (León-Cabrera et al., 2017, 2019).
Following the presentation of the final word for 200 ms, an interval of 800 ms occurred in which
no stimuli were presented. This allowed us to investigate the N400, which is canonically affected
by the manipulation of final word congruency and is sensitive to Cloze probability (LeónCabrera et al., 2019). Figure 6 depicts the progression of one trial from the experiment.
7.7 Experimental Procedure
After consent was obtained, instructions about the sentence reading task were delivered
to participants. Participants were simply told to read the sentences carefully, and that at the end
of each block of sentences, there would be a recognition task where they would be presented
with words and would respond by pressing one of two keys to indicate whether they recalled
seeing each word in the previous block of sentences. Participants then completed a practice
version of the experiment which consisted of them reading 10 sentences while seated
approximately 1m from a computer screen (Dell P2212H computer monitor, display resolution
1920 x 1080, refresh rate 60 Hz) followed by a short recognition task. The sentences were unique
from the sentences presented in the main experimental task. Participants were then fitted with an
EEG cap before performing the sentence-reading task. After completing the sentence-reading
task, participants completed computerized versions of the demographics questionnaire, SPQBRU, and DFAQ-CU, in that order. Participants were then debriefed and given the opportunity
to address questions they may have had about the experiment. Participation in the experiment
typically took under two hours. The experiment was reviewed and approved by the University of
Northern British Columbia’s Research Ethics Board.

62
7.8 EEG Recording and Preprocessing
EEG was recorded with a 32-channel EEG cap using an Electrical Geodesics Inc. 400
system with electrodes fitted according to the standard 10/20 system (shown in Figure 7). EEG
data was sampled at 250 Hz. Care was taken to ensure electrode impedances were kept below
50kΩ wherever possible. No participants were run through the experiment with more than three
electrodes displaying above 50kΩ, and no electrodes displayed impedances of above 100kΩ for
any participants. Offline, the data was re-referenced to the average of all electrode channels.
Following this, data were visually inspected for overall quality. Electrode channels which
appeared particularly noisy were interpolated to improve data quality where possible. No more
than three electrodes were interpolated for any participant, and spatially adjacent electrodes were
not interpolated together. Data were then filtered with a high-pass Butterworth filter at 0.1 Hz
(roll-off 12 dB/oct) followed by being filtered with a low-pass Butterworth filter at 40 Hz (rolloff 12 dB/oct).
Data were epoched separately for SPP and N400 analyses. For the SPP analysis, epochs
were created which contained the time interval 1100ms preceding the presentation of the final
word of each sentence to 100ms following the presentation of the final word. For the N400
analysis, epochs were created which encompassed a time window of 200ms preceding the
presentation of the final word of each sentence to 800ms following the presentation of the final
word.
Artifact rejection was carried out in a two-step procedure, separately for SPP and N400
analyses. First, a 200ms moving window peak-to-peak (moving step = 20ms) rejection method
was used to reject epochs in which the amplitude of the moving window of the
electrooculograms exceeded ±85μV. The primary purpose of this step was to exclude epochs

63
which contained eyeblinks or notable eye movements. Following this, a simple voltage rejection
method was used to reject epochs which contained amplitudes exceeding ±200μV across all
electrodes but the electrooculograms. This allowed for the exclusion of epochs which contained
substantial amplitude deviation due to artifacts such as muscle movement or skin potentials. The
artifact rejection procedure was adopted from León-Cabrera et al. (2019). For the SPP analysis,
an average of 3.5% of trials were rejected for each participant, with the highest rejection rate
being 17% of trials for one participant. For the N400 analysis, an average of 9.9% of trials were
rejected for each participant, with the highest rejection rate being 33% of trials for one
participant.
7.9 Analysis Strategy
For the memory recognition task, mean response times and proportion of correct
responses across all trials were extracted for each subject. Median splits were performed to
divide participants on the disorganized thought factor of the SPQ-BRU as well as the frequency
factory of the DFAQ-CU to investigate whether disorganized thought or cannabis use frequency
affects response time or accuracy in the task.
ERP waveforms were plotted for visual inspection of the SPP component. Statistical
analysis for the SPP was conducted on a time window 200-0 ms preceding the final word of the
sentences. Temporally, this is a region in which the SPP has been previously observed, and upon
visual inspection of the ERP waveform, a negative deflection was observed which appeared
stronger when participants read sentences high compared to low constraint sentences in this time
window over left-frontal electrodes.
Upon visual inspection, ERP waveforms for the N400 component showed the
characteristic negative voltage deflection for incongruent compared to congruent sentence

64
endings around 350-400 ms following the presentation of the final word of the sentence. The
N400 analysis was conducted on the time window 300-500 ms following the onset of the final
word, in accordance with the time window used by León-Cabrera et al. (2019).
For the SPP analysis, mean amplitudes of the 200 ms time window preceding the
presentation of the final word of each sentence for each participant x item x electrode
combination were extracted using the ERP Measurement Tool in ERPLAB v 9.00 (LopezCalderon & Luck, 2014). The extraction process for the data used for the N400 analysis was
identical apart from the time window being 300-500 ms following the onset of the final word.
Laterality was included as a factor in the SPP analysis; the left hemisphere level was created by
combining numerical values from electrodes Fp1, F3, and F7. The right hemisphere level was
created by combining values from electrodes Fp2, F4, and F8, and the midline level was created
by combining values from electrodes Fz, 27, and 28. Frontal electrodes were selected due to the
SPP canonically displaying over frontal and fronto-central sites (e.g. León-Cabrera et al. 2019).
For the N400 analysis, electrode Pz was exclusively selected, due to the N400 component
canonically being observed over centro-parietal sites (e.g. León-Cabrera et al. 2019).
Additionally, upon visual inspection, electrode Pz showed strong evidence of an N400 effect in
our data.
A cannabis use score was calculated for each participant by first extracting their
responses on each item which loaded onto the “frequency” factor of the DFAQ-CU. Next, each
of these items was z-scored, so participants were assigned a z-scored value for each item which
loaded onto the frequency factor. Finally, the average of z-scored item scores was calculated for
each participant, which yielded a unique cannabis use score for each participant based on their

65
frequency of cannabis consumption relative to others in the sample. Figure 8 shows a histogram
of the frequency of cannabis use in our sample.
Figure 8
Histogram of Frequency of Cannabis Use.

Note. This histogram displays the different frequencies of cannabis used by participants in our
sample. Cannabis use values are based on the frequency of each participant’s cannabis use
relative to others in the sample, as described in the previous paragraph.
Scores were calculated for each participant for each higher-order factor of the SPQ-BRU.
These scores were calculated by simply averaging the scores for all items which loaded onto
each factor of the questionnaire. We chose to specifically include the disorganized thought factor

66
in our analysis due to previous research suggesting that neural markers of abnormal semantic
processing in schizophrenia may be associated with the disorganized thought characteristic of
schizophrenia rather than schizophrenia itself (Pomarol-Clotet et al., 2008). Figure 9 shows a
histogram of the frequencies of participant levels of disorganized thought in our sample.
Figure 9
Histogram of Frequency of Disorganized Thought

Note. This histogram displays the frequency of participants in our sample experiencing different
levels of disorganized thought. Disorganized thought scores were calculated by averaging the
scores for all items which loaded onto the disorganized thought factor of the SPQ-BRU. Possible
values range from 0.0-5.0.

67
Statistical analyses were performed using R Statistical Software (v4.3.0, R Core Team
2023). For the SPP analysis, linear mixed effects modelling (LME) was used to estimate fixed
effects of hemisphere (left, right, central), sentence constraint (high, low, non-semantic),
disorganized thought (continuous), cannabis frequency (continuous) and all interactions. For the
N400 analysis, LME was used to estimate fixed effects of congruency (congruent, incongruent),
sentence constraint (high, low), disorganized thought (continuous), cannabis frequency
(continuous) and all interactions. Models included intercepts for random effects of participants
and stimuli. R2 values are included in text. Significant main effects and interactions are reported
(p < .05). The linear mixed-effects model fitting procedure was conducted using the lme4
package (v. 1.1.33). The ANOVA table was produced using the LMERConvenienceFunctions
package (v. 3.0). R2 and power calculations were conducted using the simr (v. 1.0.7) and MuMln
(v. 1.47.5) packages. The emmeans (v. 1.8.6) package was used to visualize interactions.
Chapter Eight
8.1 Memory Recognition Task Results
Overall, accuracy for the word recognition tasks was high, with participants correctly
classifying 83.2% of words when being asked to judge if the word had been shown in the
previous block of trials. A median split was preformed to divide participants on the disorganized
thought factor of the SPQ-BRU to investigate whether disorganized thought affects response
time or accuracy. An independent samples Welch’s t-test showed no significant difference in
response accuracy between the groups of high (M = .85, SD = .09) and low (M = .82, SD = .04)
disorganized thought, t (9.79) = .84, p = .42. Further, the difference in response time was not
significant between the high (M = 958.70, SD = 186.1) and low (M = 821.44, SD = 181.98)
disorganized thought groups, t (13.99) = 1.49, p = .16. A median split on frequency of cannabis

68
use was also performed; there was no significant difference in response accuracy between high
(M = .84, SD = .09) and low (M = .82, SD = .05) cannabis users, t (10.62) = .61, p = .55. Further,
there was no significant difference in response time between high (M = 899.86, SD = 157.66)
and low (M = 880.28, SD = 231.18) cannabis users, t (12.35) = .20, p = .85.
Table 4
Correlation Matrix Showing r Values Between Higher-Order Factors of the SPQ-BRU and
Cannabis Frequency (Taken From the DFAQ-CU) in Our Sample.
Variable

Disorganized

Disorganized

Cognitive-

Inter-

Social

Cannabis

Thought

Perceptual

Personal

Anxiety

Frequency

1.00

.44

-.05

.24

.27

1.00

-.17

.02

.45

1.00

.02

.18

1.00

.71

Thought
CognitivePerceptual
InterPersonal
Social
Anxiety
Cannabis

1.00

Frequency

8.2 Semantic Predictive Potential Waveforms and Scalp Maps
Waveforms for averages of high, low, and non-semantic conditions over electrode F3 are
shown in Figure 10. Notably, negativity begins to become more prominent in the high constraint

69
condition compared to the low constraint and non-semantic conditions around 200 ms prior to
the presentation of the final word, as highlighted by the arrows drawn to different waveforms.
This trend of the waveforms for high constraint sentences (shown in red) eliciting greater
negativity than low constraint and non-semantic conditions is visible across multiple electrodes
in Figure 11, which shows waveform plots for all electrodes. In particular, electrodes E17 (Fz)
and E4 (F4), which are located over midline and right hemispheres, respectively, and were used
in the SPP analysis, display this pattern. This difference is also visualizable in Figure 12, which
shows scalp maps of high and low constraint and non-semantic conditions. Around -100 ms,
darker colours, which indicate greater negativity, become apparent over frontal and frontocentral electrodes in the high constraint condition compared to the low constraint and nonsemantic conditions.
Figure 10
Semantic Predictive Potential Waveforms Observed Over Electrode F3.

70

Note. Different waveforms are displayed for high and low constraint and non-semantic
conditions, averaged across stimuli and participants. The temporal area highlighted in green was
used for SPP analysis (200 ms preceding the presentation of the final word to the presentation of
the final word). In particular, the waveform shown in red displays a greater negativity around the
-200 ms mark until the final word in each sentence was presented. Arrows draw attention to the
difference in waveform patterns elcicited by people reading high compared to low constraint
sentences. The temporal presentation of the final word in sentences is indicated by the black line
across the Y-axis.
Figure 11
Epoch Visualizing the Semantic Predictive Potential Across all Electrodes 1100 ms Preceding
the Presentation of the Final Words in Sentences of High, Low, and Non-Semantic Constraint.

71

72
Note. Over multiple electrodes, including E3, E4, E5, and E17, the pattern of greater negativity
elicited by high constraint compared to low constraint and non-semantic conditions is apparent,
becoming noticeable around 200 ms prior to the presentation of the final words in sentences
(temporal presentation of the final word in sentences is indicated by the black line across the Yaxis).
Figure 12
Scalp Maps of SPP Scalp Topographies.

73

74
Note. Scalp maps display voltage at 50ms intervals 300 ms preceding the presentation of the final
word in sentences to 50 ms following the presentation of the final word in sentences. Darker
colours indicate greater negativity. Around -100 ms, greater negativity becomes apparent over
frontal and fronto-central sites for the high constraint condition compared to the low constraint
and non-semantic conditions.
8.3 Semantic Predictive Potential Analysis
Table 5 shows an ANOVA table with estimated p-values and effect sizes, reported as %
deviance, for all effects and interactions on SPP amplitude. Results from the LME analysis on
the SPP revealed a main effect of hemisphere, F (28764, 28548) = 23.74, p < .001, explained
deviance (%) = 0.16. Follow-up tests revealed that negativity was greater over the left
hemisphere than the right hemisphere, t (28554) = 4.24, p < .001. Negativity was also greater
over the midline than the right hemisphere, t (28554) = 5.36, p < .001. There was no significant
difference in negativity between the left hemisphere and midline, t (28554) = 1.12, p = .502.
A main effect of constraint was also present, F (28764, 28548) = 20.80, p < .001,
explained deviance (%) = 0.14. Follow-up tests revealed that negativity was greater in high
constraint than low constraint conditions, t (28565) = 7.45, p < .001. Negativity was also greater
in high constraint than non-semantic conditions, t (256) = 3.53, p = .001. The difference in
negativity between low constraint and non-semantic conditions was not significant, t (256) = .56,
p = .844.
There was a significant interaction of constraint and disorganized thought, F (28764,
28548) = 9.85, p < .001, explained deviance (%) = 0.07. This interaction is shown below in
Figure 13. There was also a significant three-way interaction of constraint, cannabis use, and
disorganized thought, F (28764, 28548) = 11.04, p < .001, explained deviance (%) = 0.07,

75
displayed below in Figure 14. There was also a significant interaction of hemisphere and
disorganized thought, F (28764, 28548) = 4.84, p = .008, explained deviance (%) = 0.03. This
interaction is shown below in Figure 15. Finally, there was a significant three-way interaction of
hemisphere, cannabis use, and disorganized thought, F (28764, 28548) = 4.74, p = .009,
explained deviance (%) = 0.03. This interaction is shown below in Figure 16. Overall, the fixed
effects in the model explained a small amount of the variation in SPP amplitude, R2 = .006.1
Table 5
Results From LME Analysis on SPP Amplitude 200-0 ms Preceding the Presentation of Target
Words.
Effect

F

df

Value

upper

Hemisphere

23.74

28764

28548

<.001***

.1606

Constraint

20.79

28764

28548

<.001***

.1407

Cannabis frequency

0.02

28764

28548

.897

.0001

Disorganized thought

0.05

28764

28548

.829

.0002

Hemisphere x Constraint

1.19

28764

28548

.312

.0161

Hemisphere x Cannabis frequency

1.17

28764

28548

.311

.0079

Constraint x Cannabis frequency

0.73

28764

28548

.481

.0050

Hemisphere x Disorganized thought

4.84

28764

28548

.008**

.0327

Constraint x Disorganized thought

9.85

28764

28548

<.001***

.0666

1

df lower

p value

Explained
deviance (%)

We performed post-hoc power estimates for highest-order interactions using the powerSim function from the simr
package in R. For the three-way interaction of hemisphere, cannabis use, and disorganized thought, the power was
100.0%, 95% CI [96.4 100.0]. For the three-way interaction of constraint, cannabis use, and disorganized thought,
the power was 100.0%, 95% CI [96.4 100.0].

76
Cannabis frequency x Disorganized

0.72

28764

28548

.398

.0024

1.07

28764

28548

.368

.0145

1.52

28764

28548

.192

.0206

4.74

28764

28548

.009**

.0321

11.04

28764

28548

<.001***

.0747

0.43

28764

28548

.789

.0058

thought
Hemisphere x Constraint x Cannabis
Frequency
Hemisphere x Constraint x
Disorganized thought
Hemisphere x Cannabis frequency x
Disorganized thought
Constraint x Cannabis frequency x
Disorganized thought
Hemisphere x Constraint x Cannabis
frequency x Disorganized thought
***p < .001, ** p < .01
Constraint x Disorganized Thought
Figure 13
Visualization of the Interaction Between Constraint and Disorganized Thought on SPP
Amplitude.

77

Note. High and low disorganized thought in all figures represents estimates at the extreme values
of the range of the disorganized thought variable collected in our sample. Error bars represent the
95% confidence interval containing the population mean at different levels of constraint after
accounting for fixed effects of disorganized thought and random effects of participants and
stimuli. Estimates are based on averages 200 ms preceding the presentation of the final word
until the presentation of the final word (0 ms).

78
Figure 13 shows the interaction between disorganized thought and constraint on SPP
amplitudes. At low levels of disorganized thought, negativity was greater for high constraint
sentences compared to low constraint sentences, t (28672) = 5.20, p < .001. At low levels of
disorganized thought, negativity was also greater for high constraint sentences compared to nonsemantic sentences, t (5865) = 5.96, p < .001. At high levels of disorganized thought, no
differences between high and low constraint or non-semantic sentences reached statistical
significance.
Constraint x Disorganized Thought x Cannabis Frequency
Figure 14
Visualization of the Interaction between Constraint, Cannabis Frequency, and Disorganized
Thought on SPP amplitudes.

79

Note. High and low disorganized thought and cannabis use in all figures represent estimates at
the extreme values of the ranges of the disorganized thought and cannabis use variables collected
in our sample. Error bars represent the 95% confidence interval containing the population mean
at different levels of hemisphere after accounting for fixed effects of disorganized thought and

80
cannabis use and random effects of participants and stimuli. Estimates are based on averages 200
ms preceding the presentation of the final word until the presentation of the final word (0 ms).
Figure 14 shows the interaction between cannabis use, constraint, and disorganized
thought. In all cases, the trend was that negativity was greater for high compared to low
constraint sentences, except when both cannabis use and disorganized thought were high. When
cannabis use and disorganized thought were both high, negativity was significantly greater for
low constraint compared to high constraint sentences, t (28601) = 3.03, p = .007. When cannabis
use and disorganized thought were both low, the difference in negativity between high and low
constraint sentences was not significant, t (28401) = 1.78, p = .18. When disorganized thought
was low and cannabis use was high, negativity was significantly greater for high constraint
compared to low constraint sentences, t (28713) = 4.58, p < .001. Likewise, when disorganized
thought was high and cannabis use was low, negativity was significantly greater for high
constraint compared to low constraint sentences, t (28711) = 4.15, p < .001. All post-hoc
comparisons of experimental conditions across levels of high and low disorganized thought and
cannabis use are reported in Appendix B.
Hemisphere x Disorganized Thought
Figure 15
Visualization of the Interaction Between Hemisphere and Disorganized Thought on SPP
Amplitude.

81

Note. High and low disorganized thought in all figures represents estimates at the extreme values
of the range of the disorganized thought variable collected in our sample. Error bars represent the
95% confidence interval containing the population mean at different levels of disorganized
thought after accounting for fixed effects of hemisphere and random effects of participants and

82
stimuli. Estimates are based on averages 200 ms preceding the presentation of the final word
until the presentation of the final word (0 ms).
Figure 15 shows the interaction between hemisphere and disorganized thought. Over the
left hemisphere, amplitudes were greater at high compared to low levels of disorganized thought.
This pattern was reversed over the right hemisphere, where amplitudes were greater at low
compared to high levels of disorganized thought. Over midline electrodes, the pattern was the
same as at right hemisphere electrodes, although the magnitude was decreased.
Hemisphere x Cannabis Frequency x Disorganized Thought
Figure 16
Visualization of the Interaction Between Hemisphere, Cannabis Frequency, and Disorganized
Thought on SPP Amplitude.

83

Note. High and low disorganized thought and cannabis use in all figures represent estimates at
the extreme values of the ranges of the disorganized thought and cannabis use variables collected
in our sample. Error bars represent the 95% confidence interval of the population mean at
different levels of disorganized thought after accounting for fixed effects of cannabis use and

84
hemisphere and random effects of participants and stimuli. Estimates are based on averages 200
ms preceding the presentation of the final word until the presentation of the final word (0 ms).
Figure 16 shows the interaction between cannabis frequency, hemisphere, and
disorganized thought. The patten at midline and left hemisphere electrodes was consistent, but
different than the pattern at right hemisphere electrodes. At midline and left hemisphere
electrodes, when disorganized thought was low, negativities were greater for low cannabis
compared to high cannabis users. Also at midline and left hemisphere electrodes, when
disorganized thought was high, negativities were greater for high compared to low cannabis
users. However, at right hemisphere electrodes, the discrepancy in negativities between low and
high cannabis users at low and high levels of disorganized thought was reduced compared to
midline and left hemisphere electrodes. In addition, over right hemisphere electrodes when
disorganized thought was high, negativities were marginally greater for low compared to high
cannabis users. Interestingly, the pattern of negativity being lower at midline compared to left
and right hemisphere electrodes was consistent throughout, with two exceptions. When
visualized this way, no post-hoc comparisons in this three-way interaction reached statistical
significance (p < .05).
In sum, we found that when people were low in disorganized thought, the SPP effect
observed by León-Cabrera et al. (2017, 2019) was replicated in that SPP negativities were
greater for high constraint compared to low constraint sentences. However, when people were
high in disorganized thought, the SPP effect was not apparent. This was driven by the three-way
interaction between constraint, disorganized thought, and cannabis use. When people were high
in both cannabis use and disorganized thought, the SPP effect actually reversed direction;

85
negativities were greater for low constraint compared to high constraint sentences. We also
found that hemisphere interacted with disorganized thought and cannabis use.
8.4 N400 Waveforms and Scalp Maps
Figure 17 shows N400 waveforms for the different experimental conditions at electrode
Pz. The area highlighted in green was used for statistical analysis. Notably, the high constraint
congruent condition displayed less apparent negativity than all other waveforms. This trend can
be visualized across multiple parietal and centro-parietal electrodes in Figure 18. Further, Figure
19 shows scalp maps where greater negativities are apparent for the incongruent condition
compared to the congruent condition, beginning around 300 ms following the presentation of
final words in sentences until approximately 500 ms following the presentation of final words in
sentences.
Figure 17
N400 Waveforms Observed Over Electrode Pz.

86

Note. Waveforms are displayed for high constraint congruent, high constraint incongruent, low
constraint congruent, low constraint incongruent, and non-semantic sentences, averaged across
all stimuli and participants. The temporal area highlighted in green was used for N400 analysis
(300-500 ms following the presentation of the final word in sentences). The arrow shows the
high constraint congruent waveform, where N400 negativity is less pronounced compared to
other experimental conditions. The temporal presentation of the final word in sentences is
indicated by the black line across the Y-axis.
Figure 18
Epoch Visulalizing the N400 Potential Across all Electrodes.

87

88
Note. Waveforms show the temporal window from 200 ms preceding the presentation of the
final words in high constraint congruent (HCC), high constraint incongruent (HCI), low
constraint congruent (LCC), low constraint incongruent (LCC) and non-semantic (NS) sentences
to 800 ms following the presentation of final words in sentences. Over multiple electrodes,
including E6, E7, E8, and E19, the pattern of reduced negativity for HCC compared to all other
conditions is apparent, beginning around 300 ms following the presentation of the final words
until approximately 500 ms following the presentation of final words (temporal presentation of
the final word in sentences is indicated by the black line across the Y-axis).
Figure 19
Scalp Maps Showing N400 Scalp Topographies.

89

90
Note. Scalp maps display negativities at 100 ms intervals from 0-600 ms following the
presentation of the final word in sentences. Darker colours indicate greater negativity. Notably,
greater negativities become apparent over parietal and centro-parietal sites for incongruent
compared to congruent words around 300 ms following the presentation of final words. This
trend remains consistent until approximately 500 ms following the presentation of final words.
8.5 N400 Analysis
Table 6 shows an ANOVA table with estimated p-values and effect sizes, reported as %
deviance, for all effects and interactions on N400 amplitude. Results from the LME analysis on
the N400 revealed a main effect of constraint, F (2544, 2368) = 12.28, p < .001, explained
deviance (%) = .46. Negativity was significantly greater for low constraint compared to high
constraint sentences, t (2376) = 3.36, p < .001. There was also a main effect of congruency, F
(2544, 2368) = 13.10, p < .001, explained deviance (%) = 0.49. Negativity was significantly
greater for incongruent compared to congruent sentence endings, t (199) = 4.06, p < .001. There
was a significant interaction of constraint and congruency, F (2544, 2368) = 5.97, p = .01,
explained deviance (%) = 0.22. This interaction is shown below in Figure 20. There was also a
significant interaction of congruency and cannabis frequency, F (2544, 2368) = 4.09, p = .04,
explained deviance (%) = 0.15. This interaction is shown below in Figure 21. Overall, the fixed
effects in the model explained a small amount of the variation in N400 amplitude, R2 = .02.2
Table 6
Results From LME Analysis on N400 Amplitude 300-500 ms Following the Presentation of
Target Words.

2

We performed post-hoc power estimates for highest-order interactions using the powerSim function from the simr
package in R. For the interaction of congruency and constraint, the power was 86.0%, 95% CI [77.6 92.1]. For the
interaction of congruency and cannabis use, the power was 98.0%, 95% CI [93.0 99.8].

91
Effect

F

df

df lower

p value

Explained

Value

upper

Constraint

12.28

2544

2368

<.001***

.46

Congruency

13.10

2544

2368

<.001***

.49

Cannabis frequency

.02

2544

2368

.89

.0008

Disorganized thought

3.41

2544

2368

.07

.13

Constraint x Congruency

5.97

2544

2368

.01*

.22

Constraint x Disorganized thought

.60

2544

2368

.44

.02

Congruency x Disorganized thought

.96

2544

2368

.33

.04

Constraint x Cannabis frequency

.77

2544

2368

.38

.03

Congruency x Cannabis frequency

4.09

2544

2368

.04*

.15

Cannabis frequency x Disorganized

.53

2544

2368

.46

.02

.01

2544

2368

.91

.0005

3.11

2544

2368

.08

.11

.05

2544

2368

.83

.002

3.75

2544

2368

.05

.14

1.19

2544

2368

.27

.05

deviance (%)

thought
Constraint x Congruency x
Disorganized thought
Constraint x Congruency x Cannabis
frequency
Constraint x Cannabis frequency x
Disorganized thought
Congruency x Cannabis frequency x
Disorganized thought
Congruency x Constraint x Cannabis
frequency x Disorganized thought

92
***p < .001, ** p < .01, * p < .05
Constraint x Congruency
Figure 20
Visualization of Constraint x Congruency Interaction for N400 Amplitudes Over Electrode Pz.

Note. Error bars represent the 95% confidence interval containing the population mean at
different levels of constraint after accounting for fixed effects of congruency and random effects
of participants and stimuli. Estimates are based on averages 300 ms following the presentation of
the final word until 500 ms following the presentation of the final words.

93
Figure 20 shows the effect of the constraint x congruency interaction on the N400. At the
level of high constraint, N400 amplitude was more negative for incongruent endings compared to
congruent endings, t (674) = 4.25, p < .001. At the level of low constraint, the difference in N400
amplitude between congruent endings and incongruent endings was not significant, t (674) =
1.60, p = .38. Thus, the strength of the N400 effect is greater in high compared to low constraint
contexts. P-values for all post hoc comparisons are reported in Appendix B.
Congruency x Cannabis Frequency
Figure 21
Visualization of Congruency x Cannabis Frequency Interaction for N400 Amplitudes Over
Electrode Pz.

94

Note. High and low cannabis use in all figures represents estimates at the extreme values of the
range of the cannabis use variable collected in our sample. Error bars represent the 95%
confidence interval containing the population mean at different levels of congruency after
accounting for fixed effects of cannabis use and random effects of participants and stimuli.
Estimates are based on averages 300 ms following the presentation of the final word until 500 ms
following the presentation of the final word.

95
Figure 21 shows the interaction between cannabis use and congruency. When cannabis
use was low, the N400 effect was not significant, t (503.8) = 1.52, p = .43. However, when
cannabis use was high, there was a classic N400 effect in which N400 potentials were more
negative in response to incongruent compared to congruent words, t (2170.0) = 3.94, p < .001. Pvalues for post-hoc comparisons across levels of cannabis use and congruency are reported in
Appendix B. In sum, the strength of the N400 effect increased as cannabis use increased.
Chapter Nine
9.1 General Discussion
In this study, we investigated the effects of schizotypal thought and cannabis use on
neural activity while participants were engaged in a sentence-reading paradigm. Theoretically,
the SPP and N400 may be indices of predictive processing during language comprehension, and
may reflect comparisons of predictions with sensory input and computation of prediction errors.
For the purposes of my discussion, I refer to the discrepancy in voltages elicited by high and low
sentences leading up to the final word of these sentences as the SPP effect. Recall that the N400
effect refers to discrepancies in voltages when comparing congruent and incongruent endings to
sentences. People with schizophrenia display abnormalities in neural markers thought to be
related to predictive processing, and it may be that the precision with which prediction errors are
compared to existing models of the world is improperly regulated in schizophrenia. We
hypothesized that aberrant prediction error computation in schizophrenia would result in reduced
SPP and N400 effects for people high in schizotypal traits. Further, we speculated that high
levels of cannabis use may also result in reduced SPP and N400 effects, and may interact with
schizotypy to influence the SPP and N400 potential. For the purposes of this discussion, I choose
to refer to disorganized thought rather than schizotypy where appropriate, as our results hinged
on the disorganized thought factor of the SPQ-BRU.

96
Firstly, our prediction about the interaction between disorganized thought and sentence
constraint was supported, as shown in Figure 13. People high in disorganized thought displayed a
reduced SPP effect compared to people low in disorganized thought. This provides support for
our hypothesis that aberrant prediction error computation may underlie cognitive deficits in
people with schizophrenia. When a person reads a sentence, prediction errors are constantly
being generated and compared with prior models, which become iteratively updated as the
person reads more of the sentence. Thus, the model someone has about the meaning of a
sentence is constantly refined during reading as a result of comparisons between predictions and
errors. If prediction errors during sentence reading in schizophrenia are not able to influence
change on prior models as they would in a neurotypical person due to improper regulation of the
precision of prediction errors, it may be the case that people with schizophrenia rely on models
which are improperly or insufficiently updated, and these models account for less variability in
the state of the world. This results in their brains making predictions which “explain” less of the
incoming sensory information, causing them to rely more heavily on the influence of prediction
errors generated by comparisons between incoming sensory information and prior models.
However, because the precision of these prediction errors inappropriately regulated, the model is
insufficiently adjusted, and the brain remains unable to develop sufficient high-level models
capable of making precise predictions about incoming sensory information.
We posit that the difference between high- and low-level predictions is important to
consider when discussing the possible influence of schizotypy on the SPP effect. If a person high
or low in schizotypy is attempting to create a model to predict the upcoming semantic content of
sentences, we suggest that it is unlikely they will successfully and accurately predict each word
which follows from the onset of the sentence. For instance, consider if someone low in

97
schizotypy were to read the sentence “The goalkeeper managed to catch the ball”. After reading
the sentence fragment “The goalkeeper managed to” it is unlikely their brain would be predicting
the specific word “catch”, as numerous other words could fit here (e.g., kick, stop, prevent). As
such, prediction error would be generated at lower levels of the predictive hierarchy upon
encountering the word “catch”, as the unexpected low-level properties of the word catch (e.g. the
lines and shapes that construct the word “catch”) would be novel information which needs to be
incorporated into the model. Further, the meaning of the word “catch” would need to be
incorporated at higher levels. However, this word “catch” is still plausible under high-level
predictions of the types of things that a goalkeeper might do, and thus this prediction error is
resolved at low levels of the predictive hierarchy, and does not necessitate a radical overhaul of
the type of high-level model someone might be developing about the actions of a goalkeeper in a
particular scenario.
If someone were high in schizotypy, however, and read the sentence fragment “The
goalkeeper managed to”, they would again develop low-level prediction errors, as there would be
little reason to predict the specific word “catch” over the numerous other contenders offered
earlier (e.g. kick, stop, prevent). However, because of the deficits in regulating the precision of
prediction errors, the incorporation of this prediction error into the predictive model may not
proceed as it would in someone low in schizotypy. If the prediction error was weighted with too
much precision, the error may not be resolved at low levels, and instead may cause changes to
the developing high-level model about what a goalkeeper in this situation might do. This may
prevent high-level models from developing the ability to confidently predict the upcoming highlevel semantic content of the remainder of the sentence. Conversely, if the precision of this
prediction error was too low, lower levels of the predictive hierarchy may not readily incorporate

98
the new information about the word “catch” into the lower-level model, resulting in a failure to
make use of the available information, again ultimately preventing accurate and confident
predictions from being made at higher levels of the predictive hierarchy due to suboptimal
integration of the available information.
It is possible that the observed SPP voltage difference between high and low constraint
sentences, which may represent the degree to which people are predicting the ending of
sentences, is indicative of the extent to which high-level predictive models of the content of the
sentence are developed. A reduced SPP effect in people high in schizotypy may reflect that the
difference in the development of high-level predictive models as a function of the available
information is reduced compared to people low in schizotypy. In sum, we posit that differences
in the SPP effect between people high and low in schizotypy may be a result of improper
regulation of the precision of prediction errors at lower levels of the predictive hierarchy,
resulting in ineffective and poorly developed high-level predictions being formed.
Intriguingly, the two-way interaction of disorganized thought and sentence constraint
appeared to be qualified over by people’s level of cannabis use, as shown in Figure 14. When
people were high in disorganized thought but did not use cannabis frequently, they displayed the
canonical SPP effect seen in people low in disorganized thought. Further, when people were low
in disorganized thought and either used or did not use cannabis frequently, the directionality of
the SPP effect remained consistent with reports from previous literature (León-Cabrera et al.,
2019). Notably, the strength of the SPP effect was greater in people low in disorganized thought
who did not use cannabis than people low in disorganized thought who did use cannabis. If
people have thought which is typically organized, regular cannabis use may cause the brain to

99
improperly weight the precision of prediction errors, resulting in high-level predictions which are
less developed.
However, when people were high in both disorganized thought and cannabis use, the
canonical SPP effect was not observed (in fact, the direction of the SPP effect reversed). We
speculate that this combination of disorganized thought and cannabis use causes the brain to
experience further difficulty in properly regulating the precision of prediction errors. Indeed, it
has been suggested that regular cannabis use may contribute to the development of formal
thought disorder (e.g. Argote et al., 2022). Just as people with schizophrenia may be prone to
cannabis use, people high in disorganized thought may be prone to aberrant prediction error
signaling during language comprehension, and cannabis use may facilitate this phenomenon.
Of particular interest is the potential role of pyramidal cells in this association between
cannabis use, schizotypy, and prediction error signaling during language comprehension. Indeed,
it has been proposed that pyramidal cells are responsible for comparing top-down predictions
with bottom-up input (Friston, 2016). Firstly, it has been demonstrated that pyramidal cells
display alterations both in people with schizophrenia and in cannabis users (Glantz et al., 2000;
Miller et al., 2019). Glantz et al. (2000) found that dendritic spine density in layer three
pyramidal cells was decreased in the dorsolateral prefrontal cortex of patients with
schizophrenia. Miller et al. (2019) showed that adolescent cannabis exposure was associated with
reduced dendritic spine density in layer three pyramidal cells in rats. Friston et al. (2016) suggest
that higher-level representations are compared in deep pyramidal cells while lower-level
representations are compared in superficial pyramidal cells. It is possible that the combination of
schizophrenia and cannabis use causes particular abnormalities in dendritic spine density of
prefrontal neurons which underlies deficits in prediction error signaling during language

100
comprehension. Indeed, our data showed that schizotypy and cannabis use altered SPP effects
over frontal and prefrontal sites, consistent with the idea that prefrontal neurons are implicated in
prediction error computation during language processing.
We postulate that the combination of high schizotypy and cannabis use results in specific
alterations to dendritic spines of prefrontal pyramidal neurons involved in comparisons between
prior models and bottom-up input, which ultimately cause model overfitting in instances where
bottom-up information is noisy. Our results showed that people high in both cannabis use and
schizotypy exhibited more negative SPPs to low constraint and non-semantic sentences
compared to high constraint sentences. If people high in schizotypy regularly use cannabis,
prediction errors which result in noisy environments (e.g. when sentences are low in semantic
constraint or are non-semantic) may be integrated into models in such a way that the model is
generating a prediction associated with a more highly developed and vivid representation
compared to the prediction made by a neurotypical person or a person who does not regularly use
cannabis.
This line of reasoning is somewhat consistent with our observation of increased N400
effects at high levels of cannabis use, shown in Figure 21. This result did not support our initial
hypothesis that higher cannabis use would lead to reduced discrepancies in N400 amplitudes.
Further, our finding is inconsistent with literature which suggests that regular cannabis users tend
to display equal or reduced effects of neural markers related to change and expectation (e.g.
Kiang et al., 2013; Greenwood at al., 2014). Instead, if high cannabis use results in increased
N400 effects, it may be the case that greater prediction errors are generated in instances where a
highly developed prediction has been formed (e.g. in instances where high cannabis use and high
disorganized thought contribute to strong predictions being made about low constraint sentences)

101
and high-level models need to be adjusted based on unexpected incoming sensory information
(an incongruent sentence ending). However, a caveat worth mentioning is that greater N400
effects associated with high cannabis use appeared more prominent at low rather than high levels
of disorganized thought, although this interaction did not reach statistical significance (p = .053).
Some other findings are worth noting. First, we found that amplitudes were more
negative overall for left and midline electrodes compared to right hemisphere electrodes during
the SPP epoch. León-Cabrera et al. (2019) noted that hemisphere interacted with semantic
constraint, such that the SPP effect was most pronounced over left hemisphere electrodes
compared to both right hemisphere and midline electrodes. While our hemisphere effect did not
interact significantly with constraint, we replicate a main effect of hemisphere observed by LeónCabrera et al. (2019).
Second, we found a significant two-way interaction between hemisphere and
disorganized thought in our SPP analysis, where amplitudes were more negative at high
compared to low levels of disorganized thought over left hemisphere electrodes. Conversely, the
pattern was reversed over midline and right hemisphere electrodes, where amplitudes were more
negative at low compared to high levels of disorganized thought.
Third, we report a significant three-way interaction of cannabis use, disorganized
thought, and hemisphere, displayed in Figure 16. We show that discrepancies in SPP negativities
between high and low levels of cannabis use at high and low levels of disorganized thought are
more pronounced over left hemisphere and midline electrodes compared to right hemisphere
electrodes. If the SPP is more sensitive to different levels of schizotypy and cannabis use over
left compared to right hemisphere electrodes, this is consistent with the idea that left prefrontal
areas are involved in generation of the SPP (e.g. Grisoni et al., 2021; León-Cabrera et al., 2019).

102
Importantly, other studies have shown that fronto-central areas, in addition to left frontal areas,
show discrepancies in neural responses to high vs low constraint sentences (León-Cabrera et al.,
2017).
9.2 Replications of Previous Findings
There are two important replicated findings. First, we replicated previous research
suggesting that high constraint sentences elicit a greater negativity 200-0 ms before the
presentation of the final word compared to both low constraint sentences and non-semantic
sentences (León-Cabrera et al., 2019). That is, we replicated the SPP effect overall. Additionally,
the comparison in means between low constraint and non-semantic sentences was not significant
during this time period, consistent with previous findings from León-Cabrera et al. (2019). Our
research, which explored the recently popularized SPP, using a novel stimulus set created by the
author, suggests that effects of the SPP may be generalizable across languages and stimulus sets.
We also replicate the canonical N400 findings where negativity is greater for incongruent
compared to congruent sentence endings, and greater for low compared to high constraint
sentences (e.g. Kutas & Hillyard, 1984; León-Cabrera et al., 2019). Further, we showed that
when sentences were high in semantic constraint, the discrepancy in N400 amplitudes when
comparing congruent and incongruent endings was significant, but this discrepancy was not
significant when sentences were low in semantic constraint, shown in Figure 20. This supports
the hypothesis that the brain is sensitive to the extent to which the nature of upcoming semantic
content is predictable. When a sentence is highly predictable, but an unexpected final word is
presented, the brain must generate a large prediction error to adjust its model to account for the
novel and unexpected stimulus. However, if the ending of a sentence is less predictable, the brain
need not generate as large a prediction error if presented with an unexpected ending, since the

103
model was not initially developed with the same amount of confidence as a model for the
outcome of a sentence with high constraint.
In sum, we found that cannabis use and schizotypy interact to influence the SPP effect,
and that the combination of high schizotypy and high cannabis use results in strong predictions
being made when contextual semantic information is ambiguous. It may be that this interaction
of cannabis use and schizotypy affects pyramidal neurons, causing highly developed predictions
to be made in noisy environments. We also found that high cannabis use results in increased
N400 effects, which may reflect greater prediction errors being generated and computed in
instances where a strong prediction made in a noisy environment is incompatible with incoming
sensory input. We also replicate previous research suggesting that constraint and hemisphere
influence the SPP and that congruency and constraint influence the N400.
9.3 Limitations
There are a number of limitations to the present study. Firstly, a limitation of our study
was the notably non-normal distribution of cannabis use, shown in Figure 8. Many participants in
our sample reported having never used cannabis, resulting in a distribution which was skewed to
the right, with some instances of high cannabis users on the right side of the distribution. Thus,
our distribution more closely approximated a U-shape than a normal curve. This affects the
conclusions and generalizations we might make about SPP and N400 effect alterations in
cannabis users. However, it is notable that distributions of the use of substances such as coffee
and alcohol are often non-normal (Sainani, 2012).
Second, as previously mentioned, task sentences were chosen from two unique databases,
as there were an insufficient number of appropriate sentences in either of the databases alone.
Only high constraint sentences were obtained from the Block & Baldwin (2010) database, while

104
both high and low constraint sentences were obtained from the Peele et al. (2020) database.
Because sentences obtained from these databases were initially generated with slightly different
criteria, one might argue that high constraint sentences from one database were more highly
predictable than the other. However, each participant viewed each sentence stem once
throughout the experiment, meaning that potential differences in ease of predictability between
the two databases should not vary across subjects.
Third, potential criticisms may be made of the use of the MRC Psycholinguistic Database
as a tool for controlling for imageability, concreteness, and familiarity ratings, due to the age of
the database (Coltheart, 1981). Indeed, societal changes over the last 40+ years may have caused
words to become more or less familiar to a young university population, or cause people to have
more or less experience with different words. Unfortunately, more recent databases did not
include the breadth of words and parameters of words that were included in the MRC database,
and target words were excluded from our experiment if familiarity, concreteness, or imageability
ratings were unavailable. Relying on a more recent database for these ratings would have a
pernicious effect on our materials by reducing the already precarious number of appropriate
sentences. However, we were able to obtain more recent ratings for word frequency through the
English Lexicon Project (Balota, 2007).
Fourth, the non-semantic sentences in our experiment were created by scrambling the
letters from sentences in the low-constraint condition. One might argue that sentences which are
more phonetically plausible than those used in our task might be found or generated by
employing additional external databases. However, the method we used for generating nonsemantic sentences controls more effectively for low-level visual properties (sizes, shapes of
letters, etc.) than generating a completely novel list of non-semantic sentences. The method we

105
used was previously employed by León-Cabrera et al. (2017, 2019) and is an important condition
to control for non-semantic predictions.
Finally, our generation of cannabis use scores relied solely on items from the DFAQ-CU
which loaded onto the frequency factor, so information about cannabis quantity and methods of
ingestion were not accounted for in our analysis. Further, each item was given the same weight
in our analysis. We chose to include cannabis frequency, without attempting to analyze other
variables related to cannabis consumption, in order to maintain interpretability and avoid the
possibility of models failing to converge. Further, our method of calculating cannabis frequency,
which relies on nine items which load onto the frequency factor of the DFAQ-CU, should
capture more variability in frequency of cannabis use compared to the assessments used in much
cannabis research, in which only one or two items are used to probe the frequency of a
participant’s cannabis use.
Conclusion
In conclusion, we show that schizotypy and cannabis use interact in complex ways with
sentence constraint and congruency to influence amplitudes of the SPP and N400 in a sentencereading task. Our research suggests that the brain aims to predict the nature of upcoming
semantic content, and that variables including disorganized thought and frequency of cannabis
use interact with variables which have been previously shown to affect neural markers related to
predictive processing (semantic constraint, semantic congruency). Further, our research lays the
foundation for other authors looking to explore nuances of the SPP related to schizophrenia, and
provides a novel stimulus set in English which is suitable for conducting research related to the
SPP and N400. Ultimately, it is our hope that by elucidating the cognitive mechanisms by which
schizotypy and cannabis use influence the brain’s ability to predict the nature of meaningful

106
content, this research will contribute to informing healthcare policy relating to language
comprehension in schizophrenia and cannabis use.

107
References
Ahmed, S. (2019). EXAMINING AN EVENT-RELATED BRAIN POTENTIAL INDEX OF
SEMANTIC PRIMING IN CANNABIS-USING INDIVIDUALS AT CLINICAL HIGH-RISK
FOR PSYCHOSIS. [Masters dissertation, University of Toronto].
https://tspace.library.utoronto.ca/bitstream/1807/97832/3/ _Sarah_%20_201911_MSc_thesis.pdf
Alamia, A., Timmermann, C., Vanrullen, R., & Carhart-Harris, R. L. (2020). DMT alters cortical
travelling waves. ELife, 9, 1–16. https://doi.org/10.7554/eLife.59784
Alamia, A., & VanRullen, R. (2019). Alpha oscillations and traveling waves: Signatures of predictive
coding? PLoS Biology, 17(10). https://doi.org/10.1371/journal.pbio.3000487
Alonso-solís, A., Corripio, I., Castro-manglano, P. de, Duran-, S., Garcia-garcia, M., Proal, E.,
Nuñez-marín, F., Soutullo, C., Gómez-ansón, B., Kelly, C., & Castellanos, F. X. (2013). Altered
default network resting state functional connectivity. Schizophrenia, 139, 13–18.
https://doi.org/10.1016/j.schres.2012.05.005.Altered
Altmann, G. T. M., & Kamide, Y. (1999). Incremental interpretation at verbs: Restricting the domain
of subsequent reference. Cognition, 73(3), 247–264. https://doi.org/10.1016/s00100277(99)00059-1
Argote, M., Sescousse, G., Brunelin, J., Fakra, E., Nourredine, M., & Rolland, B. (2022). Association
between formal thought disorder and cannabis use: A systematic review and meta-analysis.
Schizophrenia, 8(1). https://doi.org/10.1038/s41537-022-00286-0

108
Atakan, Z. (2012). Cannabis, a complex plant: Different compounds and different effects on
individuals. Therapeutic Advances in Psychopharmacology, 2(6), 241–254.
https://doi.org/10.1177/2045125312457586
Balota, D. A., Yap, M. J., Hutchison, K. A., Cortese, M. J., Kessler, B., Loftis, B., Neely, J. H.,
Nelson, D. L., Simpson, G. B., & Treiman, R. (2007). The English Lexicon Project. Behavior
Research Methods, 39(3), 445–459. https://doi.org/10.3758/bf03193014
Battisti, R. A., Roodenrys, S., Johnstone, S. J., Respondek, C., Hermens, D. F., & Solowij, N. (2010).
Chronic use of cannabis and poor neural efficiency in verbal memory ability.
Psychopharmacology, 209(4), 319–330. https://doi.org/10.1007/s00213-010-1800-4
Beirness, D. J., & Porath-Waller, A. J. (2015). Clearing the Smoke on Cannabis Series. Initiatives,
March, 1–8.
Besche, C., Passerieux, C., Hardy-Baylé, M.-C., Sarfati, Y., Segui, J., & Laurent, J.-P. (1997).
Syntactic and semantic processing in schizophrenic patients evaluated by lexical-decision tasks.
Neuropsychology, 11(4), 498–505. https://doi.org/10.1037/0894-4105.11.4.498
Birur, B., Kraguljac, N. V., Shelton, R. C., & Lahti, A. C. (2017). Brain structure, function, and
neurochemistry in schizophrenia and bipolar disorder- A systematic review of the magnetic
resonance neuroimaging literature. In npj Schizophrenia (Vol. 3, Issue 1). Nature Publishing
Group. https://doi.org/10.1038/s41537-017-0013-9
Block, C. K., & Baldwin, C. L. (2010). Cloze probability and completion norms for 498 sentences:
Behavioral and neural validation using event-related potentials. Behavior Research Methods,
42(3), 665–670. https://doi.org/10.3758/BRM.42.3.665

109
Boggs, D. L., Peckham, A., Boggs, A. A., & Ranganathan, M. (2016). Delta-9-tetrahydrocannabinol
and cannabidiol: Separating the chemicals from the “weed,” a pharmacodynamic discussion.
Mental Health Clinician, 6(6), 277–284. https://doi.org/10.9740/mhc.2016.11.277
Bonn-Miller, M. O., & Zvolensky, M. J. (2009). An evaluation of the nature of marijuana use and its
motives among young adult active users marijuana smoking history bonn-miller & zvolensky.
American Journal on Addictions, 18(5), 409–416. https://doi.org/10.1080/10550490903077705
Bowie, C. R., & Harvey, P. D. (2006). Cognitive deficits and functional outcome in schizophrenia. In
Neuropsychiatric Disease and Treatment (Vol. 2, Issue 4).
British Columbia Schizophrenia Society. (2022). What is Schizophrenia? https://www.bcss.org/
Bugelski, B. R., & Alampay, D. A. (1961). The role of frequency in developing perceptual sets.
Canadian Journal of Psychology Revue Canadienne De Psychologie, 15(4), 205–211.
https://doi.org/10.1037/h0083443
Callaway, D. A., Cohen, A. S., Matthews, R. A., & Dinzeo, T. (2014). Schizotypal Personality
Questionnaire - Brief Revised: Psychometric replication and extension. Personality Disorders:
Theory, Research, and Treatment, 5(1), 32–38. https://doi.org/10.1037/per0000041
Canadian Mental Health Association British Columbia. (2014). Schizophrenia.
https://cmha.bc.ca/documents/schizophrenia-3/.
Carhart-Harris, R. L., & Friston, K. J. (2019). REBUS and the anarchic brain: Toward a unified model
of the brain action of psychedelics. Pharmacological Reviews, 71(3), 316–344.
https://doi.org/10.1124/pr.118.017160

110
Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive
science. Behavioral and Brain Sciences, 36(3), 181–204.
https://doi.org/10.1017/S0140525X12000477
Cohen, A. S., Matthews, R. A., Najolia, G. M., & Brown, L. A. (2010). Toward a more
psychometrically sound brief measure of Schizotypal traits: Introducing the SPQ-Brief revised.
Journal of Personality Disorders, 24(4), 516–537. https://doi.org/10.1521/pedi.2010.24.4.516
Collerton, D., Perry, E., & McKeith, I. (2005). Why people see things that are not there: A novel
Perception and Attention Deficit model for recurrent complex visual hallucinations. Behavioral
and Brain Sciences, 28(6), 737–757. https://doi.org/10.1017/S0140525X05000130
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing.
Psychological Review, 82(6), 407–428. https://doi.org/10.1037/0033-295X.82.6.407
Coltheart, M. (1981). The MRC psycholinguistic database. The Quarterly Journal of Experimental
Psychology Section A, 33(4), 497–505. https://doi.org/10.1080/14640748108400805
Corlett, P. R., Frith, C. D., & Fletcher, P. C. (2009). From drugs to deprivation: A Bayesian
framework for understanding models of psychosis. Psychopharmacology, 206(4), 515–530.
https://doi.org/10.1007/s00213-009-1561-0
Crean, R. D., Crane, N. A., & Mason, B. J. (2011). An evidence-based review of acute and long-term
effects of cannabis use on executive cognitive functions. In Journal of Addiction Medicine (Vol.
5, Issue 1, pp. 1–8). https://doi.org/10.1097/ADM.0b013e31820c23fa

111
Cuttler, C., & Spradlin, A. (2017). Measuring cannabis consumption: Psychometric properties of the
Daily Sessions, Frequency, Age of Onset, and Quantity of Cannabis Use Inventory (DFAQ-CU).
PLoS ONE, 12(5). https://doi.org/10.1371/journal.pone.0178194
Davidson, C. A., Hoffman, L., & Spaulding, W. D. (2016). Schizotypal personality questionnaire brief revised (updated): An update of norms, factor structure, and item content in a large nonclinical young adult sample. Psychiatry Research, 238, 345–355.
https://doi.org/10.1016/j.psychres.2016.01.053
DeBruine, L. M., & Barr, D. J. (2021). Understanding Mixed-Effects Models Through Data
Simulation. Advances in Methods and Practices in Psychological Science, 4(1).
https://doi.org/10.1177/2515245920965119
DeLong, K. A., Urbach, T. P., & Kutas, M. (2005). Probabilistic word pre-activation during language
comprehension inferred from electrical brain activity. Nature Neuroscience, 8(8), 1117–1121.
https://doi.org/10.1038/nn1504
den Ouden, H. E. M., Kok, P., & de Lange, F. P. (2012). How prediction errors shape perception,
attention, and motivation. In Frontiers in Psychology (Vol. 3, Issue DEC).
https://doi.org/10.3389/fpsyg.2012.00548
Doss, M. K., Weafer, J., Gallo, D. A., & de Wit, H. (2018). Δ9-Tetrahydrocannabinol at Retrieval
Drives False Recollection of Neutral and Emotional Memories. Biological Psychiatry, 84(10),
743–750. https://doi.org/10.1016/j.biopsych.2018.04.020

112
Dumas, P., Saoud, M., Bouafia, S., Gutknecht, C., Ecochard, R., Dalery, J., & Rochet, T. (2002).
Cannabis use correlates with schizotypal personality traits in healthy students. In Psychiatry
Research (Vol. 109).
Elston, G. N. (2003). Cortex, Cognition and the Cell: New Insights into the Pyramidal Neuron and
Prefrontal Function. Cerebral Cortex, 13(11), 1124–1138. https://doi.org/10.1093/cercor/bhg093
Emrich, H. M., Weber, M. M., Wendl, A., Zihl, J., von Meyer, L., & Hanisch, W. (1991). Reduced
binocular depth inversion as an indicator of cannabis-induced censorship impairment.
Pharmacology, Biochemistry and Behavior, 40(3), 689–690. https://doi.org/10.1016/00913057(91)90383-D
Epstein, R., Harris, A., Stanley, D., & Kanwisher, N. (1999). The Parahippocampal Place Area.
Neuron, 23(1), 115–125. https://doi.org/10.1016/s0896-6273(00)80758-8
Erickson, M. A., Ruffle, A., & Gold, J. M. (2016). A Meta-Analysis of Mismatch Negativity in
Schizophrenia: From Clinical Risk to Disease Specificity and Progression. Biological Psychiatry,
79(12), 980–987. https://doi.org/10.1016/j.biopsych.2015.08.025
Fergusson, D. M., Fergusson, D., Horwood, L. J., & Ridder, E. M. (2005). Tests of causal linkages
between cannabis use and psychotic symptoms. Addiction, 100, 354–366.
https://doi.org/10.1111/j.1360-0443.2005.001001.x
Fioravanti, M., Bianchi, V., & Cinti, M. E. (2012). Cognitive deficits in schizophrenia: an updated
metanalysis of the scientific evidence. http://www.biomedcentral.com/1471-244X/12/64

113
Fletcher, P. C., & Frith, C. D. (2009). Perceiving is believing: A Bayesian approach to explaining the
positive symptoms of schizophrenia. In Nature Reviews Neuroscience (Vol. 10, Issue 1, pp. 48–
58). https://doi.org/10.1038/nrn2536
Fonsesca-Pedrero, E., Cohen, A., Ortuño-Sierra, J., Pérez de Álbeniz, A., & Muñiz, J. (2016).
DIMENSIONAL STRUCTURE AND MEASUREMENT INVARIANCE OF THE
SCHIZOTYPAL PERSONALITY QUESTIONNAIRE-BRIEF REVISED (SPQ-BR) SCORES
ACROSS AMERICAN AND SPANISH SAMPLES. In Journal of Personality Disorders (Vol.
30).
Ford, J. M., Roach, B. J., William Faustman, B. O., & Mathalon, D. H. (2007). Article Synch Before
You Speak: Auditory Hallucinations in Schizophrenia. In Am J Psychiatry (Vol. 164).
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience,
11(2), 127–138. https://doi.org/10.1038/nrn2787
Friston, K., Brown, H. R., Siemerkus, J., & Stephan, K. E. (2016). The dysconnection hypothesis
(2016). In Schizophrenia Research (Vol. 176, Issues 2–3, pp. 83–94). Elsevier B.V.
https://doi.org/10.1016/j.schres.2016.07.014
Glantz, L. A., & Lewis, D. A. (2000). Decreased dendritic spine density on prefrontal cortical
pyramidal neurons in schizophrenia. Archives of General Psychiatry, 57(1), 65.
https://doi.org/10.1001/archpsyc.57.1.65
Greenwood, L., Broyd, S. J., Croft, R., Todd, J., Michie, P. T., Johnstone, S., Murray, R., & Solowij,
N. (2014). Chronic effects of cannabis use on the auditory mismatch negativity. Biological
Psychiatry, 75(6), 449–458. https://doi.org/10.1016/j.biopsych.2013.05.035

114
Grisoni, L., Tomasello, R., & Pulvermüller, F. (2021). Correlated Brain Indexes of Semantic
Prediction and Prediction Error: Brain Localization and Category Specificity. Cerebral Cortex,
31(3), 1553–1568. https://doi.org/10.1093/cercor/bhaa308
Guo, J. Y., Ragland, J. D., & Carter, C. S. (2019). Memory and cognition in schizophrenia. In
Molecular Psychiatry (Vol. 24, Issue 5, pp. 633–642). Nature Publishing Group.
https://doi.org/10.1038/s41380-018-0231-1
Haney, M., Ward, A. S., Comer, S. D., Foltin, R. W., & Fischman, M. W. (1999). Abstinence
symptoms following smoked marijuana in humans. Psychopharmacology, 141(4), 395–404.
https://doi.org/10.1007/s002130050849
Hanna, R. C., Shalvoy, A., Cullum, C. M., Ivleva, E. I., Keshavan, M., Pearlson, G., Hill, S. K.,
Sweeney, J. A., Tamminga, C. A., & Ghose, S. (2016). Cognitive Function in Individuals with
Psychosis: Moderation by Adolescent Cannabis Use. Schizophrenia Bulletin, 42(6), 1496–1503.
https://doi.org/10.1093/schbul/sbw030
Hart, C. L., van Gorp, W., Haney, M., Foltin, R. W., & Fischman, M. W. (2001). Effects of acute
smoked marijuana on complex cognitive performance. Neuropsychopharmacology, 25(5), 757–
765. https://doi.org/10.1016/S0893-133X(01)00273-1
Hill, E. L., Gallopin, T., Férézou, I., Cauli, B., Rossier, J., Schweitzer, P., & Lambolez, B. (2007).
Functional CB1 receptors are broadly expressed in neocortical GABAergic and glutamatergic
neurons. Journal of Neurophysiology, 97(4), 2580–2589. https://doi.org/10.1152/jn.00603.2006

115
Ilan, A. B., Gevins, A., Coleman, M., Elsohly, M. A., & de Wit, H. (2005). Neurophysiological and
subjective profile of marijuana with varying concentrations of cannabinoids. In Behavioural
Pharmacology (Vol. 16). Lippincott Williams & Wilkins.
Ishkhanyan, B., Michel Lange, V., Boye, K., Mogensen, J., Karabanov, A., Hartwigsen, G., &amp;
Siebner, H. R. (2020). Anterior and posterior left inferior frontal gyrus contribute to the
implementation of grammatical determiners during language production. Frontiers in
Psychology, 11. https://doi.org/10.3389/fpsyg.2020.00685
Iversen, L. (2019). Science of Marijuana. Oxford University Press.
Jessen, F., Scheef, L., Germeshausen, L., Tawo, Y., Kockler, M., Kuhn, K.-U., Maier, W., Schild, H.
H., & Heun, R. (2003). Reduced Hippocampal Activation During Encoding and Recognition of
Words in Schizophrenia Patients. In Am J Psychiatry (Vol. 160). http://ajp.psychiatryonline.org
Kay, S. R., Opler, L. A., & Lindenmayer, J. P. (1989). The Positive and Negative Syndrome Scale
(PANSS): rationale and standardisation. The British Journal of Psychiatry. Supplement, 13(7),
59–67.
Keith, D. R., Hart, C. L., McNeil, M. P., Silver, R., & Goodwin, R. D. (2015). Frequent marijuana
use, binge drinking and mental health problems among undergraduates. American Journal on
Addictions, 24(6), 499–506. https://doi.org/10.1111/ajad.12201
Kiang, M., Christensen, B. K., Streiner, D. L., Roy, C., Patriciu, I., & Zipursky, R. B. (2013).
Association of abnormal semantic processing with delusion-like ideation in frequent cannabis
users: An electrophysiological study. Psychopharmacology, 225(1), 95–104.
https://doi.org/10.1007/s00213-012-2800-3

116
Kiang, M., & Kutas, M. (2005). Association of schizotypy with Semantic Processing Differences: An
event-related brain potential study. Schizophrenia Research, 77(2-3), 329–342.
https://doi.org/10.1016/j.schres.2005.03.021
Kilbourne, A. M., Morden, N. E., Austin, K., Ilgen, M., Mccarthy, J. F., Dalack, G., & Blow, F. C.
(2009). Excess heart-disease-related mortality in a national study of patients with mental
disorders: identifying modifiable risk factors. Gen Hosp Psychiatry, 31(6), 555–563.
https://doi.org/10.1016/j.genhosppsych.2009.07.008
Kirschstein, T., & Köhling, R. (2009). What is the source of the EEG? Clinical EEG and
Neuroscience, 40(3), 146–149. https://doi.org/10.1177/155005940904000305
Koethe, D., Gerth, C. W., Neatby, M. A., Haensel, A., Thies, M., Schneider, U., Emrich, H. M.,
Klosterkötter, J., Schultze-Lutter, F., & Leweke, F. M. (2006). Disturbances of visual
information processing in early states of psychosis and experimental delta-9tetrahydrocannabinol altered states of consciousness. Schizophrenia Research, 88(1–3), 142–
150. https://doi.org/10.1016/j.schres.2006.07.023
Kotov, R., Krueger, R. F., & Watson, D. (2018). A paradigm shift in psychiatric classification: the
Hierarchical Taxonomy Of Psychopathology (HiTOP). World Psychiatry, 17(1), 24–25.
https://doi.org/10.1002/wps.20478
Kowal, M. A., Hazekamp, A., Colzato, L. S., van Steenbergen, H., van der Wee, N. J. A., Durieux, J.,
Manai, M., & Hommel, B. (2015). Cannabis and creativity: Highly potent cannabis impairs
divergent thinking in regular cannabis users. Psychopharmacology, 232(6), 1123–1134.
https://doi.org/10.1007/s00213-014-3749-1

117
Kuperberg, G. R., & Jaeger, T. F. (2016). What do we mean by prediction in language
comprehension? Language, Cognition and Neuroscience, 31(1), 32–59.
https://doi.org/10.1080/23273798.2015.1102299
Kutas, M. (1993). In the company of other words: Electrophysiological evidence for single-word and
sentence context effects. Language and Cognitive Processes, 8(4), 533–572.
https://doi.org/10.1080/01690969308407587
Kutas, M., & Hillyard, S. (1984). Brain potentials during reading reflect word expectancy and
semantic association. Nature, 307, 161–163.
Lane, S. D., Cherek, D. R., Tcheremissine, O. v., Lieving, L. M., & Pietras, C. J. (2005). Acute
marijuana effects on human risk taking. Neuropsychopharmacology, 30(4), 800–809.
https://doi.org/10.1038/sj.npp.1300620
Lawn, W., Freeman, T. P., Pope, R. A., Joye, A., Harvey, L., Hindocha, C., Mokrysz, C., Moss, A.,
Wall, M. B., Bloomfield, M. A., Das, R. K., Morgan, C. J., Nutt, D. J., & Curran, H. V. (2016).
Acute and chronic effects of cannabinoids on effort-related decision-making and reward
learning: an evaluation of the cannabis ‘amotivational’ hypotheses. Psychopharmacology,
233(19–20), 3537–3552. https://doi.org/10.1007/s00213-016-4383-x
León-Cabrera, P., Flores, A., Rodríguez-Fornells, A., & Morís, J. (2019). Ahead of time: Early
sentence slow cortical modulations associated to semantic prediction. NeuroImage, 189(October
2018), 192–201. https://doi.org/10.1016/j.neuroimage.2019.01.005

118
León-Cabrera, P., Rodríguez-Fornells, A., & Morís, J. (2017). Electrophysiological correlates of
semantic anticipation during speech comprehension. Neuropsychologia.
https://doi.org/10.1016/j.neuropsychologia.2017.02.026
Li, S., Hu, N., Zhang, W., Tao, B., Dai, J., Gong, Y., Tan, Y., Cai, D., & Lui, S. (2019).
Dysconnectivity of multiple brain networks in schizophrenia: A meta-analysis of resting-state
functional connectivity. In Frontiers in Psychiatry (Vol. 10, Issue JULY). Frontiers Media S.A.
https://doi.org/10.3389/fpsyt.2019.00482
Liao, W., Mantini, D., Zhang, Z., Pan, Z., Ding, J., Gong, Q., Yang, Y., & Chen, H. (2010).
Evaluating the effective connectivity of resting state networks using conditional Granger
causality. Biological Cybernetics, 102(1), 57–69. https://doi.org/10.1007/s00422-009-0350-5
Lopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: An open-source toolbox for the analysis of
event-related potentials. Frontiers in Human Neuroscience, 8.
https://doi.org/10.3389/fnhum.2014.00213
Luck, S. J. (2014). An Introduction to the event-related potential technique. MIT Press.
Lynott, D., Connell, L., Brysbaert, M., Brand, J., & Carney, J. (2020). The Lancaster Sensorimotor
Norms: multidimensional measures of perceptual and action strength for 40,000 English words.
Behavior Research Methods, 52(3), 1271–1291. https://doi.org/10.3758/s13428-019-01316-z
Mathys, C., Daunizeau, J., Friston, K. J., & Stephan, K. E. (2011). A Bayesian foundation for
individual learning under uncertainty. Frontiers in Human Neuroscience, MAY, 9.
https://doi.org/10.3389/fnhum.2011.00039
Mcdonald, C., Murphy, K. C., & Sci, M. (2003). The new genetics of schizophrenia.

119
Mckay, A. P., Mckenna, P. J., Bentham, P., Mortimer, A. M., Holbery, A., & Hodges, J. R. (1996).
Semantic Memory Is Impaired in Schizophrenia.
McKetin, R., Parasu, P., Cherbuin, N., Eramudugolla, R., & Anstey, K. J. (2016). A longitudinal
examination of the relationship between cannabis use and cognitive function in mid-life adults.
Drug and Alcohol Dependence, 169, 134–140. https://doi.org/10.1016/j.drugalcdep.2016.10.022
Meehl, P. (1990). Toward an Integrated Theory of Schizotaxia, Schizotypy, and Schizophrenia.
Journal of Personality Disorders, 4(1), 1–99. https://doi.org/10.1002/smj.2712
Meehl, P. E. (1962). Schizotaxia, schizotypy, schizophrenia. American Psychologist, 17(12), 827–
838. https://doi.org/10.1037/h0041029
Mestres-Misse, A., Rodriguez-Fornells, A., & Munte, T. F. (2006). Watching the brain during
meaning acquisition. Cerebral Cortex, 17(8), 1858–1866. https://doi.org/10.1093/cercor/bhl094
Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recognizing pairs of words: Evidence of a
dependence between retrieval operations. Journal of Experimental Psychology, 90(2), 227–234.
https://doi.org/10.1037/h0031564
Miller, M. L., Chadwick, B., Dickstein, D. L., Purushothaman, I., Egervari, G., Rahman, T.,
Tessereau, C., Hof, P. R., Roussos, P., Shen, L., Baxter, M. G., & Hurd, Y. L. (2019).
Adolescent exposure to Δ9-tetrahydrocannabinol alters the transcriptional trajectory and
dendritic architecture of prefrontal pyramidal neurons. Molecular Psychiatry, 24(4), 588–600.
https://doi.org/10.1038/s41380-018-0243-x
Mohammad, O. M., & de Lisi, L. E. (2013). N400 in schizophrenia patients. In Current Opinion in
Psychiatry (Vol. 26, Issue 2, pp. 196–207). https://doi.org/10.1097/YCO.0b013e32835d9e56

120
Morgan, C. J. A., Rothwell, E., Atkinson, H., Mason, O., & Curran, H. V. (2010). Hyper-priming in
cannabis users: A naturalistic study of the effects of cannabis on semantic memory function.
Psychiatry Research, 176(2–3), 213–218. https://doi.org/10.1016/j.psychres.2008.09.002
Moritz, S., Mersmann, K., Kloss, M., Jacobsen, D., Wilke, U., Andresen, B., Naber, D., & Pawlik, K.
(2001). “Hyper-priming” in thought-disordered schizophrenic patients. Psychological Medicine,
31(2), 221–229. https://doi.org/10.1017/S0033291701003105
Morrison, P. D., Zois, V., McKeown, D. A., Lee, T. D., Holt, D. W., Powell, J. F., Kapur, S., &
Murray, R. M. (2009). The acute effects of synthetic intravenous 9- tetrahydrocannabinol on
psychosis, mood and cognitive functioning. Psychological Medicine, 39(10), 1607–1616.
https://doi.org/10.1017/S0033291709005522
Nastase, S. A., & Haxby, J. v. (2016). Structural basis of semantic memory. The Curated Reference
Collection in Neuroscience and Biobehavioral Psychology, January, 133–151.
https://doi.org/10.1016/B978-0-12-809324-5.21073-0
Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless
spreading activation and limited-capacity attention. Journal of Experimental Psychology:
General, 106(3), 226–254. https://doi.org/10.1037/0096-3445.106.3.226
Nichols, D. E. (2016). Psychedelics. Pharmacological Reviews, 68(2), 264–355.
https://doi.org/10.1124/pr.115.011478
Nordgaard, J., Gravesen-Jensen, M., Buch-Pedersen, M., & Parnas, J. (2021). Formal Thought
Disorder and Self-Disorder: An Empirical Study. Frontiers in Psychiatry, 12.
https://doi.org/10.3389/fpsyt.2021.640921

121
Patel, K. R., Cherian, J., Gohil, K., & Atkinson, D. (2014). Schizophrenia: overview and treatment
options. P & T : a peer-reviewed journal for formulary management, 39(9), 638–645.
Patel, S. J., Khan, S., M, S., & Hamid, P. (2020). The Association Between Cannabis Use and
Schizophrenia: Causative or Curative? A Systematic Review. Cureus.
https://doi.org/10.7759/cureus.9309
Pearson, N. T., & Berry, J. H. (2019). Cannabis and psychosis through the lens of dsm-5.
International Journal of Environmental Research and Public Health, 16(21).
https://doi.org/10.3390/ijerph16214149
Peelle, J. E., Miller, R. L., Rogers, C. S., Spehar, B., Sommers, M. S., & van Engen, K. J. (2020).
Completion norms for 3085 English sentence contexts. Behavior Research Methods, 52(4),
1795–1799. https://doi.org/10.3758/s13428-020-01351-1
Pomarol-Clotet, E., Oh, T. M. S. S., Laws, K. R., & McKenna, P. J. (2008). Semantic priming in
schizophrenia: Systematic review and meta-analysis. In British Journal of Psychiatry (Vol. 192,
Issue 2, pp. 92–97). https://doi.org/10.1192/bjp.bp.106.032102
Pulvermüller, F., & Grisoni, L. (2020). Semantic Prediction in Brain and Mind. Trends in Cognitive
Sciences, 24(10), 781–784. https://doi.org/10.1016/j.tics.2020.07.002
Quillian, M. R. (1969). The teachable language comprehender: A simulation program and theory of
language. Communications of the ACM, 12(8), 459–476. https://doi.org/10.1145/363196.363214
Ragland, J. D., Gur, R. C., Valdez, J., Turetsky, B. I., Elliott, M., Kohler, C., Siegel, S., Kanes, S., &
Gur, R. E. (2004). Event-Related fMRI of Frontotemporal Activity During Word Encoding and
Recognition in Schizophrenia.

122
Ragland, J. D., Laird, A. R., Ranganath, C., Blumenfeld, R. S., Gonzales, S. M., & Glahn, D. C.
(2009). Prefrontal activation deficits during episodic memory in schizophrenia. In American
Journal of Psychiatry (Vol. 166, Issue 8, pp. 863–874).
https://doi.org/10.1176/appi.ajp.2009.08091307
Ragland, J. D., Ranganath, C., Phillips, J., Boudewyn, M. A., Kring, A. M., Lesh, T. A., Long, D. L.,
Luck, S. J., Niendam, T. A., Solomon, M., Swaab, T. Y., & Carter, C. S. (2015). Cognitive
control of episodic memory in schizophrenia: Differential role of dorsolateral and ventrolateral
prefrontal cortex. Frontiers in Human Neuroscience, 9(NOVEMBER).
https://doi.org/10.3389/fnhum.2015.00604
Raine, A. (1991). The spq: A scale for the assessment of schizotypal personality based on DSM-III-r
criteria. Schizophrenia Bulletin, 17(4), 555–564. https://doi.org/10.1093/schbul/17.4.555
Ramaekers, J. G., Kauert, G., van Ruitenbeek, P., Theunissen, E. L., Schneider, E., & Moeller, M. R.
(2006). High-potency marijuana impairs executive function and inhibitory motor control.
Neuropsychopharmacology, 31(10), 2296–2303. https://doi.org/10.1038/sj.npp.1301068
R Core Team (2023). R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria.
URL https://www.R-project.org/.
Robinson, S. M., Sobell, L. C., Sobell, M. B., & Leo, G. I. (2014). Reliability of the Timeline
Followback for cocaine, cannabis, and cigarette use. Psychology of Addictive Behaviors, 28(1),
154–162. https://doi.org/10.1037/a0030992

123
Rogers, B. P., Morgan, V. L., Newton, A. T., & Gore, J. C. (2007). Assessing Functional Connectivity
in the Human Brain by FMRI.
Sainani, K. L. (2012). Dealing with non-normal data. PM&R, 4(12), 1001–1005.
https://doi.org/10.1016/j.pmrj.2012.10.013
Salisbury, D. F., Shenton, M. E., Nestor, P. G., & McCarley, R. W. (2002). Semantic bias, homograph
comprehension, and event-related potentials in schizophrenia. Clinical Neurophysiology, 113(3),
383–395. https://doi.org/10.1016/s1388-2457(02)00003-2
Shao, X., Li, M., Yang, Y., Li, X., & Han, Z. (2021). The Neural Basis of Semantic Prediction in
Sentence Comprehension. Journal of Cognitive Neuroscience, 1–22.
https://doi.org/10.1162/jocn_a_01793
Shin, K. S., Kim, J. S., Kim, S. N., Koh, Y., Jang, J. H., An, S. K., O’Donnell, B. F., Chung, C. K., &
Kwon, J. S. (2012). Aberrant auditory processing in schizophrenia and in subjects at ultra-highrisk for psychosis. Schizophrenia Bulletin, 38(6), 1258–1267.
https://doi.org/10.1093/schbul/sbr138
Sinnett, S., Smilek, D., & Kingstone, A. (2016). Cognition. Oxford University Press.
Skosnik, P. D., D’Souza, D. C., Steinmetz, A. B., Edwards, C. R., Vollmer, J. M., Hetrick, W. P., &
O’Donnell, B. F. (2012). The effect of chronic cannabinoids on broadband eeg neural
oscillations in humans. Neuropsychopharmacology, 37(10), 2184–2193.
https://doi.org/10.1038/npp.2012.65
Sobell, L. C., & Sobell, M. B. (1992). Timeline follow-back. Measuring Alcohol Consumption, 41–
72. https://doi.org/10.1007/978-1-4612-0357-5_3

124
Solowij, N., Jones, K. A., Rozman, M. E., Davis, S. M., Ciarrochi, J., Heaven, P. C. L., Lubman, D.
I., & Yücel, M. (2011). Verbal learning and memory in adolescent cannabis users, alcohol users
and non-users. Psychopharmacology, 216(1), 131–144. https://doi.org/10.1007/s00213-0112203-x
Squire, L. R., & Zola-Morgan, S. (1991). The Medial Temporal Lobe Memory System. In New Series
(Vol. 253, Issue 5026).
Statistics Canada. (2021, June 23). Retail trade sales by province and territory.
https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=2010000801&pickMembers%5B0%5D=2
.30&pickMembers%5B1%5D=3.1.
Statistics Canada. (2020, February 19). What has changed since cannabis was legalized? Retrieved
September 12, 2020, from https://www150.statcan.gc.ca/n1/pub/82-003x/2020002/article/00002-eng.htm
Staub, A., Grant, M., Astheimer, L., & Cohen, A. (2015). The influence of cloze probability and item
constraint on Cloze Task Response Time. Journal of Memory and Language, 82, 1–17.
https://doi.org/10.1016/j.jml.2015.02.004
Stephan, K. E., Friston, K. J., & Frith, C. D. (2009). Dysconnection in schizophrenia: From abnormal
synaptic plasticity to failures of self-monitoring. Schizophrenia Bulletin, 35(3), 509–527.
https://doi.org/10.1093/schbul/sbn176
Sterzer, P., Adams, R. A., Fletcher, P., Frith, C., Lawrie, S. M., Muckli, L., Petrovic, P., Uhlhaas, P.,
Voss, M., & Corlett, P. R. (2018). The Predictive Coding Account of Psychosis. Biological
Psychiatry, 84(9), 634–643. https://doi.org/10.1016/j.biopsych.2018.05.015

125
Struve, F. A., Manno, B. R., Kemp, P., Patrick, G., & Manno, J. E. (2003). Acute Marihuana (THC)
Exposure Produces a “Transient” Topographic Quantitative EEG Profile Identical to the
“Persistent” Profile Seen in Chronic Heavy Users. Clinical EEG and Neuroscience, 34(2), 75–
83. https://doi.org/10.1177/155005940303400206
Stuke, H., Weilnhammer, V. A., Sterzer, P., & Schmack, K. (2019). Delusion Proneness is Linked to
a Reduced Usage of Prior Beliefs in Perceptual Decisions. Schizophrenia Bulletin, 45(1).
https://doi.org/10.1093/schbul/sbx189
Tan, E. J., Neill, E., Tomlinson, K., & Rossell, S. L. (2020). Semantic Memory Impairment Across
the Schizophrenia Continuum: A Meta-Analysis of Category Fluency Performance.
Schizophrenia Bulletin Open, 1(1). https://doi.org/10.1093/schizbullopen/sgaa054
Van Engen, K. J., Dey, A., Runge, N., Spehar, B., Sommers, M. S., & Peelle, J. E. (2020). Effects of
age, word frequency, and noise on the time course of spoken word recognition. Collabra:
Psychology, 6(1). https://doi.org/10.1525/collabra.17247
Wainberg, M., Jacobs, G. R., di Forti, M., & Tripathy, S. J. (2021). Cannabis, schizophrenia genetic
risk, and psychotic experiences: a cross-sectional study of 109,308 participants from the UK
Biobank. Translational Psychiatry, 11(1). https://doi.org/10.1038/s41398-021-01330-w
Walsh, K. S., McGovern, D. P., Clark, A., & O’Connell, R. G. (2020). Evaluating the
neurophysiological evidence for predictive processing as a model of perception. In Annals of the
New York Academy of Sciences (Vol. 1464, Issue 1, pp. 242–268). Blackwell Publishing Inc.
https://doi.org/10.1111/nyas.14321

126
Wang, K., Cheung, E. F. C., Gong, Q. yong, & Chan, R. C. K. (2011). Semantic processing
disturbance in patients with schizophrenia: A meta-analysis of the N400 component. PLoS ONE,
6(10). https://doi.org/10.1371/journal.pone.0025435
Woodberry, K. A., Anthony Giuliano, A. J., & Seidman, L. J. (2008). Reviews and Overviews
Premorbid IQ in Schizophrenia: A Meta-Analytic Review. In Am J Psychiatry (Vol. 165, Issue
5).
Yee, E., & Thompson-Schill, S. L. (2016). Putting concepts into context. Psychonomic Bulletin and
Review, 23(4), 1015–1027. https://doi.org/10.3758/s13423-015-0948-7
Zeiger, J. S., Haberstick, B. C., Corley, R. P., Ehringer, M. A., Crowley, T. J., Hewitt, J. K., Hopfer,
C. J., Stallings, M. C., Young, S. E., & Rhee, S. H. (2010). Subjective effects to marijuana
associated with marijuana use in community and clinical subjects. Drug and Alcohol
Dependence, 109(1–3), 161–166. https://doi.org/10.1016/j.drugalcdep.2009.12.026
Zemla, J. C., & Austerweil, J. L. (2018). Estimating Semantic Networks of Groups and Individuals
from Fluency Data. Computational Brain & Behavior, 1(1), 36–58.
https://doi.org/10.1007/s42113-018-0003-7

127
Appendix A
Study Questionnaires
This section contains the Schizophrenia Personality Questionnaire Brief Revised Updated
(SPQ-BRU; Davidson et al., 2016), the Daily Sessions, Frequency, Age of Onset, and Quantity
of Cannabis Use Inventory (DFAQ-CU; Cuttler & Spradlin, 2017), and the demographic
questionnaire used in our lab.

128
SPQ-BRU Items (Davidson et al., 2016)
item
#
1

Higher
-Order
CP

Subfactor
IR

Text
I sometimes feel that people are talking about me.

2

CP

IR

I sometimes feel that other people are watching me.

3

CP

IR

When shopping, I get the feeling that other people are taking notice of me.

4

CP

SU

I often feel that others have it in for me.

5

CP

SU

I sometimes get concerned that friends or co-workers are not really loyal or trustworthy.

6

CP

SU

7

IP

CF

I often have to keep an eye out to stop people from taking advantage of me.
I feel that I cannot get 'close' to people.

8

IP

CF

9

IP

CF

10

IP

CA

I find it hard to be emotionally close to other people.
I feel that there is no one I am really close to outside of my immediate family, or people I can confide
in or talk to about personal problems.
I tend to keep my feelings to myself.

11

IP

CA

I rarely laugh and smile.

12

IP

CA

I am not good at expressing my true feelings by the way I talk and look.

13

DO

EB

Other people see me as slightly eccentric (odd).

14

DO

EB

I am an odd, unusual person

15

DO

EB

I have some eccentric (odd) habits.

16

DO

EB

People sometimes comment on my unusual mannerisms and habits.

17

IP/SA

SA

I often feel nervous when I am in a group of unfamiliar people.

18

IP/SA

SA

I get anxious when meeting people for the first time.

19

IP/SA

SA

I feel very uncomfortable in social situations involving unfamiliar people.

20

IP/SA

SA

I sometimes avoid going to places where there will be many people because I will get anxious.

21

CP

MT

I believe in telepathy (mind-reading).

22

CP

MT

I believe in clairvoyance (psychic forces, fortune telling).

23

CP

MT

I have had experiences with astrology, seeing the future, UFO's, ESP, or a sixth sense.

24

CP

MT

I have felt that I was communicating with another person telepathically (by mind-reading).

25

DO

OS

I sometimes jump quickly from one topic to another when speaking.

26

DO

OS

I tend to wander off the topic when having a conversation.

27

DO

OS

I often ramble on too much when speaking.

28

DO

OS

I sometimes forget what I am trying to say.

29

CP

UP

I often hear a voice speaking my thoughts aloud.

30

CP

UP

When I look at a person or at myself in a mirror, I have seen the face change right before my eyes.

31

CP

UP

My thoughts are sometimes so strong that I can almost hear them.

32

CP

UP

Everyday things seem unusually large or small.

Response format: 1. Strongly Disagree 2. Disagree 3. Neutral 4. Agree 5. Strongly Agree
CP: Cognitive Perceptual; IP: Interpersonal; DO: Disorganized; IR: Ideas of Reference; SU: Suspiciousness; CF: No
Close Friends; CA: Constricted Affect; EB: Eccentric Behavior; SA: Social Anxiety; MT: Magical Thinking; OS:
Odd Speech; UP: Unusual Perceptions.

129
DFAQ-CU Inventory (Cuttler & Spradlin, 2017)
Instructions: Please read each of the following questions and mark the response alternative that
best describes your use of cannabis. Note that the term cannabis is being used to refer to
marijuana, cannabis concentrates, and cannabis-infused edibles.
1. Have you ever used cannabis?
0 = No
1 = Yes
*If response = 0 then skip to end of questionnaire
2. Which of the following best captures when you last used cannabis?
1 = over a year ago
7 = last week
2 = 9 – 12 months ago
8 = this week
3 = 6 – 9 months ago
9 = yesterday
4 = 3 – 6 months ago
10 = today*
5 = 1 – 3 months ago
11 = I am currently high*
6 = less than 1 month ago
*If response = 10 (today) or 11 (I am currently high) then answer 2b below
2b. How high are you right now?
0 = I am not at all high
1 = I am a little bit high
2 = I am moderately high
3 = I am very high
4 = I am extremely high
3. Which of the following best captures the average frequency you currently use cannabis?
0 = I do not use cannabis
7 = once a week
1 = less than once a year
8 = twice a week
2 = once a year
9 = 3 – 4 times a week
3 = once every 3-6 months (2-4 times/yr))
10 = 5 – 6 times a week
4 = once every 2 months (6 times/yr)
11 = once a day
5 = once a month (12 times/yr)
12 = more than once a day
6 = 2 – 3 times a month
4. Which of the following best captures how long you have been using cannabis at this
frequency?
1 = less than 1 month
7 = 2 – 3 years
2 = 1 – 3 months
8 = 3 – 5 years
3 = 3 – 6 months
9 = 5 – 10 years
4 = 6 – 9 months
10 = 10 – 15 years
5 = 9 – 12 months
11 = 15 – 20 years
6 = 1 – 2 years
12 = more than 20 years
5. Before the period of time you indicated above, how frequently did you use cannabis?
0 = I did not use cannabis
1 = less than once a year

130
2 = once a year
3 = once every 3-6 months (2-4 times/yr.)
4 = once every 2 months (6 times/yr.)
5 = once a month
6 = 2 – 3 times a month
7 = once a week

8 = twice a week
9 = 3 – 4 times a week
10 = 5 – 6 times a week
11 = once a day
12 = more than once a day

6. How many days of the past week did you use cannabis?
0 = 0 days
4 = 4 days
1 = 1 day
5 = 5 days
2 = 2 days
6 = 6 days
3 = 3 days
7 = 7 days
7. Approximately how many days of the past month did you use cannabis? ____________
8. Which of the following best captures the number of times you have used cannabis in your
entire life?
1 = 1 – 5 times in my life
6 = 501 – 1000 times in my life
2 = 6 – 10 times in my life
7 = 1001 – 2000 times in my life
3 = 11 – 50 times in my life
8 = 2001 – 5000 times in my life
4 = 51 –100 times in my life
9 = 5001 – 10,000 times in my life
5 = 101 – 500 times in my life
10 = More than 10,000 times in my life
9. Which of the following best captures your pattern of cannabis use throughout the week?
0 = I do not use cannabis at all
1 = I only use cannabis on weekends
2 = I only use cannabis on weekdays
3 = I use cannabis on weekends and weekdays
10. How many hours after waking up do you typically first use cannabis?
0 = I do not use cannabis at all
5 = 1 – 3 hours after waking up
1 = 12 – 18 hours after waking up
6 = within 1 hour of waking up
2 = 9 – 12 hours after waking up
7 = within ½ hour of waking up
3 = 6 – 9 hours after waking up
8 = immediately upon waking up
4 = 3 – 6 hours after waking up
11. How many times a day, on a typical weekday, do you use cannabis? ____________
12. How many times a day, on a typical weekend, do you use cannabis? ____________

13. What is the primary method you use to ingest cannabis?
0 = I do not use cannabis
1 = Joints

131
2 = Blunts (cigar sized joints)
3 = Hand pipe
4 = Bong (water pipe)
5 = Hookah

6 = Vaporizer (e.g., Volcano, Vape pen)
7 = Edibles
8 = Other _______________________

14. Which of the following other methods to ingest cannabis do you use regularly (at least 25%
of the time use you cannabis)? [Mark all that apply]
0 = None
5 = Hookah
1 = Joints
6 = Vaporizer (e.g., Volcano, Vape pen)
2 = Blunts (cigar sized joints)
7 = Edibles
3 = Hand pipe
8 = Other _______________________
4 = Bong (water pipe)
15. What is the primary form of cannabis you use?
0 = None****
A = Marijuana***
B = Concentrates (e.g., Oil, Wax, Shatter, Butane Hash Oil, Dabs)**
C = Edibles*
D = Other____________________
16. What other forms of cannabis do you use regularly (at least 25% of the time you use
cannabis)? [Mark all that apply]
0 = None****
A = Marijuana***
B = Concentrates (e.g., Oil, Wax, Shatter, Butane Hash Oil, Dabs)**
C = Edibles*
D = Other____________________
****If response to questions 15 and 16 = 0 (None) then skip to question 29
***If responses to questions 15 or 16 = A (Marijuana) then answer questions 17-21
**If responses to question 15 or 16 = B (Concentrates) then answer questions 22-26
*If responses to question 15 or 16 = C (Edibles) then answer question 27
Note: If you use more than one form of cannabis then complete all of the associated questions
listed above.

132
***If responses to questions 15 or 16 = A (Marijuana) then answer questions 17-21 below.
Please use the image below to refer to various quantities of marijuana. The image is not to scale;
the dollar bill is included to help provide size perspective.

For questions 17 to 19 below, clearly indicate the number of grams of marijuana you use with a
number between 0 – 100. Do NOT include other forms of cannabis you may use (such as
concentrates). You may use up to 3 decimals to indicate amounts under 1 gram.
Note: 1/8 of a gram = 0.125 grams, ¼ of a gram = 0.25 grams, ½ of a gram = 0.5 grams, ¾ of a
gram = 0.75 grams. 1/8 of a ounce = 3.5 grams, ¼ of an ounce = 7 grams, ½ ounce = 14 grams, 1
ounce = 28 grams
17. In a typical session, how much marijuana do you personally use? ______________________
18. On a typical day you use marijuana, how much do you personally use? _________________
19. In a typical week you use marijuana, how much marijuana do you personally use? ________
20. On a typical day you use marijuana, how many sessions do you have? __________________

133
21. What is the average THC content of the marijuana you typically use? Leave blank if you do
not know.
1 = 0 – 4%
5 = 20 – 24%
2 = 5 – 9%
6 = 25 – 30%
3 = 10 – 14%
7 = greater than 30%
4 = 15 – 19%
**If response to questions 15 or 16 = B (Concentrates) then answer questions 22-26 below
22. In a typical session you use cannabis concentrates, how many hits do you personally take? __
23. On a typical day you use cannabis concentrates, how many hits do you personally take? ____
24. How many hits of cannabis concentrates did you personally take yesterday? _____________
25. On a typical day you use cannabis concentrates, how many sessions do you have? _________
26. What is the average THC content of the concentrates you typically use? Leave blank if you
do not know.
1 = 0 – 9%
6 = 50 – 59%
2 = 10 – 19%
7 = 60 – 69%
3 = 20 – 29%
8 = 70 – 79%
4 = 30 – 39%
9 = 80 – 90%
5 = 40 – 49%
10 = greater than 90%
**If response to questions 15 or 16 = C (Edibles) then answer question 27 below
27. When you eat edibles how many milligrams of THC do you personally ingest in a typical
session? ___________
28. What is your current age? ___________
29. How many years in total have you used cannabis? ___________
30. How old were you when you FIRST tried cannabis? ___________
31. Has there been any time in your life when you used cannabis regularly (2 or more times per
month for 6 months or longer)?
0 = No
1 = Yes*
*If response = 1 (Yes) then answer questions 31b and 31c below
31b. How old were you when you FIRST STARTED using cannabis regularly (2 or more

134
times/month)? ___________
31c. Has there been any time in your life when you used cannabis on a daily or near daily
basis for 6 months or longer?
0 = No
1 = Yes*
*If response = 1 (Yes) then answer question 31ci below
31ci. How old were you when you FIRST STARTED using cannabis on a daily or
near daily basis? ___________
32. Which of the following best captures the average frequency that you used cannabis before the
age of 16?
0 = more than once a day
7 = once a month
1 = once a day
8 = once every 2 months (6 times/yr.)
2 = 5 – 6 times a week
9 = once every 3-6 months (2-4 times/yr.)
3 = 3 – 4 times a week
10 = once a year
4 = twice a week
11 = less than once a year
5 = once a week
12 = never
6 = 2 – 3 times a month
33. Do you have a physician’s recommendation to use cannabis for medicinal purposes?
0 = No
1 = Yes*
2 = Yes, but I use it for both medicinal and recreational purposes*
*If response = 1 or 2 (Yes) then answer questions 33b and 33c
33b. Which medical condition(s) do you use cannabis for?
________________________________________________________________________
33c. What percentage of the time do you use cannabis for recreational (rather than
medicinal) purposes? ________________

135
DFAQ-CU Scoring
Daily Sessions Items: 20, 25
Frequency Items: 2, 3, 6, 7, 8, 9, 10, 11, 12
Age of Onset Items: 30, 31b, 31ci, 32
Marijuana Quantity Items: 17, 18, 19
Concentrate Quantity Items: 22, 23, 24,
Edibles Quantity Item: 27
Note: Standardize (z-transform) scores prior to calculating the mean of each of the 6 factors
(daily sessions, frequency, age of onset, marijuana quantity, concentrate quantity, edibles
quantity).
Screening/Characterization Items: 1, 2b, 4, 5, 13, 14, 15, 16, 21, 26, 28, 29, 31, 31c, 33, 33b, 33c
Contact us at: DFAQCU@gmail.com if you would like a Qualtrics version of the DFAQ-CU
shared to your Qualtrics account.

136

Demographics and Health Form

Participant number: ______________

Which hand do you write with: _______

Age: ____________

What was the first language you spoke: __________

Highest level of education: _________

Do you speak any other languages: _____________
If so, what languages do you speak and at which age
did you learn them? __________________________

Biological sex: ___

Have you been diagnosed with any of the following:
_______ Attention Deficit Hyperactivity Disorder (ADHD)
_______ Dyslexia
_______ Other learning disabilities
_______ Epilepsy
Have you ever sustained a head injury?
If so, please describe what happened: ___________________________________________
Are you taking any prescription medications?
If so, please describe how and why you use them : _________________________________
_________________________________

137
Appendix B
Tables of Means and Post-Hoc Comparisons
This section contains tables showing means for all highest-order cells of the fitted models
predicting SPP and N400 amplitude. In addition, tables for post-hoc comparisons between means
of all highest-order interactions are included. Note that low and high cannabis use values
reported in this section come from estimated means of the most extreme values from our sample
on the continuous cannabis frequency variable at values of -.67 and 1.79, respectively. Likewise,
low and high disorganized thought values come from estimated means of the most extreme
values from our sample on the continuous disorganized thought variable at values of 1.75 and 5,
respectively.

138
Table 7
Mean SPP Amplitudes for Each Combination of Fixed Effects 200-0 ms Prior to the Presentation
of Final Words in the Sentence-Reading Task
Constraint

Disorganized

Hemisphere

Thought

Cannabis

M

SE

Frequency

High

Low

Left

Low

-.84

.37

High

Low

Left

High

-.95

1.08

High

Low

Mid

Low

-1.62

.37

High

Low

Mid

High

-1.39

1.08

High

Low

Right

Low

-.62

.37

High

Low

Right

High

-1.82

1.08

High

High

Left

Low

-.97

.44

High

High

Left

High

-.74

.74

High

High

Mid

Low

-.55

.44

High

High

Mid

High

-.90

.75

High

High

Right

Low

-.57

.44

High

High

Right

High

.06

.75

Low

Low

Left

Low

-.83

.37

Low

Low

Left

High

1.16

1.08

Low

Low

Mid

Low

-1.14

.37

Low

Low

Mid

High

.58

1.08

Low

Low

Right

Low

-.32

.37

Low

Low

Right

High

.04

1.08

139
Low

High

Left

Low

-.26

.44

Low

High

Left

High

-1.76

.75

Low

High

Mid

Low

.43

.44

Low

High

Mid

High

-1.97

.75

Low

High

Right

Low

-.08

.44

Low

High

Right

High

-.58

.75

NS

Low

Left

Low

-.79

.42

NS

Low

Left

High

.87

1.21

NS

Low

Mid

Low

-.73

.42

NS

Low

Mid

High

1.16

1.21

NS

Low

Right

Low

.27

.42

NS

Low

Right

High

1.88

1.21

NS

High

Left

Low

-.74

.49

NS

High

Left

High

-1.02

.83

NS

High

Mid

Low

-.54

.49

NS

High

Mid

High

-2.61

.83

NS

High

Right

Low

-.63

.49

NS

High

Right

High

-1.34

.83

140
Table 8
Comparisons of SPP Amplitudes for High Constraint, Low Constraint, and Non-Semantic
Sentences Across Levels of High and Low Disorganized Thought and Cannabis Use
Disorganized Cannabis Contrast ΔM

SE

df

t-ratio

p-value

Thought

Use

Low

Low

HC-LC

-.27

.15

28401

1.78

.18

Low

Low

HC-NS

-.61

.20

1887

3.03

.007**

Low

Low

LC-NS

-.35

.20

1887

1.71

.20

Low

High

HC-LC

-1.98

.43

28713

4.58

< .001***

Low

High

HC-NS

-2.69

.54

23542

5.02

< .001***

Low

High

LC-NS

-.71

.54

23542

1.33

.38

High

Low

HC-LC

-.73

.18

28711

4.15

< .001***

High

Low

HC-NS

-.06

.23

3195

.29

.96

High

Low

LC-NS

.67

.23

3195

2.87

.01

High

High

HC-LC

.91

.30

28601

3.04

.007**

High

High

HC-NS

1.13

.37

13555

3.03

.007**

High

High

LC-NS

.23

.37

13555

.60

.82

Note. HC = High Constraint, LC = Low Constraint, NS = Non-Semantic. ***p < .001, ** p <
.01, * p < .05. Tukey’s method used for correction of multiple comparisons.
Table 9
Mean N400 Amplitudes for Each Combination of Fixed Effects 300-500 ms After the
Presentation of Final Words in the Sentence-Reading Task

141
Constraint

Disorganized

Congruency

Thought

Cannabis

M

SE

Frequency

High

Low

Congruent

Low

.16

1.05

High

Low

Congruent

High

3.31

3.04

High

Low

Incongruent

Low

-2.23

1.05

High

Low

Incongruent

High

-1.80

3.04

High

High

Congruent

Low

3.16

1.24

High

High

Congruent

High

1.11

2.09

High

High

Incongruent

Low

1.42

1.24

High

High

Incongruent

High

.30

2.09

Low

Low

Congruent

Low

-2.15

1.05

Low

Low

Congruent

High

4.33

3.04

Low

Low

Incongruent

Low

-.79

1.05

Low

Low

Incongruent

High

-6.06

3.04

Low

High

Congruent

Low

.77

1.24

Low

High

Congruent

High

-1.75

2.09

Low

High

Incongruent

Low

.59

1.24

Low

High

Incongruent

High

-.57

2.09

Table 10
Post-Hoc Comparisons For Congruency x Constraint Interaction of N400 Amplitudes at
Electrode Pz.
Contrast

ΔM

SE

df

t-ratio

p-value

142
HCC - LCC

1.96

.526

2376

3.73

.001**

HCC – HCI

2.28

.537

674

4.25

<.001***

HCC – LCI

2.82

.537

674

5.26

<.001***

HCI – LCC

.32

.537

674

.58

.93

HCI – LCI

.54

.526

2376

1.02

.74

LCC - LCI

.86

.537

674

1.60

.38

Note. ***p < .001, ** p < .01, * p < .05. HCC = High Constraint Congruent, HCI = High
Constraint Incongruent, LCC = Low Constraint Congruent, LCI = Low Constraint Incongruent.
Tukey’s method used for correction of multiple comparisons.

Table 11
Post-Hoc Comparisons For Congruency x Cannabis Frequency Interaction of N400 Amplitudes
at Electrode Pz.
Contrast

ΔM

SE

df

t-ratio

p-value

HC – HI

3.59

.91

2170.0

3.94

< .001***

HC – LC

1.09

1.18

20.1

.92

.80

HC – LI

1.84

1.14

17.3

1.61

.40

HI – LC

2.50

1.14

17.3

2.19

.16

HI – LI

1.75

1.18

20.1

1.48

.47

LC - LI

.75

.49

503.8

1.52

.43

Note. ***p < .001, ** p < .01, * p < .05. HC = High Cannabis Use Congruent, HI = High
Cannabis Use Incongruent, LC = Low Cannabis Use Congruent, LI = Low Cannabis Use
Incongruent. Tukey’s method used for correction of multiple comparisons.